diff --git a/build_wheel.bat b/build_wheel.bat new file mode 100644 index 000000000..68fcb3e57 --- /dev/null +++ b/build_wheel.bat @@ -0,0 +1,35 @@ +@echo off +REM ============================================================================ +REM build_wheel.bat - One-shot: clean, build, and verify the graphify wheel. +REM +REM Produces dist\graphifyy--py3-none-any.whl and checks it actually +REM contains the cpp_deep package + the [cpp] extra before declaring success. +REM +REM Usage: build_wheel.bat +REM ============================================================================ +setlocal +cd /d "%~dp0" + +echo [1/4] Cleaning old build artifacts... +if exist dist rmdir /s /q dist +if exist build rmdir /s /q build +for /d %%D in (*.egg-info) do rmdir /s /q "%%D" +for /d %%D in (graphify\*.egg-info) do rmdir /s /q "%%D" + +echo [2/4] Ensuring the 'build' backend is available... +python -c "import build" >nul 2>&1 || python -m pip install build -q + +echo [3/4] Building wheel... +python -m build --wheel +if errorlevel 1 ( + echo [ERROR] wheel build failed. + exit /b 1 +) + +echo [4/4] Verifying wheel contents... +python -c "import glob,os,sys,zipfile; whls=sorted(glob.glob('dist/graphifyy-*.whl'));_=sys.exit('no wheel') if not whls else None; whl=whls[-1]; zf=zipfile.ZipFile(whl); names=zf.namelist(); cpp=[n for n in names if n.startswith('graphify/cpp_deep/') and n.endswith('.py')]; meta=zf.read([n for n in names if n.endswith('METADATA')][0]).decode('utf-8','replace'); skills=[n for n in names if n.endswith('skill.md') or '/skills/' in n]; checks=[('cpp_deep modules', len(cpp)>=10, str(len(cpp))+' py'),('[cpp] extra libclang','libclang' in meta,''),('skill docs',len(skills)>=1,str(len(skills))),('resolver.py','graphify/cpp_deep/resolver.py' in names,'')]; [print(' [%s] %s %s'%('OK ' if c else 'FAIL',l,d)) for l,c,d in checks]; bad=[l for l,c,d in checks if not c]; sys.exit('INCOMPLETE wheel: '+', '.join(bad)) if bad else print('\n[DONE] '+whl); print(' Copy this wheel + install.bat to the target machine.')" +if errorlevel 1 ( + echo [ERROR] wheel verification failed - do not distribute. + exit /b 1 +) +endlocal diff --git a/build_wheel.sh b/build_wheel.sh new file mode 100644 index 000000000..3ac993f85 --- /dev/null +++ b/build_wheel.sh @@ -0,0 +1,63 @@ +#!/usr/bin/env bash +# ============================================================================= +# build_wheel.sh - One-shot: clean, build, and verify the graphify wheel. +# +# Produces dist/graphifyy--py3-none-any.whl and checks it actually +# contains the cpp_deep package + the [cpp] extra before declaring success. +# +# Usage: ./build_wheel.sh +# ============================================================================= +set -euo pipefail + +SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" +cd "$SCRIPT_DIR" +PY="${PYTHON:-python}" + +echo "[1/4] Cleaning old build artifacts..." +rm -rf dist/ build/ ./*.egg-info graphify/*.egg-info 2>/dev/null || true + +echo "[2/4] Ensuring the 'build' backend is available..." +if ! "$PY" -c "import build" >/dev/null 2>&1; then + if command -v uv >/dev/null 2>&1; then + uv pip install build -q + else + "$PY" -m pip install build -q + fi +fi + +echo "[3/4] Building wheel..." +"$PY" -m build --wheel + +echo "[4/4] Verifying wheel contents..." +"$PY" - <<'PYEOF' +import glob, os, sys, zipfile + +whls = sorted(glob.glob("dist/graphifyy-*.whl")) +if not whls: + print(" [ERROR] no wheel produced in dist/", file=sys.stderr); sys.exit(1) +whl = whls[-1] +zf = zipfile.ZipFile(whl) +names = zf.namelist() + +cpp = [n for n in names if n.startswith("graphify/cpp_deep/") and n.endswith(".py")] +meta = zf.read([n for n in names if n.endswith("METADATA")][0]).decode("utf-8", "replace") +skills = [n for n in names if n.endswith("skill.md") or "/skills/" in n] + +ok = True +def check(label, cond, detail=""): + global ok + mark = "OK " if cond else "FAIL" + if not cond: ok = False + print(f" [{mark}] {label}{(' - ' + detail) if detail else ''}") + +check(f"wheel present ({os.path.basename(whl)}, {os.path.getsize(whl)//1024} KB)", True) +check("cpp_deep modules bundled", len(cpp) >= 10, f"{len(cpp)} .py files") +check("[cpp] extra -> libclang", "libclang" in meta) +check("skill docs bundled", len(skills) >= 1, f"{len(skills)} files") +check("cpp-deep resolver present", "graphify/cpp_deep/resolver.py" in names) + +if not ok: + print("\n Wheel is INCOMPLETE - do not distribute.", file=sys.stderr); sys.exit(1) +print(f"\n[DONE] {whl}") +print(" Copy this wheel + install.sh/install.bat to the target machine.") +PYEOF diff --git a/docs/cpp-deep-enhancement-design.md b/docs/cpp-deep-enhancement-design.md new file mode 100644 index 000000000..85b058a6b --- /dev/null +++ b/docs/cpp-deep-enhancement-design.md @@ -0,0 +1,116 @@ +# Graphify C/C++ 增强设计方案 + +> 目标:让 Graphify 对 C/C++ 生成的知识图谱能够 (a) 完整遍历顶层→底层调用链,含**宏展开**与**跨 DLL/模块**调用;(b) 社区聚合度更高;(c) 更少重复节点(头文件分离导致的重复)。 + +## 决策(已确认) + +- 深层引擎:**libclang (`clang.cindex`)** — 直接拿完全预处理后的 AST(宏已实例化)+ 语义级 USR 符号身份。 +- 工程类型:**Windows `.sln` / `.vcxproj`**,先转换/提取为 `compile_commands.json`(含 include 路径与宏定义)。 +- 跨 DLL:**所有 DLL 均有源码** → 做**源码级跨模块链接**(caller → 其它模块内的定义),按模块/DLL 分组成调用链。 +- 落地:**opt-in 增强模块 + 新 flag**,默认行为不变;libclang 缺失时优雅回退到纯 tree-sitter。 + +## 现状关键事实(研究结论) + +- C/C++ 全走通用 `_extract_generic`(`_C_CONFIG`/`_CPP_CONFIG`);无专用 C/C++ 提取体。 +- **宏是最大缺口**:tree-sitter 不展开宏;`preproc_def`/`preproc_function_def` 不产生任何节点;函数式宏调用点变成解析不到的悬空 raw_call。 +- **代码节点去重只按 ID**(`dedup.py` 对 `file_type=="code"` 跳过 label 合并)。重复头文件符号只有共享 canonical ID 才会合并。ID 由 `ids.py make_id` + `_file_stem`(去扩展名)决定。 +- 已有干净扩展点:`resolver_registry.register(LanguageResolver(name, suffixes, resolve_fn))`,`resolve_fn(per_file, all_nodes, all_edges)` 在 `extract.py:16827` 自动运行。现有 `_resolve_cpp_member_calls` 即样板。 +- 跨模块外部符号唯一 seam:`global_graph.py:121-142` 按 label 合并 `source_file`-空的 external 节点;C 家族在 `build.py:596 _LANG_FAMILY` 同一桶,impl→header 的 INFERRED calls 可存活。 +- 内聚度 = 社区内边密度(`cluster.py:257`)。加入**真实且类型正确**的跨文件/跨模块边会直接提升内聚度。 +- 环境:`clang.cindex` 当前未安装、PATH 无 clang;Python 3.14。CLI 为手写 `sys.argv` 解析。 + +## 总体架构:四阶段,新增独立模块 + +新增包 `graphify/cpp_deep/`(不污染 16939 行的 `extract.py`),通过 resolver 注册接入: + +``` +graphify/cpp_deep/ + __init__.py # register_cpp_deep_resolver(); 特性开关与 libclang 探测 + compile_db.py # .sln/.vcxproj → compile_commands.json;include/宏推断回退 + clang_index.py # libclang 封装:TU 解析、USR、宏实例化、调用/引用游标 + macro.py # 宏节点 + 宏展开边(定义→展开目标,调用点→真实 callee) + crossmod.py # 跨模块/DLL 调用链:按 USR 的源码级链接 + 模块分组 + merge.py # 把 clang 结果并入 all_nodes/all_edges,ID 对齐 tree-sitter + usr_id.py # USR → 稳定 canonical ID(TU 无关,消重复头文件节点) +``` + +### Stage 1 — tree-sitter(保持不变) +初步语法解析,产出现有节点/边。作为 libclang 不可用时的回退基线。 + +### Stage 2 — libclang 深层解析(新) +1. **compile_commands.json 获取**(`compile_db.py`): + - 若已存在 → 直接用。 + - 否则解析 `.sln` 找 `.vcxproj`;从 vcxproj 读 `AdditionalIncludeDirectories`、`PreprocessorDefinitions`、`AdditionalOptions`、语言标准,合成每个 TU 的编译命令(MSVC 模式 `clang-cl` 兼容参数)。 + - 都没有 → 启发式:扫描工程推断 include 根目录 + 常见宏,尽力解析,失败文件回退 tree-sitter。 +2. **TU 解析**(`clang_index.py`):对每个翻译单元 `clang.cindex.Index.parse(..., options=DETAILED_PREPROCESSING_RECORD | ...)`。 + - 用 **USR**(Unified Symbol Resolution)作为符号的 TU 无关身份 → 解决“头文件分散、重复引用”导致的重复节点。 + - 采集:函数/方法定义与声明、`CALL_EXPR` 调用边(游标级,精确到被调用定义的 USR)、类型引用。 + +### Stage 3 — 宏展开(新,`macro.py`) +- 为 `MACRO_DEFINITION` 建节点(object-like / function-like)。 +- 用 `DETAILED_PREPROCESSING_RECORD` 拿 `MACRO_INSTANTIATION`:在每个展开点,把**宏 → 展开后真实调用的函数**连边(`expands_to` / `calls`),并把宏节点挂到调用者上。 +- 对“宏生成的函数定义”(如 `DEFINE_HANDLER(foo)`):libclang 看到的是**展开后**的真实 `FUNCTION_DECL`,直接建函数节点与其内部调用边 — 补上 tree-sitter 完全丢失的部分。 +- 条件编译(`#ifdef`):libclang 只解析激活分支 → 消除 tree-sitter 双分支重复/矛盾节点。 + +### Stage 4 — 合并 + 调优(`merge.py` / `crossmod.py` + 复用现有 dedup/cluster) +1. **ID 对齐**(`usr_id.py`):USR → canonical ID,规则与 `_file_stem`/`make_id` 对齐,使 clang 节点与 tree-sitter 节点在同一符号上**同 ID**,从而: + - 复用现有 `_merge_decl_def_classes` + 按-ID 去重,自动折叠“头文件声明 + 源文件定义 + N 个包含点”的重复。 + - 头文件符号 ID 锚定到**声明所在头文件**的路径(TU 无关),彻底消除跨库重复头引用节点。 +2. **跨模块/DLL 调用链**(`crossmod.py`):clang 的 CALL_EXPR 已指向被调用定义的 USR;按 USR 在**全工程**内解析(即使定义在另一个 DLL 的源码中)→ EXTRACTED `calls` 边。给节点标注 `module`/`dll`(来自 vcxproj 输出目标),供调用链按 DLL 分组遍历与可视化。 +3. **调优**: + - 边优先 EXTRACTED(clang 精确),覆盖/增强 tree-sitter 的 INFERRED。 + - 通过更多真实跨文件/跨模块边提升社区内聚度;宏与 include 折叠减少重复节点。 + - 复用现有 `cluster`(Leiden/Louvain)与 `dedup`,不改其算法,仅喂给它更干净的输入。 + +## 接入方式(opt-in) + +- 新 flag:`--cpp-deep`(在 `/graphify` 与 `graphify extract` 都识别)。可选 `--compile-db `、`--cpp-deep-macros`。 +- `graphify/cpp_deep/__init__.py` 在**导入时探测** libclang;`register` 仅在 `--cpp-deep` 且探测成功时激活 resolver。缺失时打印一次提示并回退,绝不阻塞。 +- 依赖:`pyproject.toml` 增加 optional extra `cpp = ["libclang"]`(`pip install 'graphifyy[cpp]'`)。 + +## 落地步骤(增量、可验证) + +1. **脚手架 + 探测**:建 `cpp_deep/` 包、`--cpp-deep` flag、libclang 探测与回退;不产出边时全链路仍正常。 +2. **compile_db.py**:`.sln`/`.vcxproj` → `compile_commands.json`;单测覆盖一个样例 vcxproj。 +3. **clang_index.py + usr_id.py**:TU 解析、USR→ID、函数/调用采集;与 tree-sitter 节点 ID 对齐的单测。 +4. **macro.py**:宏节点 + 展开边 + 宏生成函数;样例含函数式宏与 `#ifdef`。 +5. **crossmod.py**:跨模块 `calls` 边 + `module`/`dll` 标注;多 vcxproj 样例形成跨 DLL 调用链。 +6. **merge.py**:并入 `all_nodes/all_edges`,跑通现有 dedup/cluster;端到端验证顶层→底层链路可 `graphify query`/`path` 遍历。 +7. **样例工程 + 集成测试**:`tests/fixtures/cpp_multi_dll/`(2~3 个互相调用、含宏、头文件分离的 DLL 源码工程),断言:宏展开链、跨 DLL 链、无重复头节点、内聚度提升。 + +## 风险与回退 + +- libclang 版本/DLL 定位:`clang_index.py` 统一处理 `Config.set_library_file`,失败即回退 tree-sitter。 +- MSVC 参数兼容:优先 `clang-cl` 驱动模式解析 vcxproj 参数。 +- 解析慢/大工程:TU 级缓存(复用现有 per-file cache 思路,key 含编译命令 hash)。 +- 全程 opt-in:任何阶段失败都不影响现有 `/graphify` 默认行为。 + +## 实现状态(已完成) + +包 `graphify/cpp_deep/` 已实现并通过测试(24 个测试全绿,含跨 DLL 集成测试): + +| 模块 | 行数 | 职责 | +|---|---|---| +| `__init__.py` | 105 | 特性开关 `GRAPHIFY_CPP_DEEP` + libclang 探测 + resolver 注册 | +| `resolver.py` | 53 | 注册入口;**运行时**再次校验开关(防跨调用泄漏) | +| `pipeline.py` | 106 | 从 baseline 节点重建 C/C++ 文件集与 root,编排四阶段 | +| `compile_db.py` | 372 | compile_commands.json / .vcxproj / 启发式 三级回退 | +| `clang_index.py` | 417 | libclang 定位 + TU 解析 + 游标遍历(符号/调用/宏) | +| `records.py` | 76 | 阶段间中间记录(不泄漏 clang 类型) | +| `usr_id.py` | 76 | USR/位置 → 与 tree-sitter 对齐的 canonical ID | +| `macro.py` | 106 | 宏节点 + `contains`/`uses` 边 | +| `crossmod.py` | 116 | 按 USR 的跨模块调用解析 + 模块归属(定义方优先) | +| `merge.py` | 165 | 折叠进 all_nodes/all_edges;EXTRACTED 覆盖 INFERRED | + +**接入点**:`extract.py` 在 `run_language_resolvers` 前调用 `cpp_deep.maybe_register()`(失败隔离);CLI `--cpp-deep` / `--compile-db` 设置环境变量;`pyproject.toml` 新增 `cpp = ["libclang>=16"]`。 + +**已验证的关键成果**: +- 宏 `DOUBLE` 成为一等节点(baseline 完全没有);`compute() --uses--> DOUBLE` 记录展开出处。 +- 函数式宏展开后的真实调用由 libclang 自动捕获(post-expansion AST)。 +- 头文件声明 + 源文件定义 = **单一节点**(ID 对齐);跨 TU 重复消除。 +- 跨 DLL 调用 `AppExe::compute --calls--> UtilLib::helper` 标为 `cross_module_call`(EXTRACTED),节点带 `module` 标签,归属取**定义方**模块。 +- baseline 的 INFERRED 调用被精确 EXTRACTED 边取代(去重、提升社区内聚度)。 + +**测试**:`tests/test_cpp_deep.py`(gate/ID 对齐/compile_db,无需 libclang)+ `tests/test_cpp_deep_integration.py`(多 DLL 端到端,libclang 缺失时自动 skip)。 + +**已知后续项**:`.sln` 顺序解析(当前用 rglob 全量扫 vcxproj,功能等价)、TU 级解析缓存、二进制 PE 导入/导出表(当前仅源码级,符合"全部有源码"的前提)。 diff --git a/docs/cpp_deep_code_review_checklist.md b/docs/cpp_deep_code_review_checklist.md new file mode 100644 index 000000000..1b45096c1 --- /dev/null +++ b/docs/cpp_deep_code_review_checklist.md @@ -0,0 +1,749 @@ +# Graphify cpp_deep — 全提交代码审核清单(含代码+diff) + +**审查范围**: `c67bf2f` → `ac54db6`(8 个提交，`graphify/cpp_deep/` 全部模块) +**审查日期**: 2026-07-09 +**总计**: **6 HIGH + 18 MEDIUM + 12 LOW = 36 项** + +--- + +## 一、HIGH Priority(6 项) + +--- + +### H1. `merge.py:95` — Edge target uses `s.node_id` not `canonical_id` + +**提交**: `8ff714e`(支持自动遍历工程文件) +**严重度**: HIGH +**影响**: 跨目录 USR 去重后，`contains`/`method` 边指向不存在的旧 node_id(dangling edge)。定义在 `src/foo.cpp`，USR 映射后节点 id 为 `foo_src_foo()`，但边 target 指向 `foo_include_foo()`(不存在)。 + +**当前代码**: + +```python +# merge.py:92-103 +if s.parent_id and s.parent_id in existing_ids: + all_edges.append({ + "source": s.parent_id, + "target": s.node_id, # ← BUG: 应为 canonical_id + "relation": s.parent_relation, + "context": "cpp_deep", + "confidence": "EXTRACTED", + "confidence_score": 1.0, + "source_file": s.source_file, + "source_location": f"L{s.line}" if s.line else None, + "weight": 1.0, + }) +``` + +**修改建议**: + +```diff + if s.parent_id and s.parent_id in existing_ids: ++ # When USR folds the symbol to a different canonical id, the edge ++ # target must match the node we actually created. ++ edge_target = canonical_id + all_edges.append({ + "source": s.parent_id, +- "target": s.node_id, ++ "target": edge_target, + "relation": s.parent_relation, +``` + +--- + +### H2. `pipeline.py:130-133` — 缓存恢复无文件内容失效检查 + +**提交**: `cbd6471`(缓存崩溃恢复) +**严重度**: HIGH +**影响**: 解析中途崩溃 → 修改源码 → 重跑，已缓存的 TU 返回过时结果。 + +**当前代码**: + +```python +# pipeline.py:128-133 +def _parse_and_cache(unit): + from graphify.cpp_deep.cache import exists as _cache_exists, write as _cache_write + t0 = time.perf_counter() + if _cache_exists(str(unit.path), unit.args): # ← 只检查文件存在,不检查内容是否变化 + elapsed = 0.0 + diags: list[str] = [] + return unit, diags, elapsed, True +``` + +**修改建议**: + +```diff + # cache.py: cache_key() 中加入文件内容 hash + def cache_key(tu_path: str, args: list[str]) -> str: + h = hashlib.sha256() + h.update(Path(tu_path).resolve().as_posix().encode()) + h.update(b"\x00") + h.update(" ".join(sorted(args)).encode()) ++ # Include source-file content hash so editing the file invalidates the cache. ++ try: ++ h.update(Path(tu_path).read_bytes()) ++ except Exception: ++ pass # if the file can't be read, the parse will fail anyway + return h.hexdigest()[:16] +``` + +--- + +### H3. `extract.py:16827-16831` — Bare `except Exception: pass` + +**提交**: `c67bf2f`(初始脚手架) +**严重度**: HIGH +**影响**: 吞掉 `MemoryError`、`RecursionError` 及 `maybe_register()` 内部的任何 bug。 + +**当前代码**: + +```python +# extract.py:16827-16831 +try: + from graphify.cpp_deep import maybe_register as _cpp_deep_maybe_register + _cpp_deep_maybe_register() +except Exception: + pass +``` + +**修改建议**: + +```diff + try: + from graphify.cpp_deep import maybe_register as _cpp_deep_maybe_register + _cpp_deep_maybe_register() +-except Exception: ++except ImportError: + pass ++except Exception: ++ import logging ++ _log = logging.getLogger(__name__) ++ _log.warning("cpp_deep registration failed; falling back to tree-sitter", exc_info=True) +``` + +--- + +### H4. `__init__.py` + `resolver.py` — Env var gate 进程内泄漏 + +**提交**: `c67bf2f`(初始脚手架)，已在 `b063306` 部分修复 +**严重度**: HIGH(已在最新代码中缓解) +**状态**: ✅ 已有运行时 `is_enabled()` 重检，但 `GRAPHIFY_CPP_DEEP` env var 仍未在第二次调用前清除 + +**当前代码**(已修复后的 resolver.py): + +```python +# resolver.py:38-42 +def resolve_cpp_deep(per_file, all_nodes, all_edges): + from graphify.cpp_deep import is_enabled + if not is_enabled(): # ← 运行时重检(后续 commit 加入) + return +``` + +**结论**: 当前代码已通过运行时重检缓解。建议额外在 `extract()` 结束时恢复 env var。 + +```diff +# extract.py:16822-16831 建议 + try: + from graphify.cpp_deep import maybe_register as _cpp_deep_maybe_register + _cpp_deep_maybe_register() + except ImportError: + pass ++finally: ++ # Restore the env var so a subsequent extract() without --cpp-deep ++ # does not leak the deep pass. ++ os.environ.pop("GRAPHIFY_CPP_DEEP", None) +``` + +--- + +### H5. `cache.py:72-73` — `write()` bare `except:pass` + +**提交**: `a2539f6`(缓存架构) +**严重度**: HIGH +**影响**: 磁盘满/权限拒绝 → 缓存文件未写入 → merge 阶段静默丢失该 TU。 + +**当前代码**: + +```python +# cache.py:67-74 +try: + payload = json.dumps(result.to_dict(), ensure_ascii=False, separators=(",", ":")) + tmp = cp.with_suffix(".tmp") + tmp.write_text(payload, encoding="utf-8") + tmp.replace(cp) +except Exception: + pass +``` + +**修改建议**: + +```diff + try: + payload = json.dumps(result.to_dict(), ensure_ascii=False, separators=(",", ":")) + tmp = cp.with_suffix(".tmp") + tmp.write_text(payload, encoding="utf-8") + tmp.replace(cp) + except Exception: +- pass ++ _LOG.warning("cpp-deep cache: failed to write %s", Path(result.path).name) +``` + +--- + +### H6. `cache.py:95-108` — `load_all()` 死代码+文档误导 + +**提交**: `a2539f6`(缓存架构) +**严重度**: HIGH(死代码+文档声称流式实为全量) + +**当前代码**: + +```python +# cache.py:95-108 +def load_all(unit_paths: list[str], args: list[str]) -> list[TUResult]: + """Stream-load all cached TUResults in batch order. + Each file is loaded and immediately yielded so memory holds at most one + deserialised TUResult at a time. # ← 文档: "at most one" (假的) + """ + results: list[TUResult] = [] # ← 实现: 全量列表 + for path in unit_paths: + r = load(path, args) + if r is not None: + results.append(r) + else: + _LOG.warning("cpp-deep cache: missing entry for %s", Path(path).name) + return results # ← 一次返回全部 +# 全局搜索: 无人调用此函数 +``` + +**修改建议**: + +```diff +-def load_all(unit_paths: list[str], args: list[str]) -> list[TUResult]: +- """Stream-load all cached TUResults in batch order. +- +- Each file is loaded and immediately yielded so memory holds at most one +- deserialised TUResult at a time. +- """ +- results: list[TUResult] = [] ++def load_all(unit_paths: list[str], args: list[str]) -> Generator[TUResult, None, None]: ++ """Yield cached TUResults one at a time so memory holds at most one result.""" + for path in unit_paths: + r = load(path, args) + if r is not None: +- results.append(r) ++ yield r + else: + _LOG.warning("cpp-deep cache: missing entry for %s", Path(path).name) +- return results +``` + +如果保持死代码状态，则直接删除。 + +--- + +## 二、MEDIUM Priority(18 项) + +--- + +### M1. `merge.py:94` — Edge source 未 canonicalize + +**提交**: `8ff714e` +**当前代码**: 同 H1 的 92-103 行 +**影响**: 方法所属类被 USR 折叠到另一文件时，`contain`/`method` 边静默丢弃 + +```diff + if s.parent_id and s.parent_id in existing_ids: ++ # If the class was itself USR-remapped, use its canonical id as the source. ++ edge_source = usr_to_id.get(getattr(s, 'parent_usr', ''), s.parent_id) + all_edges.append({ +- "source": s.parent_id, ++ "source": edge_source if edge_source in existing_ids else s.parent_id, + "target": canonical_id, +``` + +--- + +### M2. `pipeline.py:275-288` — 方法节点反推 stem 错误 + +**提交**: `c67bf2f` +**当前代码**: + +```python +# pipeline.py: _baseline_file_ids +if not (isinstance(label, str) and label.endswith("()")): + continue +name = label[:-2].lstrip(".") +# ... 剥离 suffix = "_" + normalize_id(name) +``` + +**影响**: `Foo::bar()` → id=`"src_main_foo_bar"`, 剥离 `_bar` → `"src_main_foo"`(应是 `"src_main"`) + +```diff + if not (isinstance(label, str) and label.endswith("()")): + continue + name = label[:-2].lstrip(".") + if not name: + continue ++ # Skip method nodes (label starts with ".") — they have an extra ++ # class-name segment in their id that we can't strip cleanly. ++ # File-level functions give us the correct stem directly. ++ if label.startswith("."): ++ continue + suffix = "_" + normalize_id(name) + sid = str(nid) + if sid.endswith(suffix): +``` + +--- + +### M3. `clang_index.py:449` — extent 匹配只用 basename + +**提交**: `c67bf2f` +**当前代码**: + +```python +# clang_index.py:449-450 +fname = Path(loc.file.name).name +line = loc.line +for ext_file, start, end, fid in extents_by_line: + if ext_file == fname and start <= line <= end: # ← basename only +``` + +**修改建议**: + +```diff +-fname = Path(loc.file.name).name ++fname = Path(loc.file.name).resolve().as_posix() # full path for disambiguation + line = loc.line + for ext_file, start, end, fid in extents_by_line: + if ext_file == fname and start <= line <= end: +``` + +相应的 `func_extents` 记录处也需改为完整路径: + +```diff +# clang_index.py: func_extents 记录处 + func_extents.append(( +- Path(ext.start.file.name).name, ++ Path(ext.start.file.name).resolve().as_posix(), + ext.start.line, ext.end.line, nid, + )) +``` + +--- + +### M4. `pipeline.py:337-350` — `_common_root` 推断与 baseline 不一致 + +**提交**: `c67bf2f` +**影响**: fallback ID 计算用错误的 root + +```diff +# pipeline.py: _collect_cpp_files + root = _common_root(files) ++ _LOG.debug("cpp-deep: inferred scan root: %s", root) + return sorted(files), root +``` + +此问题的主要缓解已在后续 commit(`b063306`)通过 `_baseline_file_ids` + `stem_of` 回调实现——大部分符号 ID 不再依赖 `_common_root`。遗留风险仅限于无基线符号的文件(如独立头文件)。 + +--- + +### M5. `cache.py:50` — `sorted(args)` 破坏有序 flag + +**提交**: `a2539f6` +**当前代码**: + +```python +# cache.py:48-51 +h.update(Path(tu_path).resolve().as_posix().encode()) +h.update(b"\x00") +h.update(" ".join(sorted(args)).encode()) # ← sorted() 破坏 -include / -I 顺序 +``` + +**修改建议**: + +```diff + h.update(Path(tu_path).resolve().as_posix().encode()) + h.update(b"\x00") +-h.update(" ".join(sorted(args)).encode()) ++h.update(" ".join(args).encode()) # preserve flag order +``` + +--- + +### M6. `pipeline.py:222-227` — Cache miss 只有 LOG.warning + +**提交**: `a2539f6` +**当前代码**: + +```python +# pipeline.py: merge 阶段 +else: + _LOG.warning("cpp-deep: cache miss for %s", unit.path.name) +``` + +**修改建议**: + +```diff + else: +- _LOG.warning("cpp-deep: cache miss for %s", unit.path.name) ++ _LOG.error("cpp-deep: cache miss for %s — TU will be absent from the graph", unit.path.name) ++ cache_misses.append(unit.path.name) ++# After the merge loop: ++if cache_misses: ++ _LOG.error("cpp-deep: %d TUs had cache misses and are MISSING from the graph", len(cache_misses)) +``` + +--- + +### M7. `pipeline.py` — 串行路径缺少 SLOW 标签 + +**提交**: `f4e73cc` +**当前代码**: + +```python +# pipeline.py: 串行路径 +_log_progress(i + 1, unit.path.name, elapsed, cached=cached) +``` + +并行路径的 `_log_progress` 已有 `elapsed > 30` 检查(第145行)，串行路径的日志也走同一个 `_log_progress`。✅ **此问题已不存在**。 + +--- + +### M8. `pipeline.py` — `_maybe_gc(force)` 无效 + 线程时机错误 + +**提交**: `b063306` +**当前代码**: + +```python +# pipeline.py:151-155 +def _maybe_gc(force: bool = False) -> None: + if force: + gc.collect() +``` + +**修改建议**: + +```diff +-def _maybe_gc(force: bool = False) -> None: +- """...""" +- if force: +- gc.collect() ++def _maybe_gc() -> None: ++ """Force a GC cycle. Called from the main thread after Futures are consumed.""" ++ gc.collect() +``` + +调用处同步简化: + +```diff + if done_count % _GC_INTERVAL == 0: +- _maybe_gc(force=True) ++ _maybe_gc() +``` + +--- + +### M9. `compile_db.py:131` — compile_commands.json 路径无根范围检查 + +**提交**: `c67bf2f` +**当前代码**: + +```python +# compile_db.py:131 +src = (directory / file_field).resolve() if not Path(file_field).is_absolute() \ + else Path(file_field).resolve() +``` + +**修改建议**: + +```diff + src = (directory / file_field).resolve() if not Path(file_field).is_absolute() \ + else Path(file_field).resolve() ++# Validate that the resolved path is within the expected root ++try: ++ src.relative_to(root) # raises ValueError if outside root ++except ValueError: ++ _LOG.warning("cpp-deep: compile_db entry %s is outside scan root, skipping", src) ++ continue +``` + +--- + +### M10. `compile_db.py:201-204` — `_abspath` 无路径验证 + +**提交**: `c67bf2f` + +```diff + def _abspath(p: str, directory: Path) -> str: + p = p.strip('"') + pp = Path(p) +- return str(pp if pp.is_absolute() else (directory / pp).resolve()) ++ resolved = pp if pp.is_absolute() else (directory / pp).resolve() ++ # Log a warning for include dirs outside the project boundary ++ if not str(resolved).startswith(str(directory.resolve())): ++ _LOG.debug("cpp-deep: external include dir %s", resolved) ++ return str(resolved) +``` + +--- + +### M11. `clang_index.py:42-52` — `GRAPHIFY_LIBCLANG` 不加验证的 DLL 加载 + +**提交**: `c67bf2f` + +```diff + env = os.environ.get(_LIB_ENV, "").strip() + if env: ++ if not Path(env).is_file(): ++ _LOG.warning("GRAPHIFY_LIBCLANG=%s is not a file, ignoring", env) ++ else: + cands.append(env) +``` + +--- + +### M12. `cache.py:89` — `exists()` 让异常传播 + +**提交**: `a2539f6` + +```diff + def exists(tu_path: str, args: list[str]) -> bool: + """True if this TU has a valid cache file.""" +- cp = cache_path(tu_path, args) +- return cp is not None and cp.is_file() ++ try: ++ cp = cache_path(tu_path, args) ++ return cp is not None and cp.is_file() ++ except Exception: ++ return False +``` + +--- + +### M13. `__main__.py:4668-4678` — FileHandler 创建失败静默吞 + +**提交**: `ac54db6` + +```diff + try: + _fh = logging.FileHandler(...) + ... + print(f"[graphify extract] Progress log: {graphify_out / 'cpp_deep.log'}") + except Exception: +- pass ++ import traceback ++ _LOG = logging.getLogger(__name__) ++ _LOG.warning("Failed to create progress log file: %s", traceback.format_exc()) +``` + +--- + +### M14. `cache.py:72` — `tmp.replace(cp)` Windows 上不原子 + +**提交**: `a2539f6` +**说明**: Windows 上 `Path.replace()` 在目标已存在时非原子。可用重试+回退策略: + +```diff + tmp.write_text(payload, encoding="utf-8") +-tmp.replace(cp) ++try: ++ tmp.replace(cp) ++except OSError: ++ # Windows: target may be locked by AV/another process ++ cp.unlink(missing_ok=True) ++ tmp.rename(cp) +``` + +--- + +### M15. `records.py:38-124` — `from_dict(**d)` 无输入验证 + +**提交**: `a2539f6` + +```diff + @classmethod + def from_dict(cls, d: dict[str, Any]) -> SymbolRecord: ++ # Keep only known fields to reject extra/crafted keys ++ known = {"usr", "node_id", "label", "kind", "source_file", "line", ++ "module", "is_definition", "parent_id", "parent_relation"} ++ filtered = {k: v for k, v in d.items() if k in known} +- return cls(**d) ++ return cls(**filtered) +``` + +(同样适用于 `CallRecord`、`MacroRecord`、`MacroUse`) + +--- + +### M16. `b063306` — Commit message 误导("修bug"实为性能优化) + +**当前 message**: `[Update]增强cpp-deep内存优化、并行解析、进度日志、Python工具链和Roslyn原型` 中 claiming "修复known_macros空集合导致宏use边全丢的bug" + +**建议**: 无需修改代码。纯文档/commit message 问题，已记录。后续 commit message 建议更精确地区分 "fix" 和 "perf"。 + +--- + +### M17. `pipeline.py:34-44` — `_status_file().mkdir()` 副作用 + +**提交**: `799fe60` + +```diff + def _status_file() -> Path | None: + env = os.environ.get("GRAPHIFY_OUT", "").strip() + base = Path(env) if env else Path("graphify-out") ++ # Don't create the directory tree if it doesn't exist yet — ++ # the output directory is created by the main extract flow. ++ if not base.exists(): ++ return None +- base.mkdir(parents=True, exist_ok=True) ++ if not base.is_dir(): ++ return None + return base / ".cpp_deep_status.json" +``` + +--- + +### M18. `compile_db.py:156-158` — `_split_command` 不处理引号 + +**提交**: `c67bf2f` + +```diff ++import shlex ++ + def _split_command(command: str) -> list[str]: + """Split a shell-like compiler command line into arguments.""" +- return [tok for tok in re.split(r"\s+", command.strip()) if tok] ++ try: ++ return shlex.split(command.strip(), posix=False) ++ except ValueError: ++ # Fall back to the simple split for malformed commands ++ return [tok for tok in re.split(r"\s+", command.strip()) if tok] +``` + +--- + +## 三、LOW Priority(12 项) + +--- + +### L1. `compile_db.py:138-139` — bare `except: continue` + +```diff + except Exception: ++ _LOG.debug("cpp-deep: skipping malformed compile_db entry: %s", exc_info=True) + continue +``` + +### L2. `clang_index.py:200,210` — Index/TU 未显式释放 + +**状态**: ✅ 已在 `b063306` 修复(`del tu`) + +### L3. `compile_db.py:222-228` — 未映射 MSVC 标准值静默 + +```diff +-return m.get(std.lower(), "") ++result = m.get(std.lower(), "") ++if not result and std: ++ _LOG.debug("cpp-deep: unmapped MSVC std '%s', falling back to clang default", std) ++return result +``` + +### L4. `clang_index.py:384` — 不捕获 `CXX_OPERATOR_CALL_EXPR` + +```diff +# _walk_calls 函数中, CALL_EXPR 检查后增加 + if node.kind == _CURSORKIND.CALL_EXPR: + ... ++elif node.kind == _CURSORKIND.CXX_OPERATOR_CALL_EXPR: ++ callee = node.referenced ++ if callee is not None: ++ result.calls.append(CallRecord( ++ caller_id=caller_id, ++ callee_usr=callee.get_usr() or "", ++ callee_name=callee.spelling or node.spelling or "", ++ source_file=_rel(node.location.file.name) if node.location and node.location.file else "", ++ line=node.location.line if node.location else 0, ++ )) +``` + +### L5. `clang_index.py:432` — 全 TU walk 的单异常中止 + +```diff +# _walk 中的 get_children() 循环改为逐子防御: + for child in cursor.get_children(): +- k = child.kind +- # class/struct ... ++ try: ++ k = child.kind ++ # class/struct ... ++ except Exception as child_exc: ++ result.diagnostics.append(f"child walk failed: {child_exc}") ++ continue +``` + +### L6. `clang_index.py:222-226` — `_rel` fallback 返回绝对路径 + +```diff + def _rel(path: str) -> str: + try: + return Path(path).resolve().relative_to(root).as_posix() + except Exception: ++ _LOG.debug("cpp-deep: path %s outside root %s, using absolute", path, root) + return Path(path).as_posix() +``` + +### L7. `usr_id.py:73-76` — file_node_id fallback 与基线不一致 + +**状态**: ✅ 已通过 `file_id_of` 回调缓解(当基线文件节点存在时不走 fallback)。仅影响无基线文件节点的文件(罕见)。 + +### L8. `macro.py:69` — fallback file_node_id 边悬空 + +```diff +-file_id = m.file_id or file_node_id(m.source_file) +-_add_edge(file_id, m.node_id, "contains", m.source_file, m.line, "macro_def") ++file_id = m.file_id or file_node_id(m.source_file) ++if file_id in valid_file_ids: # 只对实际存在的文件节点添加边 ++ _add_edge(file_id, m.node_id, "contains", m.source_file, m.line, "macro_def") +``` + +### L9. `clang_index.py:470-477` — `del tu` 不保证同步释放 + +**状态**: 已部分缓解。完整修复需要直接调用 ctypes 的 dispose 函数。 + +```python +# 更可靠的方式(如果 binding 支持): +from clang.cindex import TranslationUnit_dispose +TranslationUnit_dispose(tu) +``` + +### L10. `cache.py:70-72` — .tmp 孤儿文件 + +```diff +# 在 write() 中, replace 前加入清理逻辑 ++tmp_orphan = cp.with_suffix(".tmp") ++if tmp_orphan.exists(): ++ tmp_orphan.unlink(missing_ok=True) +``` + +或在 `clear()` 中一并清理。 + +### L11. `pipeline.py:67-69` — `_write_status` .tmp 残留 + +**同 L10**，状态文件的原子写入也有同样问题。 + +### L12. `__main__.py:4636,4670` — 重复 main() 堆积 handler + +```diff + # __main__.py:4636 — 加 guard + _log = logging.getLogger("graphify.cpp_deep") + _log.setLevel(logging.INFO) + _log.propagate = False ++if not _log.handlers: # prevent duplicates on repeated main() calls + _sh = logging.StreamHandler(sys.stderr) + _sh.setFormatter(_fmt) + _log.addHandler(_sh) +``` + +--- + +*审查由 open-code-review 插件 + 两个 Explore 代理并行执行。代码行与 diff 均基于当前 working tree 验证。* diff --git a/docs/translations/README.fa-IR.md b/docs/translations/README.fa-IR.md index 2851c0803..87ae87b6c 100644 --- a/docs/translations/README.fa-IR.md +++ b/docs/translations/README.fa-IR.md @@ -1,5 +1,5 @@

- Graphify + Graphify

@@ -511,11 +511,11 @@ graphify --version ## ساخته‌شده روی graphify — Penpax -[**Penpax**](https://graphify.com) لایه همیشه‌روشن ساخته‌شده روی graphify است — همان رویکرد گراف را بر کل زندگی کاری شما اعمال می‌کند: جلسات، تاریخچه مرورگر، ایمیل‌ها، فایل‌ها و کد، به‌صورت مداوم در پس‌زمینه به‌روزرسانی می‌شود. +[**Penpax**](https://graphifylabs.ai) لایه همیشه‌روشن ساخته‌شده روی graphify است — همان رویکرد گراف را بر کل زندگی کاری شما اعمال می‌کند: جلسات، تاریخچه مرورگر، ایمیل‌ها، فایل‌ها و کد، به‌صورت مداوم در پس‌زمینه به‌روزرسانی می‌شود. ساخته‌شده برای کسانی که کارشان در صدها مکالمه و سند پراکنده است. بدون ابر، کاملاً روی دستگاه. -**آزمایش رایگان به‌زودی راه‌اندازی می‌شود.** [به لیست انتظار بپیوندید ←](https://graphify.com) +**آزمایش رایگان به‌زودی راه‌اندازی می‌شود.** [به لیست انتظار بپیوندید ←](https://graphifylabs.ai) --- diff --git a/docs/translations/README.fil-PH.md b/docs/translations/README.fil-PH.md index e318eb68e..6a8ef4cda 100644 --- a/docs/translations/README.fil-PH.md +++ b/docs/translations/README.fil-PH.md @@ -72,6 +72,6 @@ Ang mga code file ay prinoseso nang lokal sa pamamagitan ng tree-sitter AST. Ang ## Binuo sa ibabaw ng graphify — Penpax -Ang [**Penpax**](https://graphify.com) ay ang enterprise layer sa ibabaw ng graphify. **Malapit nang magkaroon ng libreng trial.** [Sumali sa waitlist →](https://graphify.com) +Ang [**Penpax**](https://graphifylabs.ai) ay ang enterprise layer sa ibabaw ng graphify. **Malapit nang magkaroon ng libreng trial.** [Sumali sa waitlist →](https://graphifylabs.ai) [![Star History Chart](https://api.star-history.com/svg?repos=safishamsi/graphify&type=Date)](https://star-history.com/#safishamsi/graphify&Date) diff --git a/docs/translations/README.he-IL.md b/docs/translations/README.he-IL.md index 65e9ae75d..c1c193e84 100644 --- a/docs/translations/README.he-IL.md +++ b/docs/translations/README.he-IL.md @@ -1,5 +1,5 @@

- Graphify + Graphify

@@ -777,11 +777,11 @@ graphify label ./my-project --backend=openai --model gpt-4o # כפיית backe ## נבנה על graphify — ‏Penpax -[**Penpax**](https://graphify.com) היא השכבה התמידית שנבנתה מעל graphify — היא מיישמת את אותה גישת גרף על כל חיי העבודה שלכם: פגישות, היסטוריית דפדפן, מיילים, קבצים וקוד, ומתעדכנת ברציפות ברקע. +[**Penpax**](https://graphifylabs.ai) היא השכבה התמידית שנבנתה מעל graphify — היא מיישמת את אותה גישת גרף על כל חיי העבודה שלכם: פגישות, היסטוריית דפדפן, מיילים, קבצים וקוד, ומתעדכנת ברציפות ברקע. נבנתה לאנשים שהעבודה שלהם מפוזרת על פני מאות שיחות ומסמכים שהם לעולם לא יוכלו לשחזר במלואם. ללא ענן, לגמרי על המכשיר. -**גרסת ניסיון חינמית תושק בקרוב.** [הצטרפו לרשימת ההמתנה ←](https://graphify.com) +**גרסת ניסיון חינמית תושק בקרוב.** [הצטרפו לרשימת ההמתנה ←](https://graphifylabs.ai) --- diff --git a/docs/translations/README.uk-UA.md b/docs/translations/README.uk-UA.md index 2fa632544..337a3c465 100644 --- a/docs/translations/README.uk-UA.md +++ b/docs/translations/README.uk-UA.md @@ -1,5 +1,5 @@

- Graphify + Graphify

@@ -529,11 +529,11 @@ graphify cluster-only ./my-project --exclude-hubs 99 # виключи ## Побудовано на graphify — Penpax -[**Penpax**](https://graphify.com) — це завжди активний шар поверх graphify, він застосовує той самий графовий підхід до всього робочого життя: зустрічей, історії браузера, email-ів, файлів і коду, постійно оновлюючись у фоновому режимі. +[**Penpax**](https://graphifylabs.ai) — це завжди активний шар поверх graphify, він застосовує той самий графовий підхід до всього робочого життя: зустрічей, історії браузера, email-ів, файлів і коду, постійно оновлюючись у фоновому режимі. Створений для людей, чия робота розкидана по сотнях розмов і документів, які неможливо повністю відтворити. Без хмари, повністю на пристрої. -**Безкоштовна пробна версія незабаром.** [Приєднайтесь до списку очікування →](https://graphify.com) +**Безкоштовна пробна версія незабаром.** [Приєднайтесь до списку очікування →](https://graphifylabs.ai) --- diff --git a/graphify/__main__.py b/graphify/__main__.py index 708c45819..d7467d4db 100644 --- a/graphify/__main__.py +++ b/graphify/__main__.py @@ -3,6 +3,7 @@ from __future__ import annotations import functools import json +import logging import os import platform import re @@ -23,114 +24,30 @@ # same override (#1423). from graphify.paths import GRAPHIFY_OUT as _GRAPHIFY_OUT -# Install/uninstall subsystem moved to graphify/install.py; re-exported here so -# `from graphify.__main__ import ` keeps working unchanged. -from graphify.install import ( # noqa: E402,F401 - dispatch_install_cli, - _agents_install, - _agents_platform_install, - _agents_platform_uninstall, - _agents_uninstall, - _always_on, - _amp_install, - _amp_legacy_cleanup, - _amp_uninstall, - _antigravity_finalize, - _antigravity_install, - _antigravity_uninstall, - _canonical_platform, - _claude_pretooluse_hooks, - _copy_skill_file, - _cursor_install, - _cursor_uninstall, - _devin_rules_install, - _devin_rules_uninstall, - _gemini_hook, - _install_claude_hook, - _install_codebuddy_hook, - _install_codex_hook, - _install_gemini_hook, - _install_kilo_plugin, - _install_opencode_plugin, - _install_skill_references, - _kilo_config_path, - _kilo_config_write_path, - _kilo_install, - _kilo_uninstall, - _kilo_uninstall_global, - _kiro_install, - _kiro_uninstall, - _load_json_like, - _packaged_skill_refs_dir, - _platform_skill_destination, - _print_banner, - _print_install_usage, - _print_project_git_add_hint, - _project_install, - _project_scope_root, - _project_uninstall, - _project_uninstall_all, - _refresh_all_version_stamps, - _remove_claude_skill_registration, - _remove_skill_file, - _replace_or_append_section, - _resolve_graphify_exe, - _skill_registration, - _strip_graphify_hook, - _strip_graphify_md_section, - _strip_json_comments, - _uninstall_claude_hook, - _uninstall_codebuddy_hook, - _uninstall_codex_hook, - _uninstall_gemini_hook, - _uninstall_kilo_plugin, - _uninstall_opencode_plugin, - claude_install, - claude_uninstall, - codebuddy_install, - codebuddy_uninstall, - gemini_install, - gemini_uninstall, - install, - uninstall_all, - vscode_install, - vscode_uninstall, - _PLATFORM_ALIASES, - _CLAUDE_MD_MARKER, - _CODEBUDDY_MD_MARKER, - _AGENTS_MD_MARKER, - _GEMINI_MD_MARKER, - _VSCODE_INSTRUCTIONS_MARKER, - _ANTIGRAVITY_RULES_PATH, - _ANTIGRAVITY_WORKFLOW_PATH, - _ANTIGRAVITY_WORKFLOW, - _CURSOR_RULE_PATH, - _CURSOR_RULE, - _DEVIN_RULES_PATH, - _DEVIN_RULES, - _KILO_PLUGIN_JS, - _KILO_PLUGIN_PATH, - _KILO_CONFIG_JSON_PATH, - _KILO_CONFIG_JSONC_PATH, - _OPENCODE_PLUGIN_JS, - _OPENCODE_PLUGIN_PATH, - _OPENCODE_CONFIG_PATH, - _PLATFORM_CONFIG, -) -from graphify.cli import ( # noqa: E402,F401 - dispatch_command, - _StageTimer, - _clone_repo, - _default_graph_path, - _enforce_graph_size_cap_or_exit, - _run_hook_guard, - _SEARCH_NUDGE, - _READ_NUDGE, - _HOOK_SOURCE_EXTS, - _GEMINI_NUDGE_TEXT, -) +@functools.lru_cache(maxsize=None) +def _always_on(basename: str) -> str: + """Read a packaged always-on instruction block from graphify/always_on/. + The six always-on blocks (CLAUDE.md / AGENTS.md / GEMINI.md / VS Code + Copilot instructions / Antigravity rules / Kiro steering) live as committed + markdown next to this module, generated by tools/skillgen from a single + human-edited fragment and guarded against drift by ``skillgen --check``. The + installer injects them verbatim via ``_replace_or_append_section``, so the + bytes here must match the former triple-quoted constant exactly — the + always-on-roundtrip validator proves that. + """ + path = Path(__file__).parent / "always_on" / f"{basename}.md" + try: + return path.read_text(encoding="utf-8") + except OSError as exc: + # Defer to use-time so a missing/corrupt packaged block can't crash module + # import (which would brick every CLI command, not just install). Reached + # only by an install/integration path that actually needs this block. + raise RuntimeError( + f"graphify install is incomplete: missing always-on block '{basename}' " + f"at {path}. Reinstall graphifyy (e.g. `uv tool install --reinstall graphifyy`)." + ) from exc _ALWAYS_ON_ALIASES = { @@ -154,10 +71,53 @@ def __getattr__(name: str) -> str: raise AttributeError(f"module {__name__!r} has no attribute {name!r}") +def _default_graph_path() -> str: + return str(Path(_GRAPHIFY_OUT) / "graph.json") +class _StageTimer: + """Print per-stage wall-clock timings to stderr when --timing is set (#1490). + Monotonic (perf_counter), diagnostic-only: emits ``[graphify timing] : + N.Ns`` after each stage and a final total. Off by default, so normal output is + byte-identical and machine-read stdout is untouched. + """ + def __init__(self, enabled: bool) -> None: + import time as _time + self._now = _time.perf_counter + self.enabled = enabled + self.start = self._now() + self._last = self.start + + def mark(self, stage: str) -> None: + now = self._now() + if self.enabled: + print(f"[graphify timing] {stage}: {now - self._last:.1f}s", file=sys.stderr) + self._last = now + + def total(self) -> None: + if self.enabled: + print(f"[graphify timing] total: {self._now() - self.start:.1f}s", file=sys.stderr) + + +def _enforce_graph_size_cap_or_exit(gp: Path) -> None: + """Reject oversized graph files before parsing (CLI exit-on-fail flavor). + + Delegates to ``graphify.security.check_graph_file_size_cap`` and turns the + raised ``ValueError`` into a CLI-style ``error: ...`` message + exit 1. + Use this from ``__main__.py`` subcommands that already use the ``print + + sys.exit(1)`` idiom. Library/MCP/loader callers (``serve._load_graph``, + ``build``, ``benchmark``, ``tree_html``, ``callflow_html``, ``prs``, + ``global_graph``, ``watch``, ``export``) call the security helper directly + and let the ``ValueError`` propagate. + """ + from graphify.security import check_graph_file_size_cap + try: + check_graph_file_size_cap(gp) + except ValueError as exc: + print(f"error: {exc}", file=sys.stderr) + sys.exit(1) def _check_skill_version(skill_dst: Path) -> None: @@ -226,22 +186,254 @@ def _version_tuple(version: str) -> tuple[int, ...]: return tuple(parts) +def _refresh_all_version_stamps() -> None: + """After a successful install, update .graphify_version in all other known skill dirs. + Prevents stale-version warnings from platforms that were installed previously + but not explicitly re-installed during this upgrade. + """ + for name in _PLATFORM_CONFIG: + skill_dst = _platform_skill_destination(name) + vf = skill_dst.parent / ".graphify_version" + if skill_dst.exists(): + vf.write_text(__version__, encoding="utf-8") + + +def _platform_skill_destination(platform_name: str, *, project: bool = False, project_dir: Path | None = None) -> Path: + """Return the skill destination for a platform and scope.""" + if platform_name == "gemini": + if project: + return (project_dir or Path(".")) / ".gemini" / "skills" / "graphify" / "SKILL.md" + if platform.system() == "Windows": + return Path.home() / ".agents" / "skills" / "graphify" / "SKILL.md" + return Path.home() / ".gemini" / "skills" / "graphify" / "SKILL.md" + + if platform_name == "opencode": + if project: + return (project_dir or Path(".")) / ".opencode" / "skills" / "graphify" / "SKILL.md" + return Path.home() / ".config" / "opencode" / "skills" / "graphify" / "SKILL.md" + + if platform_name == "hermes": + if project: + return (project_dir or Path(".")) / ".hermes" / "skills" / "graphify" / "SKILL.md" + # On Windows, Hermes scans %LOCALAPPDATA%\hermes\skills, not ~/.hermes (#1403). + if platform.system() == "Windows": + local_appdata = Path(os.environ.get("LOCALAPPDATA") or (Path.home() / "AppData" / "Local")) + return local_appdata / "hermes" / "skills" / "graphify" / "SKILL.md" + return Path.home() / ".hermes" / "skills" / "graphify" / "SKILL.md" + + if platform_name == "devin": + if project: + return (project_dir or Path(".")) / ".devin" / "skills" / "graphify" / "SKILL.md" + return Path.home() / ".config" / "devin" / "skills" / "graphify" / "SKILL.md" + + if platform_name == "amp": + if project: + return (project_dir or Path(".")) / ".agents" / "skills" / "graphify" / "SKILL.md" + return Path.home() / ".config" / "agents" / "skills" / "graphify" / "SKILL.md" + + if platform_name == "agents": + # The generic Agent-Skills target: project ./.agents/skills, global the + # spec's user-global ~/.agents/skills (read by `npx skills` and compliant + # frameworks), NOT amp's ~/.config/agents/skills. + if project: + return (project_dir or Path(".")) / ".agents" / "skills" / "graphify" / "SKILL.md" + return Path.home() / ".agents" / "skills" / "graphify" / "SKILL.md" + + if platform_name in ("antigravity", "antigravity-windows"): + if project: + return (project_dir or Path(".")) / ".agents" / "skills" / "graphify" / "SKILL.md" + # Global Antigravity skill dir (all workspaces): ~/.gemini/config/skills/ + return Path.home() / ".gemini" / "config" / "skills" / "graphify" / "SKILL.md" + + cfg = _PLATFORM_CONFIG[platform_name] + if project: + return (project_dir or Path(".")) / cfg["skill_dst"] + + if platform_name in ("claude", "windows") and os.environ.get("CLAUDE_CONFIG_DIR"): + return Path(os.environ["CLAUDE_CONFIG_DIR"]) / "skills" / "graphify" / "SKILL.md" + return Path.home() / cfg["skill_dst"] + + +def _packaged_skill_refs_dir(platform_name: str) -> Path | None: + """Return the packaged references source dir for a progressive platform, else None. + + A platform opts into progressive disclosure by setting ``skill_refs`` in its + ``_PLATFORM_CONFIG`` entry. The value names a bundle under + ``graphify/skills//references/``. Reuse keys (e.g. trae-cn) point at + their twin's bundle. + + ``gemini`` has no ``_PLATFORM_CONFIG`` entry: it installs claude's + ``skill.md`` body verbatim (see ``_copy_skill_file``). Since that body is the + lean progressive core that links to ``references/``, gemini needs claude's + references/ sidecar too, or its SKILL.md ships with dead pointers. So gemini + resolves to the claude bundle rather than opting out. + + Bundles ship one platform-group at a time. A host whose bundle directory + ``graphify/skills//`` is not in this build has not gone progressive + yet, so this returns None and the host installs today's monolithic SKILL.md + with no references/ sidecar. Only when the bundle directory IS present does + this return the references path; if that directory then lacks its + ``references/`` subdir, ``_copy_skill_file`` hard-fails (a malformed bundle, + the empty-sidecar regression the wheel-content test also guards). + """ + if platform_name == "gemini": + bundle = "claude" + else: + bundle = _PLATFORM_CONFIG[platform_name].get("skill_refs") + if not bundle: + return None + bundle_dir = Path(__file__).parent / "skills" / bundle + if not bundle_dir.is_dir(): + return None + return bundle_dir / "references" + + +def _install_skill_references(skill_dst: Path, refs_src: Path) -> None: + """Atomically install a packaged references/ sidecar next to SKILL.md. + + Stages the packaged dir into ``references.tmp`` (copytree), drops any stale + ``references/`` already on disk, then ``os.replace``-renames the staged dir + into place. The rename is atomic on the same filesystem, so an interrupted + install never leaves a half-written references/ visible to the agent. + """ + refs_dst = skill_dst.parent / "references" + refs_staged = skill_dst.parent / "references.tmp" + if refs_staged.exists(): + shutil.rmtree(refs_staged) + try: + shutil.copytree(refs_src, refs_staged) + if refs_dst.exists(): + shutil.rmtree(refs_dst) + os.replace(refs_staged, refs_dst) + except Exception: + if refs_staged.exists(): + shutil.rmtree(refs_staged, ignore_errors=True) + raise + + +def _copy_skill_file(platform_name: str, *, project: bool = False, project_dir: Path | None = None) -> Path: + """Copy a packaged skill file and write its version stamp. + + For progressive platforms (those with ``skill_refs`` set), the packaged + ``references/`` sidecar is installed alongside SKILL.md and the single + ``.graphify_version`` stamp covers both. For monolith platforms (no + ``skill_refs``), any orphan ``references/`` left by a prior progressive + install is removed so the on-disk layout matches the package. + """ + skill_file = "skill.md" if platform_name == "gemini" else _PLATFORM_CONFIG[platform_name]["skill_file"] + skill_src = Path(__file__).parent / skill_file + if not skill_src.exists(): + print(f"error: {skill_file} not found in package - reinstall graphify", file=sys.stderr) + sys.exit(1) + + refs_src = _packaged_skill_refs_dir(platform_name) + if refs_src is not None and not refs_src.exists(): + # Progressive platform declared a references bundle that is missing from + # the package. Fail loud rather than silently shipping an empty sidecar. + print( + f"error: references for '{platform_name}' not found in package " + f"({refs_src}) - reinstall graphify", + file=sys.stderr, + ) + sys.exit(1) + + skill_dst = _platform_skill_destination(platform_name, project=project, project_dir=project_dir) + skill_dst.parent.mkdir(parents=True, exist_ok=True) + + # Install the references/ sidecar (or clear an orphan one) BEFORE writing + # SKILL.md, so SKILL.md is the last artifact laid down. An install that is + # interrupted partway then leaves no SKILL.md rather than a SKILL.md that + # points at an absent references/ dir. + if refs_src is not None: + _install_skill_references(skill_dst, refs_src) + print(f" references -> {skill_dst.parent / 'references'}") + else: + # Monolith (or progressive-with-no-refs): clear any orphan references/. + orphan_refs = skill_dst.parent / "references" + if orphan_refs.exists(): + shutil.rmtree(orphan_refs) + + # SKILL.md last (crash-safety), via an atomic temp + rename. + tmp_dst = skill_dst.with_suffix(skill_dst.suffix + ".tmp") + try: + shutil.copy(skill_src, tmp_dst) + os.replace(tmp_dst, skill_dst) + except Exception: + try: + tmp_dst.unlink(missing_ok=True) + except OSError: + pass + raise + + (skill_dst.parent / ".graphify_version").write_text(__version__, encoding="utf-8") + print(f" skill installed -> {skill_dst}") + return skill_dst + + +def _remove_skill_file(platform_name: str, *, project: bool = False, project_dir: Path | None = None) -> bool: + """Remove a platform skill file and its version stamp without touching other scopes.""" + skill_dst = _platform_skill_destination(platform_name, project=project, project_dir=project_dir) + removed = False + if skill_dst.exists(): + skill_dst.unlink() + print(f" skill removed -> {skill_dst}") + removed = True + version_file = skill_dst.parent / ".graphify_version" + if version_file.exists(): + version_file.unlink() + removed = True + refs_dir = skill_dst.parent / "references" + if refs_dir.exists(): + shutil.rmtree(refs_dir) + removed = True + for d in (skill_dst.parent, skill_dst.parent.parent, skill_dst.parent.parent.parent): + try: + d.rmdir() + except OSError: + break + return removed + + +def _project_scope_root(path: Path, project_dir: Path) -> Path: + """Return the top-level project artifact for a project-scoped skill path.""" + try: + rel = path.relative_to(project_dir) + except ValueError: + return path + return project_dir / rel.parts[0] if rel.parts else path - - - - - - - - - - - - - +def _remove_claude_skill_registration(project_dir: Path) -> None: + """Remove the project-scoped Claude skill registration file/section.""" + claude_md = project_dir / ".claude" / "CLAUDE.md" + if not claude_md.exists(): + return + content = claude_md.read_text(encoding="utf-8") + if "# graphify" not in content: + return + cleaned = re.sub(r"\n*# graphify\n.*?(?=\n# |\Z)", "", content, flags=re.DOTALL).rstrip() + if cleaned: + claude_md.write_text(cleaned + "\n", encoding="utf-8") + print(f" CLAUDE.md -> graphify skill registration removed from {claude_md}") + else: + claude_md.unlink() + print(f" CLAUDE.md -> deleted {claude_md}") + + +def _print_project_git_add_hint(paths: list[Path]) -> None: + unique: list[str] = [] + for path in paths: + text = path.as_posix().rstrip("/") + if path.exists() and path.is_dir(): + text += "/" + if text not in unique: + unique.append(text) + if not unique: + return + print() + print("Project-scoped install. Add to version control:") + print(f" git add {' '.join(unique)}") # PreToolUse nudge payloads, emitted verbatim by the shell-agnostic # `graphify hook-guard` subcommand (see _run_hook_guard). The previous hooks @@ -253,41 +445,413 @@ def _version_tuple(version: str) -> tuple[int, ...]: # additionalContext on PreToolUse (Codex Desktop does not — that path stays a # no-op via `hook-check`). Compact separators keep the payload byte-for-byte the # same JSON the old `echo` emitted. - +_SEARCH_NUDGE = json.dumps({ + "hookSpecificOutput": { + "hookEventName": "PreToolUse", + "additionalContext": ( + 'MANDATORY: graphify-out/graph.json exists. You MUST run ' + '`graphify query ""` before grepping raw files. Only grep ' + 'after graphify has oriented you, or to modify/debug specific lines.' + ), + } +}, ensure_ascii=False, separators=(",", ":")) + "\n" + +_READ_NUDGE = json.dumps({ + "hookSpecificOutput": { + "hookEventName": "PreToolUse", + "additionalContext": ( + 'MANDATORY: graphify-out/graph.json exists. You MUST run graphify ' + 'before reading source files. Use: `graphify query ""` ' + '(scoped subgraph), `graphify explain ""`, or ' + '`graphify path "" ""`. Only read raw files after graphify has ' + 'oriented you, or to modify/debug specific lines. This rule applies to ' + 'subagents too — include it in every subagent prompt involving code ' + 'exploration.' + ), + } +}, ensure_ascii=False, separators=(",", ":")) + "\n" # Source/doc extensions the Read|Glob guard nudges on (verbatim from the old hook). # The trailing-extension test (real final path segment, then its last '.') means # '.json' never false-matches '.js', and framework files like '.astro' are kept. +_HOOK_SOURCE_EXTS = ( + '.py', '.js', '.ts', '.tsx', '.jsx', '.astro', '.vue', '.svelte', '.go', + '.rs', '.java', '.rb', '.c', '.h', '.cpp', '.hpp', '.cc', '.cs', '.kt', + '.swift', '.php', '.scala', '.lua', '.sh', '.md', '.rst', '.txt', '.mdx', +) +def _claude_pretooluse_hooks() -> "list[dict]": + """graphify's Claude/Codebuddy PreToolUse hooks, resolved at install time. + The command invokes `graphify hook-guard ` via the absolute exe + path (`_resolve_graphify_exe`), so it parses under sh, cmd.exe and PowerShell + alike — this is the #522 fix, and mirrors the codex hook. Matchers stay "Bash" + and "Read|Glob" and the command always contains "graphify", so the existing + install/uninstall filters find and replace both old bash hooks and these. + """ + exe = _resolve_graphify_exe() + if " " in exe and not exe.startswith('"'): + exe = f'"{exe}"' + return [ + {"matcher": "Bash", + "hooks": [{"type": "command", "command": f"{exe} hook-guard search"}]}, + {"matcher": "Read|Glob", + "hooks": [{"type": "command", "command": f"{exe} hook-guard read"}]}, + ] + +def _skill_registration(skill_path: str = "~/.claude/skills/graphify/SKILL.md") -> str: + return ( + "\n# graphify\n" + f"- **graphify** (`{skill_path}`) " + "- any input to knowledge graph. Trigger: `/graphify`\n" + "When the user types `/graphify`, use the installed graphify skill " + "or instructions before doing anything else.\n" + ) + + +_PLATFORM_CONFIG: dict[str, dict] = { + "claude": { + "skill_file": "skill.md", + "skill_dst": Path(".claude") / "skills" / "graphify" / "SKILL.md", + "claude_md": True, + "skill_refs": "claude", + }, + "codex": { + "skill_file": "skill-codex.md", + "skill_dst": Path(".codex") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "codex", + }, + "opencode": { + "skill_file": "skill-opencode.md", + "skill_dst": Path(".config") / "opencode" / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "opencode", + }, + "kilo": { + "skill_file": "skill-kilo.md", + "skill_dst": Path(".config") / "kilo" / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "kilo", + }, + "aider": { + # Monolith: aider ships the full SKILL.md inline, no references/ sidecar. + "skill_file": "skill-aider.md", + "skill_dst": Path(".aider") / "graphify" / "SKILL.md", + "claude_md": False, + }, + "copilot": { + "skill_file": "skill-copilot.md", + "skill_dst": Path(".copilot") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "copilot", + }, + "claw": { + "skill_file": "skill-claw.md", + "skill_dst": Path(".openclaw") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "claw", + }, + "droid": { + "skill_file": "skill-droid.md", + "skill_dst": Path(".factory") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "droid", + }, + "trae": { + "skill_file": "skill-trae.md", + "skill_dst": Path(".trae") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "trae", + }, + "trae-cn": { + # Reuses trae's split bundle (same skill body + references). + "skill_file": "skill-trae.md", + "skill_dst": Path(".trae-cn") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "trae", + }, + "hermes": { + # Reuses claw's split bundle. + "skill_file": "skill-claw.md", + "skill_dst": Path(".hermes") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "claw", + }, + "kiro": { + "skill_file": "skill-kiro.md", + "skill_dst": Path(".kiro") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "kiro", + }, + "pi": { + "skill_file": "skill-pi.md", + "skill_dst": Path(".pi") / "agent" / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "pi", + }, + "codebuddy": { + # Reuses claude's split bundle (shares skill.md). + "skill_file": "skill.md", + "skill_dst": Path(".codebuddy") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "claude", + }, + "antigravity": { + # Rides claude's split bundle (shares skill.md). + "skill_file": "skill.md", + "skill_dst": Path(".agents") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "claude", + }, + "antigravity-windows": { + # Rides windows' split bundle. + "skill_file": "skill-windows.md", + "skill_dst": Path(".agents") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "windows", + }, + "windows": { + "skill_file": "skill-windows.md", + "skill_dst": Path(".claude") / "skills" / "graphify" / "SKILL.md", + "claude_md": True, + "skill_refs": "windows", + }, + "kimi": { + # Reuses claude's split bundle (shares skill.md). + "skill_file": "skill.md", + "skill_dst": Path(".kimi") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "claude", + }, + "amp": { + # Amp searches .agents/skills (project) and ~/.config/agents/skills (user), + # not .amp/skills. The user-scope path is set in _platform_skill_destination. + "skill_file": "skill-amp.md", + "skill_dst": Path(".agents") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "amp", + }, + "agents": { + # The generic cross-framework Agent-Skills target. Global: ~/.agents/skills + # (the spec's user-global location, read by `npx skills` and compliant + # frameworks); project: ./.agents/skills. The CLI accepts `skills` as an + # alias (see _canonical_platform). Ships its own rendered bundle. + "skill_file": "skill-agents.md", + "skill_dst": Path(".agents") / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + "skill_refs": "agents", + }, + "devin": { + # Monolith: devin ships the full SKILL.md inline, no references/ sidecar. + "skill_file": "skill-devin.md", + # User scope: ~/.config/devin/skills/graphify/SKILL.md + # Project scope: .devin/skills/graphify/SKILL.md (overridden in _platform_skill_destination) + "skill_dst": Path(".config") / "devin" / "skills" / "graphify" / "SKILL.md", + "claude_md": False, + }, +} +# CLI-only platform aliases, resolved to a real _PLATFORM_CONFIG key before +# dispatch. `skills` is the friendly alias for the generic `agents` platform +# (the Agent-Skills ecosystem calls them "skills"). +_PLATFORM_ALIASES: dict[str, str] = {"skills": "agents"} +def _canonical_platform(platform_name: str) -> str: + """Resolve a CLI platform alias to its real _PLATFORM_CONFIG key.""" + return _PLATFORM_ALIASES.get(platform_name, platform_name) +def _replace_or_append_section(content: str, marker: str, new_section: str) -> str: + """Idempotently update or append a graphify-owned section in shared files. + If no line is exactly ``marker`` (the heading, at column 0), append + ``new_section`` to the end (with a blank-line separator if there's existing + content). + If a real ``marker`` heading exists, replace the existing section in place. + The section runs from that heading to the line before the next H2 heading + (``## `` at line start), or to EOF if no later H2 exists. This lets older + installs receive the updated copy without users having to uninstall and + reinstall (issue #580). + The heading is matched only when a line *is* exactly ``marker`` (after + stripping surrounding whitespace), never as a substring. Matching ``## + graphify`` inside a bullet or an inline reference used to anchor the replace + on that mention and delete every line from there to the next heading, + silently destroying hand-curated content (#1688). When several exact + headings exist, the last one is used, since graphify's section is appended. + """ + lines = content.split("\n") + starts = [i for i, line in enumerate(lines) if line.strip() == marker] + if not starts: + if content.strip(): + return content.rstrip() + "\n\n" + new_section.lstrip() + return new_section.lstrip() + + start = starts[-1] + end = len(lines) + for j in range(start + 1, len(lines)): + if lines[j].startswith("## "): + end = j + break + + head = "\n".join(lines[:start]).rstrip() + tail = "\n".join(lines[end:]).lstrip() + section = new_section.strip() + + parts: list[str] = [] + if head: + parts.append(head) + parts.append(section) + if tail: + parts.append(tail) + out = "\n\n".join(parts) + if not out.endswith("\n"): + out += "\n" + return out + + +def _print_banner() -> None: + """Amber brain banner on graphify install. TTY-only, never raises.""" + if not sys.stdout.isatty(): + return + try: + if sys.platform == "win32": + import ctypes + ctypes.windll.kernel32.SetConsoleMode( + ctypes.windll.kernel32.GetStdHandle(-11), 7 + ) + A = "\033[38;5;214m" + D = "\033[38;5;130m" + R = "\033[0m" + print(f"""{A} + ╭──◉──╮ ╭──◉──╮ + ╱ ◉ ◉ ╲ ╱ ◉ ◉ ╲ +│ ◉─◉─◉ ◉ ◉─◉─◉ │ +│ ◉ ◉ │ ◉ ◉ │ +│ ◉─◉─◉ ◉ ◉─◉─◉ │ + ╲ ◉ ◉ ╱ ╲ ◉ ◉ ╱ + ╰──◉──╯ ╰──◉──╯ + ◉ + + █▀▀ █▀█ ▄▀█ █▀█ █ █ █ █▀▀ █▄█ + █▄█ █▀▄ █▀█ █▀▀ █▀█ █ █▀ █{D} {__version__}{R} +""") + except Exception: + pass + + +def install(platform: str = "claude", *, project: bool = False, project_dir: Path | None = None) -> None: + _print_banner() + platform = _canonical_platform(platform) + if platform == "gemini": + gemini_install(project_dir=project_dir, project=project) + return + if platform == "cursor": + _cursor_install(Path(".")) + return + # On Windows, antigravity needs the PowerShell skill, not the bash one + if platform == "antigravity" and sys.platform == "win32": + platform = "antigravity-windows" + if platform not in _PLATFORM_CONFIG: + print( + f"error: unknown platform '{platform}'. Choose from: {', '.join(_PLATFORM_CONFIG)}, gemini, cursor", + file=sys.stderr, + ) + sys.exit(1) + cfg = _PLATFORM_CONFIG[platform] + project_dir = project_dir or Path(".") + skill_dst = _copy_skill_file(platform, project=project, project_dir=project_dir) + if platform == "kilo": + # Kilo Code also supports a native /graphify command file. + command_src = Path(__file__).parent / "command-kilo.md" + if not command_src.exists(): + print( + f"error: command-kilo.md not found in package - reinstall graphify", + file=sys.stderr, + ) + sys.exit(1) + command_dst = Path.home() / ".config" / "kilo" / "command" / "graphify.md" + command_dst.parent.mkdir(parents=True, exist_ok=True) + shutil.copy(command_src, command_dst) + print(f" command installed -> {command_dst}") + + if cfg["claude_md"]: + # Register in the matching Claude Code scope. + claude_md = (project_dir / ".claude" / "CLAUDE.md") if project else Path.home() / ".claude" / "CLAUDE.md" + registration = _skill_registration(".claude/skills/graphify/SKILL.md" if project else "~/.claude/skills/graphify/SKILL.md") + if claude_md.exists(): + content = claude_md.read_text(encoding="utf-8") + if "graphify" in content: + print(f" CLAUDE.md -> already registered (no change)") + else: + claude_md.write_text(content.rstrip() + registration, encoding="utf-8") + print(f" CLAUDE.md -> skill registered in {claude_md}") + else: + claude_md.parent.mkdir(parents=True, exist_ok=True) + claude_md.write_text(registration.lstrip(), encoding="utf-8") + print(f" CLAUDE.md -> created at {claude_md}") + + if platform == "codebuddy": + # Register in ~/.codebuddy/CODEBUDDY.md (CodeBuddy only) + codebuddy_md = Path.home() / ".codebuddy" / "CODEBUDDY.md" + registration = _skill_registration("~/.codebuddy/skills/graphify/SKILL.md") + if codebuddy_md.exists(): + content = codebuddy_md.read_text(encoding="utf-8") + if "graphify" in content: + print(f" CODEBUDDY.md -> already registered (no change)") + else: + codebuddy_md.write_text(content.rstrip() + registration, encoding="utf-8") + print(f" CODEBUDDY.md -> skill registered in {codebuddy_md}") + else: + codebuddy_md.parent.mkdir(parents=True, exist_ok=True) + codebuddy_md.write_text(registration.lstrip(), encoding="utf-8") + print(f" CODEBUDDY.md -> created at {codebuddy_md}") + + if platform == "opencode": + _install_opencode_plugin(project_dir if project else Path(".")) + # Refresh version stamps in all other previously-installed skill dirs so + # stale-version warnings don't fire for platforms not explicitly re-installed. + if project: + _print_project_git_add_hint([_project_scope_root(skill_dst, project_dir)]) + else: + _refresh_all_version_stamps() + print() + print("Done. Open your AI coding assistant and type:") + print() + print(" /graphify .") + print() +def _print_install_usage() -> None: + platforms = ", ".join([*_PLATFORM_CONFIG, "gemini", "cursor"]) + print("Usage: graphify install [--project] [--platform P|P]") + print(f"Platforms: {platforms}") + # The always-on instruction blocks are packaged markdown under graphify/always_on/, # generated by tools/skillgen and guarded by `skillgen --check`. Reading them at # load keeps the install-string / issue-#580 contract byte-for-byte while letting # a human edit one fragment instead of a triple-quoted literal here. +_CLAUDE_MD_MARKER = "## graphify" +_CODEBUDDY_MD_MARKER = "## graphify" # AGENTS.md section for Codex, OpenCode, and OpenClaw. # All three platforms read AGENTS.md in the project root for persistent instructions. +_AGENTS_MD_MARKER = "## graphify" +_GEMINI_MD_MARKER = "## graphify" # Gemini CLI BeforeTool hook nudge text. The hook always returns # {"decision":"allow"} (never blocks a tool) and appends this as additionalContext @@ -296,74 +860,737 @@ def _version_tuple(version: str) -> tuple[int, ...]: # `python`/`py` or absent on Windows) and embedded backticks + escaped quotes that # Windows PowerShell mangles (#522 follow-up); the subcommand form has no such # dependency and parses under every shell. +_GEMINI_NUDGE_TEXT = ( + 'graphify: knowledge graph at graphify-out/. For focused questions, run ' + '`graphify query ""` (scoped subgraph, usually much smaller than ' + 'GRAPH_REPORT.md) instead of grepping raw files. Read GRAPH_REPORT.md only ' + 'for broad architecture context.' +) +def _gemini_hook() -> dict: + """Gemini CLI BeforeTool hook, resolved to a shell-agnostic `graphify` call.""" + exe = _resolve_graphify_exe() + if " " in exe and not exe.startswith('"'): + exe = f'"{exe}"' + return { + "matcher": "read_file|list_directory", + "hooks": [{"type": "command", "command": f"{exe} hook-guard gemini"}], + } +def gemini_install(project_dir: Path | None = None, *, project: bool = False) -> None: + """Copy skill file, write GEMINI.md section, and install BeforeTool hook.""" + project_dir = project_dir or Path(".") + skill_dst = _copy_skill_file("gemini", project=project, project_dir=project_dir) + target = project_dir / "GEMINI.md" + if target.exists(): + content = target.read_text(encoding="utf-8") + new_content = _replace_or_append_section( + content, _GEMINI_MD_MARKER, _always_on("gemini-md") + ) + else: + new_content = _always_on("gemini-md") + + if target.exists() and new_content == target.read_text(encoding="utf-8"): + print(f"graphify already configured in {target.resolve()} (no change)") + else: + target.write_text(new_content, encoding="utf-8") + print(f"graphify section written to {target.resolve()}") + + # Always re-install the Gemini hook so an older payload (e.g. pre-issue-#580 + # wording) is replaced on upgrade. + _install_gemini_hook(project_dir) + if project: + _print_project_git_add_hint([_project_scope_root(skill_dst, project_dir), project_dir / "GEMINI.md", project_dir / ".gemini"]) + print() + print("Gemini CLI will now check the knowledge graph before answering") + print("codebase questions and rebuild it after code changes.") + + +def _install_gemini_hook(project_dir: Path) -> None: + settings_path = project_dir / ".gemini" / "settings.json" + settings_path.parent.mkdir(parents=True, exist_ok=True) + try: + settings = ( + json.loads(settings_path.read_text(encoding="utf-8")) + if settings_path.exists() + else {} + ) + except json.JSONDecodeError: + settings = {} + before_tool = settings.setdefault("hooks", {}).setdefault("BeforeTool", []) + settings["hooks"]["BeforeTool"] = [ + h for h in before_tool if "graphify" not in str(h) + ] + settings["hooks"]["BeforeTool"].append(_gemini_hook()) + settings_path.write_text(json.dumps(settings, indent=2), encoding="utf-8") + print(" .gemini/settings.json -> BeforeTool hook registered") + + +def _uninstall_gemini_hook(project_dir: Path) -> None: + settings_path = project_dir / ".gemini" / "settings.json" + if not settings_path.exists(): + return + try: + settings = json.loads(settings_path.read_text(encoding="utf-8")) + except json.JSONDecodeError: + return + before_tool = settings.get("hooks", {}).get("BeforeTool", []) + filtered = [h for h in before_tool if "graphify" not in str(h)] + if len(filtered) == len(before_tool): + return + settings["hooks"]["BeforeTool"] = filtered + settings_path.write_text(json.dumps(settings, indent=2), encoding="utf-8") + print(" .gemini/settings.json -> BeforeTool hook removed") +def gemini_uninstall(project_dir: Path | None = None, *, project: bool = False) -> None: + """Remove the graphify section from GEMINI.md, uninstall hook, and remove skill file.""" + project_dir = project_dir or Path(".") + _remove_skill_file("gemini", project=project, project_dir=project_dir) + target = project_dir / "GEMINI.md" + if not target.exists(): + print("No GEMINI.md found in current directory - nothing to do") + return + content = target.read_text(encoding="utf-8") + if _GEMINI_MD_MARKER not in content: + print("graphify section not found in GEMINI.md - nothing to do") + return + cleaned = re.sub( + r"\n*## graphify\n.*?(?=\n## |\Z)", "", content, flags=re.DOTALL + ).rstrip() + if cleaned: + target.write_text(cleaned + "\n", encoding="utf-8") + print(f"graphify section removed from {target.resolve()}") + else: + target.unlink() + print(f"GEMINI.md was empty after removal - deleted {target.resolve()}") + _uninstall_gemini_hook(project_dir) + + +_VSCODE_INSTRUCTIONS_MARKER = "## graphify" + + +def vscode_install(project_dir: Path | None = None) -> None: + """Install graphify skill for VS Code Copilot Chat + write .github/copilot-instructions.md.""" + skill_src = Path(__file__).parent / "skill-vscode.md" + refs_bundle = "vscode" + if not skill_src.exists(): + skill_src = Path(__file__).parent / "skill-copilot.md" + refs_bundle = "copilot" + skill_dst = Path.home() / ".copilot" / "skills" / "graphify" / "SKILL.md" + skill_dst.parent.mkdir(parents=True, exist_ok=True) + tmp_dst = skill_dst.with_suffix(skill_dst.suffix + ".tmp") + try: + shutil.copy(skill_src, tmp_dst) + os.replace(tmp_dst, skill_dst) + except Exception: + try: + tmp_dst.unlink(missing_ok=True) + except OSError: + pass + raise + # Progressive-capable: install the packaged references/ sidecar when present. + refs_src = Path(__file__).parent / "skills" / refs_bundle / "references" + if refs_src.exists(): + _install_skill_references(skill_dst, refs_src) + print(f" references -> {skill_dst.parent / 'references'}") + else: + orphan_refs = skill_dst.parent / "references" + if orphan_refs.exists(): + shutil.rmtree(orphan_refs) + (skill_dst.parent / ".graphify_version").write_text(__version__, encoding="utf-8") + print(f" skill installed -> {skill_dst}") + + instructions = (project_dir or Path(".")) / ".github" / "copilot-instructions.md" + instructions.parent.mkdir(parents=True, exist_ok=True) + if instructions.exists(): + content = instructions.read_text(encoding="utf-8") + new_content = _replace_or_append_section( + content, _VSCODE_INSTRUCTIONS_MARKER, _always_on("vscode-instructions") + ) + if new_content == content: + print(f" {instructions} -> already configured (no change)") + else: + instructions.write_text(new_content, encoding="utf-8") + print(f" {instructions} -> graphify section {'updated' if _VSCODE_INSTRUCTIONS_MARKER in content else 'added'}") + else: + instructions.write_text(_always_on("vscode-instructions"), encoding="utf-8") + print(f" {instructions} -> created") + + print() + print( + "VS Code Copilot Chat configured. Type /graphify in the chat panel to build the graph." + ) + print("Note: for GitHub Copilot CLI (terminal), use: graphify copilot install") + + +def vscode_uninstall(project_dir: Path | None = None) -> None: + """Remove graphify VS Code Copilot Chat skill and .github/copilot-instructions.md section.""" + skill_dst = Path.home() / ".copilot" / "skills" / "graphify" / "SKILL.md" + if skill_dst.exists(): + skill_dst.unlink() + print(f" skill removed -> {skill_dst}") + version_file = skill_dst.parent / ".graphify_version" + if version_file.exists(): + version_file.unlink() + refs_dir = skill_dst.parent / "references" + if refs_dir.exists(): + shutil.rmtree(refs_dir) + for d in ( + skill_dst.parent, + skill_dst.parent.parent, + skill_dst.parent.parent.parent, + ): + try: + d.rmdir() + except OSError: + break + + instructions = (project_dir or Path(".")) / ".github" / "copilot-instructions.md" + if not instructions.exists(): + return + content = instructions.read_text(encoding="utf-8") + if _VSCODE_INSTRUCTIONS_MARKER not in content: + return + cleaned = re.sub( + r"\n*## graphify\n.*?(?=\n## |\Z)", "", content, flags=re.DOTALL + ).rstrip() + if cleaned: + instructions.write_text(cleaned + "\n", encoding="utf-8") + print(f" graphify section removed from {instructions}") + else: + instructions.unlink() + print(f" {instructions} -> deleted (was empty after removal)") +_ANTIGRAVITY_RULES_PATH = Path(".agents") / "rules" / "graphify.md" +_ANTIGRAVITY_WORKFLOW_PATH = Path(".agents") / "workflows" / "graphify.md" +_ANTIGRAVITY_WORKFLOW = """\ +--- +name: graphify +description: Turn any folder of files into a navigable knowledge graph +--- +# Workflow: graphify +Follow the graphify skill installed at ~/.gemini/config/skills/graphify/SKILL.md to run the full pipeline. - - - - +If no path argument is given, use `.` (current directory). +""" _KIRO_STEERING_MARKER = "graphify: A knowledge graph of this project" +def _kiro_install(project_dir: Path) -> None: + """Write graphify skill + steering file for Kiro IDE/CLI.""" + project_dir = project_dir or Path(".") + + # Skill file + references/ sidecar + .graphify_version stamp via the shared + # progressive-disclosure helper. Previously this used a bare write_text that + # bypassed _copy_skill_file, so the references/ dir and version stamp were + # never written even though kiro declares skill_refs: "kiro" (#1142). + _copy_skill_file("kiro", project=True, project_dir=project_dir) + + # Steering file → .kiro/steering/graphify.md (always-on) + steering_dir = project_dir / ".kiro" / "steering" + steering_dir.mkdir(parents=True, exist_ok=True) + steering_dst = steering_dir / "graphify.md" + if steering_dst.exists() and steering_dst.read_text(encoding="utf-8") == _always_on("kiro-steering"): + print(f" .kiro/steering/graphify.md -> already configured (no change)") + else: + # File is wholly graphify-owned. Overwrite on upgrade so older + # report-first wording does not silently linger (issue #580). + action = "updated" if steering_dst.exists() else "written" + steering_dst.write_text(_always_on("kiro-steering"), encoding="utf-8") + print(f" .kiro/steering/graphify.md -> always-on steering {action}") + + print() + print("Kiro will now read the knowledge graph before every conversation.") + print("Use /graphify to build or update the graph.") + + +def _kiro_uninstall(project_dir: Path) -> None: + """Remove graphify skill + steering file for Kiro.""" + project_dir = project_dir or Path(".") + removed = [] + + # Skill + .graphify_version + references/ sidecar + empty-dir walk. + skill_dst = _platform_skill_destination("kiro", project=True, project_dir=project_dir) + if _remove_skill_file("kiro", project=True, project_dir=project_dir): + removed.append(str(skill_dst.relative_to(project_dir))) + + steering_dst = project_dir / ".kiro" / "steering" / "graphify.md" + if steering_dst.exists(): + steering_dst.unlink() + removed.append(str(steering_dst.relative_to(project_dir))) + + print("Removed: " + (", ".join(removed) if removed else "nothing to remove")) + + +def _antigravity_finalize(skill_dst: Path, project_dir: Path) -> None: + """Write Antigravity's always-on layer next to an installed skill. + + Injects the native tool-discovery YAML frontmatter into *skill_dst*, then + writes ``.agents/rules/graphify.md`` and ``.agents/workflows/graphify.md`` + under *project_dir*. Shared by the global ``antigravity install`` and the + project-scoped ``install --project --platform antigravity`` paths, so both lay + down the rules/workflows that the uninstall path already expects to remove. + """ + # Inject YAML frontmatter for native Antigravity tool discovery. + if skill_dst.exists(): + content = skill_dst.read_text(encoding="utf-8") + if not content.startswith("---\n"): + frontmatter = "---\nname: graphify-manager\ndescription: Rebuild the code graph or perform manual CLI queries when MCP server is offline.\n---\n\n" + skill_dst.write_text(frontmatter + content, encoding="utf-8") + + # .agents/rules/graphify.md + rules_path = project_dir / _ANTIGRAVITY_RULES_PATH + rules_path.parent.mkdir(parents=True, exist_ok=True) + if rules_path.exists(): + existing = rules_path.read_text(encoding="utf-8") + if _always_on("antigravity-rules").strip() != existing.strip(): + rules_path.write_text(_always_on("antigravity-rules"), encoding="utf-8") + print(f"graphify rule updated at {rules_path.resolve()}") + else: + print(f"graphify rule already configured at {rules_path.resolve()} (no change)") + else: + rules_path.write_text(_always_on("antigravity-rules"), encoding="utf-8") + print(f"graphify rule written to {rules_path.resolve()}") + + # .agents/workflows/graphify.md + wf_path = project_dir / _ANTIGRAVITY_WORKFLOW_PATH + wf_path.parent.mkdir(parents=True, exist_ok=True) + if wf_path.exists(): + existing = wf_path.read_text(encoding="utf-8") + if _ANTIGRAVITY_WORKFLOW.strip() != existing.strip(): + wf_path.write_text(_ANTIGRAVITY_WORKFLOW, encoding="utf-8") + print(f"graphify workflow updated at {wf_path.resolve()}") + else: + print(f"graphify workflow already configured at {wf_path.resolve()} (no change)") + else: + wf_path.write_text(_ANTIGRAVITY_WORKFLOW, encoding="utf-8") + print(f"graphify workflow written to {wf_path.resolve()}") + + +def _antigravity_install(project_dir: Path) -> None: + """Install graphify for Google Antigravity (global skill + .agents/rules + .agents/workflows).""" + # Copy the skill to ~/.gemini/config/skills/graphify/SKILL.md (global), then + # lay down the always-on rules/workflows under the project dir. + install(platform="antigravity") + _antigravity_finalize(_platform_skill_destination("antigravity"), project_dir) + + print() + print("Antigravity will now check the knowledge graph before answering") + print("codebase questions. Run /graphify first to build the graph.") + print() + print( + "To enable full MCP architecture navigation, add this to ~/.gemini/antigravity/mcp_config.json:" + ) + print(' "graphify": {') + print(' "command": "uv",') + print( + ' "args": ["run", "--with", "graphifyy", "--with", "mcp", "-m", "graphify.serve", "${workspace.path}/graphify-out/graph.json"]' + ) + print(" }") + + +def _antigravity_uninstall(project_dir: Path, *, project: bool = False) -> None: + """Remove graphify Antigravity rules, workflow, and skill files.""" + # Remove rules file + rules_path = project_dir / _ANTIGRAVITY_RULES_PATH + if rules_path.exists(): + rules_path.unlink() + print(f"graphify rule removed from {rules_path.resolve()}") + else: + print("No graphify Antigravity rule found - nothing to do") + + # Remove workflow file + wf_path = project_dir / _ANTIGRAVITY_WORKFLOW_PATH + if wf_path.exists(): + wf_path.unlink() + print(f"graphify workflow removed from {wf_path.resolve()}") + + # Remove skill file + skill_dst = _platform_skill_destination("antigravity", project=project, project_dir=project_dir) + if skill_dst.exists(): + skill_dst.unlink() + print(f"graphify skill removed from {skill_dst}") + version_file = skill_dst.parent / ".graphify_version" + if version_file.exists(): + version_file.unlink() + refs_dir = skill_dst.parent / "references" + if refs_dir.exists(): + shutil.rmtree(refs_dir) + for d in ( + skill_dst.parent, + skill_dst.parent.parent, + skill_dst.parent.parent.parent, + ): + try: + d.rmdir() + except OSError: + break + + +_CURSOR_RULE_PATH = Path(".cursor") / "rules" / "graphify.mdc" +_CURSOR_RULE = """\ +--- +description: graphify knowledge graph context +alwaysApply: true +--- + +This project has a graphify knowledge graph at graphify-out/. + +**MANDATORY: Before using Read, Grep, Glob, or Bash to explore the codebase, you MUST run graphify first:** +- `graphify query ""` — scoped subgraph for any codebase or architecture question +- `graphify path "" ""` — dependency path between two symbols +- `graphify explain ""` — all nodes related to a concept + +This applies to YOU and to every subagent you spawn. Include this rule explicitly in every subagent prompt that involves code exploration. Do not skip graphify because files are "already known" or because you are executing a plan — the graph surfaces cross-file dependencies and INFERRED edges that grep and Read cannot find. + +Only use Read/Grep/Glob directly when: +1. graphify has already oriented you and you need to modify or debug specific lines +2. `graphify-out/graph.json` does not exist yet + +- If `graphify-out/wiki/index.md` exists, navigate it instead of reading raw files +- Read `graphify-out/GRAPH_REPORT.md` only for broad architecture review when query/path/explain do not surface enough context +- After modifying code files, run `graphify update .` to keep the graph current (AST-only, no API cost) +""" + + +def _cursor_install(project_dir: Path) -> None: + """Write .cursor/rules/graphify.mdc with alwaysApply: true.""" + rule_path = (project_dir or Path(".")) / _CURSOR_RULE_PATH + rule_path.parent.mkdir(parents=True, exist_ok=True) + if rule_path.exists() and rule_path.read_text(encoding="utf-8") == _CURSOR_RULE: + print(f"graphify rule at {rule_path} already configured (no change)") + return + # File is wholly graphify-owned. Overwrite on upgrade so older + # report-first wording does not silently linger (issue #580). + action = "updated" if rule_path.exists() else "written" + rule_path.write_text(_CURSOR_RULE, encoding="utf-8") + print(f"graphify rule {action} at {rule_path.resolve()}") + print() + print("Cursor will now always include the knowledge graph context.") + print("Run /graphify . first to build the graph if you haven't already.") + + +def _cursor_uninstall(project_dir: Path) -> None: + """Remove .cursor/rules/graphify.mdc.""" + rule_path = (project_dir or Path(".")) / _CURSOR_RULE_PATH + if not rule_path.exists(): + print("No graphify Cursor rule found - nothing to do") + return + rule_path.unlink() + print(f"graphify Cursor rule removed from {rule_path.resolve()}") +# Devin CLI — .windsurf/rules/graphify.md (always-on context) +# Devin reads .windsurf/rules/*.md files the same way Windsurf IDE does. +_DEVIN_RULES_PATH = Path(".windsurf") / "rules" / "graphify.md" +_DEVIN_RULES = """\ +## graphify +This project has a graphify knowledge graph at graphify-out/. +Rules: +- For codebase or architecture questions, when `graphify-out/graph.json` exists, first run `graphify query ""` (or `graphify path "" ""` / `graphify explain ""`). These return a scoped subgraph, usually much smaller than `GRAPH_REPORT.md` or raw grep output. +- If graphify-out/wiki/index.md exists, navigate it instead of reading raw files +- Read graphify-out/GRAPH_REPORT.md only for broad architecture review or when query/path/explain do not surface enough context +- After modifying code files in this session, run `graphify update .` to keep the graph current (AST-only, no API cost) +""" +def _devin_rules_install(project_dir: Path) -> None: + """Write .windsurf/rules/graphify.md for always-on Devin context.""" + rules_path = (project_dir or Path(".")) / _DEVIN_RULES_PATH + rules_path.parent.mkdir(parents=True, exist_ok=True) + if rules_path.exists() and rules_path.read_text(encoding="utf-8") == _DEVIN_RULES: + print(f" {rules_path} -> already configured (no change)") + return + action = "updated" if rules_path.exists() else "written" + rules_path.write_text(_DEVIN_RULES, encoding="utf-8") + print(f" rules {action} -> {rules_path}") +def _devin_rules_uninstall(project_dir: Path) -> None: + """Remove .windsurf/rules/graphify.md.""" + rules_path = (project_dir or Path(".")) / _DEVIN_RULES_PATH + if not rules_path.exists(): + return + rules_path.unlink() + print(f" rules removed -> {rules_path}") + + +_KILO_PLUGIN_JS = """\ +// graphify Kilo plugin +// Injects a knowledge graph reminder before bash tool calls when the graph exists. +import { existsSync } from "fs"; +import { join } from "path"; + +export const GraphifyPlugin = async ({ directory }) => { + let reminded = false; + + return { + "tool.execute.before": async (input, output) => { + if (reminded) return; + if (!existsSync(join(directory, "graphify-out", "graph.json"))) return; + + if (input.tool === "bash") { + // Separate with ';' not '&&' — Windows PowerShell 5.1 rejects '&&' as a + // statement separator ("not a valid statement separator"), which broke + // the first bash command in every OpenCode session on Windows (#1646). + // ';' works in PowerShell 5.1, Bash, and POSIX shells alike. + output.args.command = + 'echo "[graphify] Knowledge graph available. Read graphify-out/GRAPH_REPORT.md for god nodes and architecture context before searching files." ; ' + + output.args.command; + reminded = true; + } + }, + }; +}; +""" + +_KILO_PLUGIN_PATH = Path(".kilo") / "plugins" / "graphify.js" +_KILO_CONFIG_JSON_PATH = Path(".kilo") / "kilo.json" +_KILO_CONFIG_JSONC_PATH = Path(".kilo") / "kilo.jsonc" + + +def _strip_json_comments(raw: str) -> str: + """Remove JSONC-style comments while leaving string content intact.""" + result: list[str] = [] + in_string = False + escaped = False + line_comment = False + block_comment = False + i = 0 + + while i < len(raw): + ch = raw[i] + nxt = raw[i + 1] if i + 1 < len(raw) else "" + + if line_comment: + if ch == "\n": + line_comment = False + result.append(ch) + i += 1 + continue + + if block_comment: + if ch == "*" and nxt == "/": + block_comment = False + i += 2 + else: + i += 1 + continue + + if in_string: + result.append(ch) + if escaped: + escaped = False + elif ch == "\\": + escaped = True + elif ch == '"': + in_string = False + i += 1 + continue + + if ch == "/" and nxt == "/": + line_comment = True + i += 2 + continue + if ch == "/" and nxt == "*": + block_comment = True + i += 2 + continue + + result.append(ch) + if ch == '"': + in_string = True + i += 1 + + return re.sub(r",(\s*[}\]])", r"\1", "".join(result)) + + +def _load_json_like(config_file: Path) -> dict: + if not config_file.exists(): + return {} + try: + raw = config_file.read_text(encoding="utf-8") + if config_file.suffix == ".jsonc": + raw = _strip_json_comments(raw) + loaded = json.loads(raw) + except (OSError, json.JSONDecodeError): + return {} + return loaded if isinstance(loaded, dict) else {} + + +def _kilo_config_path(project_dir: Path) -> Path: + kilo_dir = (project_dir or Path(".")) / ".kilo" + json_path = kilo_dir / _KILO_CONFIG_JSON_PATH.name + if json_path.exists(): + return json_path + jsonc_path = kilo_dir / _KILO_CONFIG_JSONC_PATH.name + if jsonc_path.exists(): + return jsonc_path + return json_path + + +def _kilo_config_write_path(project_dir: Path) -> Path: + """Write automated Kilo edits to kilo.json so existing JSONC stays untouched.""" + kilo_dir = (project_dir or Path(".")) / ".kilo" + return kilo_dir / _KILO_CONFIG_JSON_PATH.name + + +def _install_kilo_plugin(project_dir: Path) -> None: + """Write graphify.js plugin and register it without rewriting user JSONC.""" + plugin_file = project_dir / _KILO_PLUGIN_PATH + plugin_file.parent.mkdir(parents=True, exist_ok=True) + plugin_file.write_text(_KILO_PLUGIN_JS, encoding="utf-8") + print(f" {_KILO_PLUGIN_PATH} -> tool.execute.before hook written") + + config_file = _kilo_config_path(project_dir) + write_config_file = _kilo_config_write_path(project_dir) + write_config_file.parent.mkdir(parents=True, exist_ok=True) + config = _load_json_like(config_file) + plugins = config.get("plugin") + if not isinstance(plugins, list): + plugins = [] + config["plugin"] = plugins + entry = plugin_file.resolve().as_uri() + if entry not in plugins: + plugins.append(entry) + write_config_file.write_text(json.dumps(config, indent=2), encoding="utf-8") + print(f" {write_config_file.relative_to(project_dir)} -> plugin registered") + else: + print( + f" {config_file.relative_to(project_dir)} -> plugin already registered (no change)" + ) +def _uninstall_kilo_plugin(project_dir: Path) -> None: + """Remove graphify.js plugin and deregister it without rewriting user JSONC.""" + plugin_file = project_dir / _KILO_PLUGIN_PATH + if plugin_file.exists(): + plugin_file.unlink() + print(f" {_KILO_PLUGIN_PATH} -> removed") + config_file = _kilo_config_path(project_dir) + if not config_file.exists(): + return + write_config_file = _kilo_config_write_path(project_dir) + config = _load_json_like(config_file) + plugins = config.get("plugin", []) + if not isinstance(plugins, list): + plugins = [] + entry = plugin_file.resolve().as_uri() + if entry in plugins: + config["plugin"] = [plugin for plugin in plugins if plugin != entry] + if not config["plugin"]: + config.pop("plugin") + write_config_file.parent.mkdir(parents=True, exist_ok=True) + write_config_file.write_text(json.dumps(config, indent=2), encoding="utf-8") + print( + f" {write_config_file.relative_to(project_dir)} -> plugin deregistered" + ) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +# OpenCode tool.execute.before plugin — fires before every tool call. +# Injects a graph reminder into bash command output when graph.json exists. +_OPENCODE_PLUGIN_JS = """\ +// graphify OpenCode plugin +// Injects a knowledge graph reminder before bash tool calls when the graph exists. +// +// IMPORTANT: keep the reminder string free of backticks and $(...) constructs. +// The hook prepends `echo "" && ` to the user's bash command; +// backticks inside the double-quoted echo trigger bash command substitution, +// which both corrupts tool output and silently executes the very graphify +// command we are only suggesting. Plain words render fine in opencode's TUI. +import { existsSync } from "fs"; +import { join } from "path"; + +export const GraphifyPlugin = async ({ directory }) => { + let reminded = false; + + return { + "tool.execute.before": async (input, output) => { + if (reminded) return; + if (!existsSync(join(directory, "graphify-out", "graph.json"))) return; + + if (input.tool === "bash") { + // ';' not '&&' — Windows PowerShell 5.1 rejects '&&' as a statement + // separator, breaking the first bash command of the session (#1646). + output.args.command = + 'echo "[graphify] knowledge graph at graphify-out/. For focused questions, run graphify query with your question (scoped subgraph, usually much smaller than GRAPH_REPORT.md) instead of grepping raw files. Read GRAPH_REPORT.md only for broad architecture context." ; ' + + output.args.command; + reminded = true; + } + }, + }; +}; +""" + +_OPENCODE_PLUGIN_PATH = Path(".opencode") / "plugins" / "graphify.js" +_OPENCODE_CONFIG_PATH = Path(".opencode") / "opencode.json" + + +def _install_opencode_plugin(project_dir: Path) -> None: + """Write graphify.js plugin and register it in opencode.json.""" + plugin_file = project_dir / _OPENCODE_PLUGIN_PATH + plugin_file.parent.mkdir(parents=True, exist_ok=True) + plugin_file.write_text(_OPENCODE_PLUGIN_JS, encoding="utf-8") + print(f" {_OPENCODE_PLUGIN_PATH} -> tool.execute.before hook written") + + config_file = project_dir / _OPENCODE_CONFIG_PATH + if config_file.exists(): + try: + config = json.loads(config_file.read_text(encoding="utf-8")) + except json.JSONDecodeError: + config = {} + else: + config = {} + + plugins = config.setdefault("plugin", []) + entry = _OPENCODE_PLUGIN_PATH.as_posix() + if entry not in plugins: + plugins.append(entry) + config_file.write_text(json.dumps(config, indent=2), encoding="utf-8") + print(f" {_OPENCODE_CONFIG_PATH} -> plugin registered") + else: + print(f" {_OPENCODE_CONFIG_PATH} -> plugin already registered (no change)") + + +def _uninstall_opencode_plugin(project_dir: Path) -> None: + """Remove graphify.js plugin and deregister from opencode.json.""" + plugin_file = project_dir / _OPENCODE_PLUGIN_PATH + if plugin_file.exists(): + plugin_file.unlink() + print(f" {_OPENCODE_PLUGIN_PATH} -> removed") + + config_file = project_dir / _OPENCODE_CONFIG_PATH + if not config_file.exists(): + return + try: + config = json.loads(config_file.read_text(encoding="utf-8")) + except json.JSONDecodeError: + return + plugins = config.get("plugin", []) + entry = _OPENCODE_PLUGIN_PATH.as_posix() + if entry in plugins: + plugins.remove(entry) + if not plugins: + config.pop("plugin") + config_file.write_text(json.dumps(config, indent=2), encoding="utf-8") + print(f" {_OPENCODE_CONFIG_PATH} -> plugin deregistered") _CODEX_HOOK = { @@ -387,61 +1614,700 @@ def _version_tuple(version: str) -> tuple[int, ...]: } +def _resolve_graphify_exe() -> str: + """Return the absolute path to the graphify executable. + Falls back to bare 'graphify' if resolution fails. Using an absolute path + ensures the hook works in environments where the venv Scripts/ directory is + not on PATH (e.g. VS Code Codex extension on Windows). + """ + import shutil + found = shutil.which("graphify") + if found: + return found + # Derive from sys.executable: same Scripts/ (Windows) or bin/ (Unix) dir + scripts_dir = Path(sys.executable).parent + for name in ("graphify.exe", "graphify"): + candidate = scripts_dir / name + if candidate.exists(): + return str(candidate) + return "graphify" + + +def _run_hook_guard(kind: str) -> None: + """Shell-agnostic PreToolUse guard (#522). + + Reads the tool-call JSON from stdin and, when a knowledge graph exists in the + current output dir, prints a nudge (`additionalContext`) telling the agent to + use graphify instead of grepping/reading raw files. Replaces the old inline + bash hooks that failed to parse on Windows. Always fails open: any error, or a + non-matching tool call, prints nothing and the caller exits 0, so a legitimate + tool call is never blocked. Detection mirrors the previous hooks exactly. + """ + from graphify.paths import out_path, GRAPHIFY_OUT_NAME + # Gemini's BeforeTool hook takes no stdin and must ALWAYS return a decision so + # the tool is never blocked; the graph nudge is appended only when a graph + # exists. Handled before the stdin read below (which the search/read guards need). + if kind == "gemini": + payload = {"decision": "allow"} + try: + if out_path("graph.json").is_file(): + payload["additionalContext"] = _GEMINI_NUDGE_TEXT + except Exception: + pass + sys.stdout.write(json.dumps(payload, ensure_ascii=False, separators=(",", ":"))) + return + try: + d = json.loads(sys.stdin.buffer.read().decode("utf-8", "replace")) + except Exception: + return + if not isinstance(d, dict): + return + t = d.get("tool_input", d) + if not isinstance(t, dict): + return + try: + if kind == "search": + cmd_str = str(t.get("command", "") or "") + # Same set the old `case` matched: *grep*, *ripgrep*, and rg/find/fd/ + # ack/ag as a token (name followed by a space). + if any(tok in cmd_str for tok in ("grep", "ripgrep", "rg ", "find ", "fd ", "ack ", "ag ")) \ + and out_path("graph.json").is_file(): + sys.stdout.write(_SEARCH_NUDGE) + elif kind == "read": + vals = [str(t.get("file_path") or ""), str(t.get("pattern") or ""), str(t.get("path") or "")] + j = " ".join(vals).lower().replace("\\", "/") + tails = [ + "." + seg.rsplit(".", 1)[-1] + for v in vals if v + for seg in [v.lower().replace("\\", "/").rsplit("/", 1)[-1]] + if "." in seg + ] + under_out = "graphify-out/" in j or (GRAPHIFY_OUT_NAME.lower() + "/") in j + if not under_out and any(tl in _HOOK_SOURCE_EXTS for tl in tails) \ + and out_path("graph.json").is_file(): + sys.stdout.write(_READ_NUDGE) + except Exception: + pass + + +def _install_codex_hook(project_dir: Path) -> None: + """Add graphify PreToolUse hook to .codex/hooks.json.""" + hooks_path = project_dir / ".codex" / "hooks.json" + hooks_path.parent.mkdir(parents=True, exist_ok=True) + + if hooks_path.exists(): + try: + existing = json.loads(hooks_path.read_text(encoding="utf-8")) + except json.JSONDecodeError: + existing = {} + else: + existing = {} + + graphify_exe = _resolve_graphify_exe() + hook_entry = { + "hooks": { + "PreToolUse": [ + { + "matcher": "Bash", + "hooks": [{"type": "command", "command": f"{graphify_exe} hook-check"}], + } + ] + } + } + pre_tool = existing.setdefault("hooks", {}).setdefault("PreToolUse", []) + existing["hooks"]["PreToolUse"] = [h for h in pre_tool if "graphify" not in str(h)] + existing["hooks"]["PreToolUse"].extend(hook_entry["hooks"]["PreToolUse"]) + hooks_path.write_text(json.dumps(existing, indent=2), encoding="utf-8") + print(f" .codex/hooks.json -> PreToolUse hook registered ({graphify_exe} hook-check)") +def _uninstall_codex_hook(project_dir: Path) -> None: + """Remove graphify PreToolUse hook from .codex/hooks.json.""" + hooks_path = project_dir / ".codex" / "hooks.json" + if not hooks_path.exists(): + return + try: + existing = json.loads(hooks_path.read_text(encoding="utf-8")) + except json.JSONDecodeError: + return + pre_tool = existing.get("hooks", {}).get("PreToolUse", []) + filtered = [h for h in pre_tool if "graphify" not in str(h)] + existing["hooks"]["PreToolUse"] = filtered + hooks_path.write_text(json.dumps(existing, indent=2), encoding="utf-8") + print(f" .codex/hooks.json -> PreToolUse hook removed") +def _agents_install(project_dir: Path, platform: str) -> None: + """Write the graphify section to the local AGENTS.md for always-on platforms.""" + target = (project_dir or Path(".")) / "AGENTS.md" + if target.exists(): + content = target.read_text(encoding="utf-8") + new_content = _replace_or_append_section( + content, _AGENTS_MD_MARKER, _always_on("agents-md") + ) + else: + new_content = _always_on("agents-md") + + if target.exists() and new_content == target.read_text(encoding="utf-8"): + print(f"graphify already configured in {target.resolve()} (no change)") + else: + target.write_text(new_content, encoding="utf-8") + print(f"graphify section written to {target.resolve()}") + + if platform == "codex": + _install_codex_hook(project_dir or Path(".")) + elif platform == "opencode": + _install_opencode_plugin(project_dir or Path(".")) + elif platform == "kilo": + _install_kilo_plugin(project_dir or Path(".")) + + print() + print( + f"{platform.capitalize()} will now check the knowledge graph before answering" + ) + print("codebase questions and rebuild it after code changes.") + if platform not in ("codex", "opencode", "kilo"): + print() + print("Note: unlike Claude Code, there is no PreToolUse hook equivalent for") + print( + f"{platform.capitalize()} — the AGENTS.md rules are the always-on mechanism." + ) +def _amp_legacy_cleanup() -> None: + """Best-effort removal of the pre-fix ~/.amp/skills/graphify install dir. + Older graphify versions wrote the Amp skill to ~/.amp/skills, which Amp does + not search. Clean it up on install so a stale, never-loaded copy does not + linger. Failures are ignored (the new path is what matters). + """ + legacy = Path.home() / ".amp" / "skills" / "graphify" + if legacy.exists(): + shutil.rmtree(legacy, ignore_errors=True) + if not legacy.exists(): + print(f" legacy removed -> {legacy}") +def _amp_install(project_dir: Path | None = None) -> None: + """User-scope Amp install: skill into ~/.config/agents/skills + AGENTS.md.""" + _amp_legacy_cleanup() + _copy_skill_file("amp") + _agents_install(project_dir or Path("."), "amp") +def _amp_uninstall(project_dir: Path | None = None) -> None: + """User-scope Amp uninstall: remove the skill and the AGENTS.md section.""" + removed = _remove_skill_file("amp") + if removed: + print("skill removed") + _agents_uninstall(project_dir or Path("."), platform="amp") +def _agents_platform_install(project_dir: Path | None = None) -> None: + """`graphify agents install`: skill into ~/.agents/skills + AGENTS.md. + The amp-twin of the generic Agent-Skills target. Mirrors _amp_install but + lands the skill at the spec's user-global ~/.agents/skills (set in + _platform_skill_destination). Wiring AGENTS.md keeps it honest with the + rendered hooks reference, which points at `graphify agents install`. The bare + `graphify install --platform agents` path stays skill-only (via install()), + exactly as amp's `--platform amp` does. + """ + _copy_skill_file("agents") + _agents_install(project_dir or Path("."), "agents") + + +def _agents_platform_uninstall(project_dir: Path | None = None) -> None: + """`graphify agents uninstall`: remove the skill and the AGENTS.md section.""" + removed = _remove_skill_file("agents") + if removed: + print("skill removed") + _agents_uninstall(project_dir or Path("."), platform="agents") + + +def _project_install(platform_name: str, project_dir: Path | None = None) -> None: + """Install platform skill/config files in the current project.""" + project_dir = project_dir or Path(".") + platform_name = _canonical_platform(platform_name) + if platform_name in ("claude", "windows"): + install(platform=platform_name, project=True, project_dir=project_dir) + claude_install(project_dir) + _print_project_git_add_hint([project_dir / ".claude", project_dir / "CLAUDE.md"]) + elif platform_name == "gemini": + gemini_install(project_dir, project=True) + elif platform_name == "cursor": + _cursor_install(project_dir) + _print_project_git_add_hint([project_dir / ".cursor"]) + elif platform_name == "kiro": + _kiro_install(project_dir) + _print_project_git_add_hint([project_dir / ".kiro"]) + elif platform_name in ("aider", "amp", "codex", "opencode", "claw", "droid", "trae", "trae-cn", "hermes"): + skill_dst = _copy_skill_file(platform_name, project=True, project_dir=project_dir) + _agents_install(project_dir, platform_name) + hint_paths = [_project_scope_root(skill_dst, project_dir), project_dir / "AGENTS.md"] + if platform_name == "opencode": + hint_paths.append(project_dir / ".opencode") + elif platform_name == "codex": + hint_paths.append(project_dir / ".codex") + _print_project_git_add_hint(hint_paths) + elif platform_name == "devin": + skill_dst = _copy_skill_file("devin", project=True, project_dir=project_dir) + _devin_rules_install(project_dir) + _print_project_git_add_hint([_project_scope_root(skill_dst, project_dir), project_dir / ".windsurf"]) + elif platform_name == "antigravity": + # Project-scoped: skill in .agents/skills/ PLUS the .agents/rules + + # .agents/workflows always-on layer (previously this path wrote only the + # skill, leaving the rules/workflows the uninstall path removes unset). + skill_dst = _copy_skill_file("antigravity", project=True, project_dir=project_dir) + _antigravity_finalize(skill_dst, project_dir) + _print_project_git_add_hint([_project_scope_root(skill_dst, project_dir), project_dir / ".agents"]) + elif platform_name in ("copilot", "pi", "kimi", "agents"): + # Skill-only project install: drop SKILL.md (+ references) at the scope + # root. `agents` -> ./.agents/skills/graphify/SKILL.md. + skill_dst = _copy_skill_file(platform_name, project=True, project_dir=project_dir) + _print_project_git_add_hint([_project_scope_root(skill_dst, project_dir)]) + else: + install(platform=platform_name, project=True, project_dir=project_dir) + + +def _project_uninstall(platform_name: str, project_dir: Path | None = None) -> None: + """Remove project-scoped platform skill/config files only.""" + project_dir = project_dir or Path(".") + platform_name = _canonical_platform(platform_name) + if platform_name in ("claude", "windows"): + _remove_skill_file(platform_name, project=True, project_dir=project_dir) + _remove_claude_skill_registration(project_dir) + claude_uninstall(project_dir, project=True) + elif platform_name == "gemini": + gemini_uninstall(project_dir, project=True) + elif platform_name == "cursor": + _cursor_uninstall(project_dir) + elif platform_name == "kiro": + _kiro_uninstall(project_dir) + elif platform_name in ("aider", "amp", "codex", "opencode", "claw", "droid", "trae", "trae-cn", "hermes"): + _remove_skill_file(platform_name, project=True, project_dir=project_dir) + _agents_uninstall(project_dir, platform=platform_name) + if platform_name == "codex": + _uninstall_codex_hook(project_dir) + elif platform_name == "antigravity": + _antigravity_uninstall(project_dir, project=True) + elif platform_name == "devin": + removed = _remove_skill_file("devin", project=True, project_dir=project_dir) + _devin_rules_uninstall(project_dir) + if not removed: + print("nothing to remove") + elif platform_name in ("copilot", "pi", "kimi", "agents"): + removed = _remove_skill_file(platform_name, project=True, project_dir=project_dir) + if not removed: + print("nothing to remove") + elif platform_name == "codebuddy": + codebuddy_uninstall(project_dir) + else: + _remove_skill_file(platform_name, project=True, project_dir=project_dir) + + +def _project_uninstall_all(project_dir: Path | None = None) -> None: + """Remove project-scoped install files without touching user-scope installs.""" + project_dir = project_dir or Path(".") + print("Uninstalling project-scoped graphify files...\n") + for platform_name in _PLATFORM_CONFIG: + _project_uninstall(platform_name, project_dir) + for platform_name in ("gemini", "cursor"): + _project_uninstall(platform_name, project_dir) + print("\nDone.") + + +def _agents_uninstall(project_dir: Path, platform: str = "") -> None: + """Remove the graphify section from the local AGENTS.md.""" + target = (project_dir or Path(".")) / "AGENTS.md" + + if not target.exists(): + print("No AGENTS.md found in current directory - nothing to do") + if platform == "opencode": + _uninstall_opencode_plugin(project_dir or Path(".")) + elif platform == "kilo": + _uninstall_kilo_plugin(project_dir or Path(".")) + return + content = target.read_text(encoding="utf-8") + if _AGENTS_MD_MARKER not in content: + print("graphify section not found in AGENTS.md - nothing to do") + if platform == "opencode": + _uninstall_opencode_plugin(project_dir or Path(".")) + elif platform == "kilo": + _uninstall_kilo_plugin(project_dir or Path(".")) + return + cleaned = re.sub( + r"\n*## graphify\n.*?(?=\n## |\Z)", + "", + content, + flags=re.DOTALL, + ).rstrip() + if cleaned: + target.write_text(cleaned + "\n", encoding="utf-8") + print(f"graphify section removed from {target.resolve()}") + else: + target.unlink() + print(f"AGENTS.md was empty after removal - deleted {target.resolve()}") + + if platform == "opencode": + _uninstall_opencode_plugin(project_dir or Path(".")) + elif platform == "kilo": + _uninstall_kilo_plugin(project_dir or Path(".")) + + +def _kilo_uninstall_global() -> list[str]: + removed = [] + command_dst = Path.home() / ".config" / "kilo" / "command" / "graphify.md" + if command_dst.exists(): + command_dst.unlink() + removed.append(f"command removed: {command_dst}") + try: + command_dst.parent.rmdir() + except OSError: + pass + skill_dst = Path.home() / _PLATFORM_CONFIG["kilo"]["skill_dst"] + if skill_dst.exists(): + skill_dst.unlink() + removed.append(f"skill removed: {skill_dst}") + version_file = skill_dst.parent / ".graphify_version" + if version_file.exists(): + version_file.unlink() + for d in ( + skill_dst.parent, + skill_dst.parent.parent, + skill_dst.parent.parent.parent, + ): + try: + d.rmdir() + except OSError: + break + + return removed + + +def _kilo_install(project_dir: Path) -> None: + """Install native Kilo skill + command globally and always-on project wiring locally.""" + install(platform="kilo") + _agents_install(project_dir or Path("."), "kilo") + + +def _kilo_uninstall(project_dir: Path) -> None: + """Remove Kilo always-on project wiring and global skill/command files.""" + _agents_uninstall(project_dir or Path("."), platform="kilo") + removed = _kilo_uninstall_global() + print("; ".join(removed) if removed else "nothing to remove") + + +def claude_install(project_dir: Path | None = None) -> None: + """Write the graphify section to the local CLAUDE.md.""" + target = (project_dir or Path(".")) / "CLAUDE.md" + + if target.exists(): + content = target.read_text(encoding="utf-8") + new_content = _replace_or_append_section( + content, _CLAUDE_MD_MARKER, _always_on("claude-md") + ) + else: + new_content = _always_on("claude-md") + + if target.exists() and new_content == target.read_text(encoding="utf-8"): + print(f"graphify already configured in {target.resolve()} (no change)") + else: + target.write_text(new_content, encoding="utf-8") + print(f"graphify section written to {target.resolve()}") + + # Always re-install the Claude Code PreToolUse hook so an old hook + # payload (e.g. pre-issue-#580 wording) is replaced on upgrade. + _install_claude_hook(project_dir or Path(".")) + + print() + print("Claude Code will now check the knowledge graph before answering") + print("codebase questions and rebuild it after code changes.") + + +def _install_claude_hook(project_dir: Path) -> None: + """Add graphify PreToolUse hook to .claude/settings.json.""" + settings_path = project_dir / ".claude" / "settings.json" + settings_path.parent.mkdir(parents=True, exist_ok=True) + + if settings_path.exists(): + try: + settings = json.loads(settings_path.read_text(encoding="utf-8")) + except json.JSONDecodeError: + settings = {} + else: + settings = {} + + hooks = settings.setdefault("hooks", {}) + pre_tool = hooks.setdefault("PreToolUse", []) + + hooks["PreToolUse"] = [h for h in pre_tool if not (h.get("matcher") in ("Glob|Grep", "Bash", "Read|Glob") and "graphify" in str(h))] + hooks["PreToolUse"].extend(_claude_pretooluse_hooks()) + settings_path.write_text(json.dumps(settings, indent=2), encoding="utf-8") + print(f" .claude/settings.json -> PreToolUse hooks registered (Bash search + Read/Glob)") + + +def _uninstall_claude_hook(project_dir: Path) -> None: + """Remove graphify PreToolUse hook from .claude/settings.json.""" + settings_path = project_dir / ".claude" / "settings.json" + if not settings_path.exists(): + return + try: + settings = json.loads(settings_path.read_text(encoding="utf-8")) + except json.JSONDecodeError: + return + pre_tool = settings.get("hooks", {}).get("PreToolUse", []) + filtered = [h for h in pre_tool if not (h.get("matcher") in ("Glob|Grep", "Bash", "Read|Glob") and "graphify" in str(h))] + if len(filtered) == len(pre_tool): + return + settings["hooks"]["PreToolUse"] = filtered + settings_path.write_text(json.dumps(settings, indent=2), encoding="utf-8") + print(f" .claude/settings.json -> PreToolUse hook removed") + + +def uninstall_all(project_dir: Path | None = None, purge: bool = False) -> None: + """Remove graphify from every platform detected in the current project.""" + pd = project_dir or Path(".") + print("Uninstalling graphify from all detected platforms...\n") + + # Skill-file / config-section uninstallers + claude_uninstall(pd) + codebuddy_uninstall(pd) + gemini_uninstall(pd) + vscode_uninstall(pd) + _cursor_uninstall(pd) + _kiro_uninstall(pd) + _antigravity_uninstall(pd) + # AGENTS.md covers: codex, aider, opencode, claw, droid, trae, trae-cn, hermes, copilot + _agents_uninstall(pd) + # Amp also drops a user-scope skill at ~/.config/agents/skills, which the + # AGENTS.md cleanup above does not touch. + _remove_skill_file("amp") + # The generic agents platform's user-scope skill lives at ~/.agents/skills, + # which neither the AGENTS.md cleanup nor amp's removal reaches. + _remove_skill_file("agents") + _uninstall_opencode_plugin(pd) + _uninstall_codex_hook(pd) + + # Git hook + try: + from graphify.hooks import uninstall as hook_uninstall + result = hook_uninstall(pd) + if result: + print(result) + except Exception: + pass + + if purge: + import shutil as _shutil + out = pd / _GRAPHIFY_OUT + if out.exists(): + _shutil.rmtree(out) + print(f"\n {_GRAPHIFY_OUT}/ -> deleted (--purge)") + else: + print(f"\n {_GRAPHIFY_OUT}/ -> not found (nothing to purge)") + print("\nDone. Run 'pip uninstall graphifyy' to remove the package itself.") +def claude_uninstall(project_dir: Path | None = None, *, project: bool = False) -> None: + """Remove the graphify skill tree (SKILL.md + references/) and the CLAUDE.md section. + Mirrors gemini_uninstall: the bare `graphify uninstall` and `graphify claude + uninstall` must remove the installed skill, not just strip CLAUDE.md, or the + progressive-disclosure tree (SKILL.md + references/) is orphaned (#1121). + """ + project_dir = project_dir or Path(".") + _remove_skill_file("claude", project=project, project_dir=project_dir) + target = project_dir / "CLAUDE.md" + if not target.exists(): + print("No CLAUDE.md found in current directory - nothing to do") + return + content = target.read_text(encoding="utf-8") + if _CLAUDE_MD_MARKER not in content: + print("graphify section not found in CLAUDE.md - nothing to do") + return + # Remove the ## graphify section: from the marker to the next ## heading or EOF + cleaned = re.sub( + r"\n*## graphify\n.*?(?=\n## |\Z)", + "", + content, + flags=re.DOTALL, + ).rstrip() + if cleaned: + target.write_text(cleaned + "\n", encoding="utf-8") + print(f"graphify section removed from {target.resolve()}") + else: + target.unlink() + print(f"CLAUDE.md was empty after removal - deleted {target.resolve()}") + + _uninstall_claude_hook(project_dir or Path(".")) + + +def codebuddy_install(project_dir: Path | None = None) -> None: + """Install the graphify skill and CODEBUDDY.md section for CodeBuddy.""" + _copy_skill_file("codebuddy", project=bool(project_dir), project_dir=project_dir) + target = (project_dir or Path(".")) / "CODEBUDDY.md" + + if target.exists(): + content = target.read_text(encoding="utf-8") + new_content = _replace_or_append_section( + content, _CODEBUDDY_MD_MARKER, _always_on("claude-md") + ) + else: + new_content = _always_on("claude-md") + if target.exists() and new_content == target.read_text(encoding="utf-8"): + print(f"graphify already configured in {target.resolve()} (no change)") + else: + target.write_text(new_content, encoding="utf-8") + print(f"graphify section written to {target.resolve()}") + # Also write CodeBuddy PreToolUse hook to .codebuddy/settings.json + _install_codebuddy_hook(project_dir or Path(".")) + print() + print("CodeBuddy will now check the knowledge graph before answering") + print("codebase questions and rebuild it after code changes.") +def _install_codebuddy_hook(project_dir: Path) -> None: + """Add graphify PreToolUse hook to .codebuddy/settings.json.""" + settings_path = project_dir / ".codebuddy" / "settings.json" + settings_path.parent.mkdir(parents=True, exist_ok=True) + if settings_path.exists(): + try: + settings = json.loads(settings_path.read_text(encoding="utf-8")) + except json.JSONDecodeError: + settings = {} + else: + settings = {} + hooks = settings.setdefault("hooks", {}) + pre_tool = hooks.setdefault("PreToolUse", []) + hooks["PreToolUse"] = [h for h in pre_tool if not (h.get("matcher") in ("Glob|Grep", "Bash", "Read|Glob") and "graphify" in str(h))] + hooks["PreToolUse"].extend(_claude_pretooluse_hooks()) + settings_path.write_text(json.dumps(settings, indent=2), encoding="utf-8") + print(f" .codebuddy/settings.json -> PreToolUse hooks registered") +def _uninstall_codebuddy_hook(project_dir: Path) -> None: + """Remove graphify PreToolUse hook from .codebuddy/settings.json.""" + settings_path = project_dir / ".codebuddy" / "settings.json" + if not settings_path.exists(): + return + try: + settings = json.loads(settings_path.read_text(encoding="utf-8")) + except json.JSONDecodeError: + return + pre_tool = settings.get("hooks", {}).get("PreToolUse", []) + filtered = [h for h in pre_tool if not (h.get("matcher") in ("Glob|Grep", "Bash", "Read|Glob") and "graphify" in str(h))] + if len(filtered) == len(pre_tool): + return + settings["hooks"]["PreToolUse"] = filtered + settings_path.write_text(json.dumps(settings, indent=2), encoding="utf-8") + print(f" .codebuddy/settings.json -> PreToolUse hook removed") +def codebuddy_uninstall(project_dir: Path | None = None, *, project: bool = False) -> None: + """Remove the graphify skill tree (SKILL.md + references/) and the CODEBUDDY.md section.""" + project_dir = project_dir or Path(".") + _remove_skill_file("codebuddy", project=project, project_dir=project_dir) + target = project_dir / "CODEBUDDY.md" + if not target.exists(): + print("No CODEBUDDY.md found in current directory - nothing to do") + return + content = target.read_text(encoding="utf-8") + if _CODEBUDDY_MD_MARKER not in content: + print("graphify section not found in CODEBUDDY.md - nothing to do") + return - - - - - - - - - - - - - + # Remove the ## graphify section: from the marker to the next ## heading or EOF + cleaned = re.sub( + r"\n*## graphify\n.*?(?=\n## |\Z)", + "", + content, + flags=re.DOTALL, + ).rstrip() + if cleaned: + target.write_text(cleaned + "\n", encoding="utf-8") + print(f"graphify section removed from {target.resolve()}") + else: + target.unlink() + print(f"CODEBUDDY.md was empty after removal - deleted {target.resolve()}") + + _uninstall_codebuddy_hook(project_dir or Path(".")) + +def _clone_repo( + url: str, branch: str | None = None, out_dir: Path | None = None +) -> Path: + """Clone a GitHub repo to a local cache dir and return the path. + + Clones into ~/.graphify/repos// by default so repeated + runs on the same URL reuse the existing clone (git pull instead of clone). + """ + import subprocess as _sp + import re as _re + + # Normalise URL — strip trailing .git if present + url = url.rstrip("/") + if not url.endswith(".git"): + git_url = url + ".git" + else: + git_url = url + url = url[:-4] + + # Extract owner/repo from URL + m = _re.search(r"github\.com[:/]([^/]+)/([^/]+?)(?:\.git)?$", url) + if not m: + print(f"error: not a recognised GitHub URL: {url}", file=sys.stderr) + sys.exit(1) + owner, repo = m.group(1), m.group(2) + + if out_dir: + dest = out_dir + else: + dest = Path.home() / ".graphify" / "repos" / owner / repo + + if branch and branch.startswith("-"): + print(f"error: invalid branch name: {branch!r}", file=sys.stderr) + sys.exit(1) + + if dest.exists(): + print(f"Repo already cloned at {dest} - pulling latest...", flush=True) + cmd = ["git", "-C", str(dest), "pull"] + if branch: + cmd += ["origin", "--", branch] + result = _sp.run(cmd, capture_output=True, text=True) + if result.returncode != 0: + print(f"warning: git pull failed:\n{result.stderr}", file=sys.stderr) + else: + dest.parent.mkdir(parents=True, exist_ok=True) + print(f"Cloning {url} -> {dest} ...", flush=True) + cmd = ["git", "clone", "--depth", "1"] + if branch: + cmd += ["--branch", branch] + cmd += ["--", git_url, str(dest)] + result = _sp.run(cmd, capture_output=True, text=True) + if result.returncode != 0: + print(f"error: git clone failed:\n{result.stderr}", file=sys.stderr) + sys.exit(1) + + print(f"Ready at: {dest}", flush=True) + return dest def main() -> None: @@ -568,7 +2434,6 @@ def main() -> None: print(" --out DIR output dir (default: ); writes

/graphify-out/") print(" --google-workspace export .gdoc/.gsheet/.gslides shortcuts via gws before extraction") print(" --no-cluster skip clustering, write raw extraction only") - print(" --code-only index code (local AST, no API key) and skip doc/paper/image files") print(" --postgres DSN extract schema from a live PostgreSQL database") print(" maps tables, views, functions + FK relationships;") print(" column-level detail is not represented in the graph") @@ -654,6 +2519,14 @@ def main() -> None: cmd = sys.argv[1] + # Short-form command aliases. + _CMD_ALIASES = { + "q": "query", "p": "path", "e": "explain", "ex": "extract", + "u": "update", "co": "cluster-only", "opt": "optimize", + "af": "affected", "cl": "clone", "ep": "export", + } + cmd = _CMD_ALIASES.get(cmd, cmd) + # Universal help guard: -h/--help/-? anywhere after the command shows help # and stops — prevents flags from silently triggering destructive subcommands # (e.g. "cursor install --help" was silently installing into Cursor, #821). @@ -664,9 +2537,2899 @@ def main() -> None: print(f"Run 'graphify --help' for full usage.") return - if dispatch_install_cli(cmd): - return - dispatch_command(cmd) + if cmd == "install": + # Default to windows platform on Windows, claude elsewhere + default_platform = "windows" if platform.system() == "Windows" else "claude" + selected_platform: str | None = None + project_scope = False + args = sys.argv[2:] + i = 0 + while i < len(args): + arg = args[i] + if arg in ("-h", "--help"): + _print_install_usage() + return + if arg == "--project": + project_scope = True + i += 1 + elif arg.startswith("--platform="): + candidate = arg.split("=", 1)[1] + if selected_platform and selected_platform != candidate: + print("error: specify install platform only once", file=sys.stderr) + sys.exit(1) + selected_platform = candidate + i += 1 + elif arg == "--platform": + if i + 1 >= len(args): + print("error: --platform requires a value", file=sys.stderr) + sys.exit(1) + candidate = args[i + 1] + if selected_platform and selected_platform != candidate: + print("error: specify install platform only once", file=sys.stderr) + sys.exit(1) + selected_platform = candidate + i += 2 + elif arg.startswith("-"): + print(f"error: unknown install option '{arg}'", file=sys.stderr) + sys.exit(1) + else: + if selected_platform and selected_platform != arg: + print("error: specify install platform only once", file=sys.stderr) + sys.exit(1) + selected_platform = arg + i += 1 + chosen_platform = selected_platform or default_platform + if project_scope: + _project_install(chosen_platform, Path(".")) + else: + install(platform=chosen_platform) + elif cmd == "uninstall": + args = sys.argv[2:] + purge = "--purge" in args + project_scope = "--project" in args + selected_platform = None + i = 0 + while i < len(args): + arg = args[i] + if arg in ("--purge", "--project"): + i += 1 + elif arg.startswith("--platform="): + selected_platform = arg.split("=", 1)[1] + i += 1 + elif arg == "--platform": + if i + 1 >= len(args): + print("error: --platform requires a value", file=sys.stderr) + sys.exit(1) + selected_platform = args[i + 1] + i += 2 + elif arg.startswith("-"): + print(f"error: unknown uninstall option '{arg}'", file=sys.stderr) + sys.exit(1) + else: + selected_platform = arg + i += 1 + if project_scope: + if selected_platform: + _project_uninstall(selected_platform, Path(".")) + else: + _project_uninstall_all(Path(".")) + else: + uninstall_all(purge=purge) + elif cmd == "claude": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + if "--project" in sys.argv[3:]: + _project_install("claude", Path(".")) + else: + claude_install() + elif subcmd == "uninstall": + if "--project" in sys.argv[3:]: + _project_uninstall("claude", Path(".")) + else: + claude_uninstall() + else: + print("Usage: graphify claude [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "codebuddy": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + codebuddy_install() + elif subcmd == "uninstall": + codebuddy_uninstall() + else: + print("Usage: graphify codebuddy [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "gemini": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + gemini_install(project=("--project" in sys.argv[3:])) + elif subcmd == "uninstall": + gemini_uninstall(project=("--project" in sys.argv[3:])) + else: + print("Usage: graphify gemini [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "cursor": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + _cursor_install(Path(".")) + elif subcmd == "uninstall": + _cursor_uninstall(Path(".")) + else: + print("Usage: graphify cursor [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "vscode": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + vscode_install() + elif subcmd == "uninstall": + vscode_uninstall() + else: + print("Usage: graphify vscode [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "copilot": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + if "--project" in sys.argv[3:]: + _project_install("copilot", Path(".")) + else: + install(platform="copilot") + elif subcmd == "uninstall": + if "--project" in sys.argv[3:]: + _project_uninstall("copilot", Path(".")) + else: + removed = _remove_skill_file("copilot") + print("skill removed" if removed else "nothing to remove") + else: + print("Usage: graphify copilot [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "kilo": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + _kilo_install(Path(".")) + elif subcmd == "uninstall": + _kilo_uninstall(Path(".")) + else: + print("Usage: graphify kilo [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "kiro": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + _kiro_install(Path(".")) + elif subcmd == "uninstall": + _kiro_uninstall(Path(".")) + else: + print("Usage: graphify kiro [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "devin": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + if "--project" in sys.argv[3:]: + _project_install("devin", Path(".")) + else: + install(platform="devin") + elif subcmd == "uninstall": + if "--project" in sys.argv[3:]: + _project_uninstall("devin", Path(".")) + else: + removed = _remove_skill_file("devin") + print("skill removed" if removed else "nothing to remove") + else: + print("Usage: graphify devin [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "pi": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + if "--project" in sys.argv[3:]: + _project_install("pi", Path(".")) + else: + install("pi") + elif subcmd == "uninstall": + if "--project" in sys.argv[3:]: + _project_uninstall("pi", Path(".")) + else: + _remove_skill_file("pi") + else: + print("Usage: graphify pi [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "amp": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + if "--project" in sys.argv[3:]: + _project_install("amp", Path(".")) + else: + _amp_install(Path(".")) + elif subcmd == "uninstall": + if "--project" in sys.argv[3:]: + _project_uninstall("amp", Path(".")) + else: + _amp_uninstall(Path(".")) + else: + print("Usage: graphify amp [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd in ("agents", "skills"): + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + if "--project" in sys.argv[3:]: + _project_install("agents", Path(".")) + else: + _agents_platform_install(Path(".")) + elif subcmd == "uninstall": + if "--project" in sys.argv[3:]: + _project_uninstall("agents", Path(".")) + else: + _agents_platform_uninstall(Path(".")) + else: + print(f"Usage: graphify {cmd} [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd in ("aider", "codex", "opencode", "claw", "droid", "trae", "trae-cn", "hermes"): + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + if "--project" in sys.argv[3:]: + _project_install(cmd, Path(".")) + else: + _agents_install(Path("."), cmd) + elif subcmd == "uninstall": + if "--project" in sys.argv[3:]: + _project_uninstall(cmd, Path(".")) + else: + _agents_uninstall(Path("."), platform=cmd) + if cmd == "codex": + _uninstall_codex_hook(Path(".")) + else: + print(f"Usage: graphify {cmd} [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "antigravity": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + if "--project" in sys.argv[3:]: + _project_install("antigravity", Path(".")) + else: + _antigravity_install(Path(".")) + elif subcmd == "uninstall": + if "--project" in sys.argv[3:]: + _project_uninstall("antigravity", Path(".")) + else: + _antigravity_uninstall(Path(".")) + else: + print("Usage: graphify antigravity [install|uninstall]", file=sys.stderr) + sys.exit(1) + elif cmd == "provider": + from graphify.llm import _custom_providers_path, BACKENDS + import json as _json + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + global_path = _custom_providers_path(global_=True) + + if subcmd == "list": + global_path.parent.mkdir(parents=True, exist_ok=True) + existing: dict = {} + if global_path.is_file(): + try: + existing = _json.loads(global_path.read_text(encoding="utf-8")) + except Exception: + pass + if not existing: + print("No custom providers registered.") + else: + for name in existing: + print(f" {name} ({existing[name].get('base_url', '')})") + + elif subcmd == "show": + name = sys.argv[3] if len(sys.argv) > 3 else "" + if not name: + print("Usage: graphify provider show ", file=sys.stderr) + sys.exit(1) + existing = {} + if global_path.is_file(): + try: + existing = _json.loads(global_path.read_text(encoding="utf-8")) + except Exception: + pass + if name not in existing: + print(f"Provider '{name}' not found.", file=sys.stderr) + sys.exit(1) + print(_json.dumps({name: existing[name]}, indent=2)) + + elif subcmd == "add": + args = sys.argv[3:] + name = args[0] if args and not args[0].startswith("-") else "" + if not name: + print("Usage: graphify provider add --base-url URL --default-model MODEL --env-key KEY", file=sys.stderr) + sys.exit(1) + if name in BACKENDS: + print(f"Error: '{name}' is a built-in provider and cannot be overridden.", file=sys.stderr) + sys.exit(1) + base_url = "" + default_model = "" + env_key = "" + pricing_input = 0.0 + pricing_output = 0.0 + i = 1 + while i < len(args): + a = args[i] + if a == "--base-url" and i + 1 < len(args): + base_url = args[i + 1]; i += 2 + elif a.startswith("--base-url="): + base_url = a.split("=", 1)[1]; i += 1 + elif a == "--default-model" and i + 1 < len(args): + default_model = args[i + 1]; i += 2 + elif a.startswith("--default-model="): + default_model = a.split("=", 1)[1]; i += 1 + elif a == "--env-key" and i + 1 < len(args): + env_key = args[i + 1]; i += 2 + elif a.startswith("--env-key="): + env_key = a.split("=", 1)[1]; i += 1 + elif a == "--pricing-input" and i + 1 < len(args): + pricing_input = float(args[i + 1]); i += 2 + elif a == "--pricing-output" and i + 1 < len(args): + pricing_output = float(args[i + 1]); i += 2 + else: + i += 1 + if not base_url or not default_model or not env_key: + print("Error: --base-url, --default-model, and --env-key are required.", file=sys.stderr) + sys.exit(1) + from graphify.llm import provider_base_url_ok + if not provider_base_url_ok(base_url, name): + print(f"Error: refusing to add provider with unsafe base_url {base_url!r}.", file=sys.stderr) + sys.exit(1) + global_path.parent.mkdir(parents=True, exist_ok=True) + existing = {} + if global_path.is_file(): + try: + existing = _json.loads(global_path.read_text(encoding="utf-8")) + except Exception: + pass + existing[name] = { + "base_url": base_url, + "default_model": default_model, + "env_key": env_key, + "pricing": {"input": pricing_input, "output": pricing_output}, + "temperature": 0, + } + global_path.write_text(_json.dumps(existing, indent=2) + "\n", encoding="utf-8") + print(f"Provider '{name}' added. Use with: graphify extract . --backend {name}") + + elif subcmd == "remove": + name = sys.argv[3] if len(sys.argv) > 3 else "" + if not name: + print("Usage: graphify provider remove ", file=sys.stderr) + sys.exit(1) + existing = {} + if global_path.is_file(): + try: + existing = _json.loads(global_path.read_text(encoding="utf-8")) + except Exception: + pass + if name not in existing: + print(f"Provider '{name}' not found.", file=sys.stderr) + sys.exit(1) + del existing[name] + global_path.write_text(_json.dumps(existing, indent=2) + "\n", encoding="utf-8") + print(f"Provider '{name}' removed.") + + else: + print("Usage: graphify provider [add|list|show|remove]", file=sys.stderr) + if subcmd: + sys.exit(1) + elif cmd == "prs": + from graphify.prs import cmd_prs + cmd_prs(sys.argv[2:]) + elif cmd == "hook": + from graphify.hooks import ( + install as hook_install, + uninstall as hook_uninstall, + status as hook_status, + ) + + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd == "install": + print(hook_install(Path("."))) + elif subcmd == "uninstall": + print(hook_uninstall(Path("."))) + elif subcmd == "status": + print(hook_status(Path("."))) + else: + print("Usage: graphify hook [install|uninstall|status]", file=sys.stderr) + sys.exit(1) + elif cmd == "query": + if len(sys.argv) < 3: + print("Usage: graphify query \"\" [--dfs] [--context C] [--budget N] [--graph path]", file=sys.stderr) + sys.exit(1) + from graphify.serve import _query_graph_text + from graphify.security import sanitize_label + from networkx.readwrite import json_graph + from graphify import querylog + + question = sys.argv[2] + use_dfs = "--dfs" in sys.argv + budget = 2000 + graph_path = _default_graph_path() + context_filters: list[str] = [] + args = sys.argv[3:] + i = 0 + while i < len(args): + if args[i] == "--budget" and i + 1 < len(args): + try: + budget = int(args[i + 1]) + except ValueError: + print(f"error: --budget must be an integer", file=sys.stderr) + sys.exit(1) + i += 2 + elif args[i].startswith("--budget="): + try: + budget = int(args[i].split("=", 1)[1]) + except ValueError: + print(f"error: --budget must be an integer", file=sys.stderr) + sys.exit(1) + i += 1 + elif args[i] == "--context" and i + 1 < len(args): + context_filters.append(args[i + 1]) + i += 2 + elif args[i].startswith("--context="): + context_filters.append(args[i].split("=", 1)[1]) + i += 1 + elif args[i] == "--graph" and i + 1 < len(args): + graph_path = args[i + 1] + i += 2 + else: + i += 1 + gp = Path(graph_path).resolve() + if not gp.exists(): + print(f"error: graph file not found: {gp}", file=sys.stderr) + sys.exit(1) + if not gp.suffix == ".json": + print(f"error: graph file must be a .json file", file=sys.stderr) + sys.exit(1) + _enforce_graph_size_cap_or_exit(gp) + try: + import json as _json + import networkx as _nx + + _raw = _json.loads(gp.read_text(encoding="utf-8")) + if "links" not in _raw and "edges" in _raw: + _raw = dict(_raw, links=_raw["edges"]) + try: + G = json_graph.node_link_graph(_raw, edges="links") + except TypeError: + G = json_graph.node_link_graph(_raw) + try: + from graphify.build import graph_has_legacy_ids as _legacy + if _legacy(_raw.get("nodes", [])): + print( + "[graphify] note: this graph uses the pre-#1504 node-ID scheme; " + "rebuild with `graphify extract --force` to get path-qualified IDs " + "(fixes same-name-file collisions).", + file=sys.stderr, + ) + except Exception: + pass + except Exception as exc: + print(f"error: could not load graph: {exc}", file=sys.stderr) + sys.exit(1) + import time as _time + _t0 = _time.perf_counter() + _mode = "dfs" if use_dfs else "bfs" + _result = _query_graph_text( + G, + question, + mode=_mode, + depth=2, + token_budget=budget, + context_filters=context_filters, + ) + querylog.log_query( + kind="query", + question=question, + corpus=str(gp), + result=_result, + mode=_mode, + depth=2, + token_budget=budget, + duration_ms=(_time.perf_counter() - _t0) * 1000, + ) + print(_result) + elif cmd == "affected": + if len(sys.argv) < 3: + print("Usage: graphify affected \"\" [--relation R] [--depth N] [--graph path]", file=sys.stderr) + sys.exit(1) + from graphify.affected import DEFAULT_AFFECTED_RELATIONS, format_affected, load_graph + query = sys.argv[2] + graph_path = _default_graph_path() + depth = 2 + relations: list[str] = [] + args = sys.argv[3:] + i = 0 + while i < len(args): + if args[i] == "--graph" and i + 1 < len(args): + graph_path = args[i + 1] + i += 2 + elif args[i].startswith("--graph="): + graph_path = args[i].split("=", 1)[1] + i += 1 + elif args[i] == "--depth" and i + 1 < len(args): + try: + depth = int(args[i + 1]) + except ValueError: + print("error: --depth must be an integer", file=sys.stderr) + sys.exit(1) + i += 2 + elif args[i].startswith("--depth="): + try: + depth = int(args[i].split("=", 1)[1]) + except ValueError: + print("error: --depth must be an integer", file=sys.stderr) + sys.exit(1) + i += 1 + elif args[i] == "--relation" and i + 1 < len(args): + relations.append(args[i + 1]) + i += 2 + elif args[i].startswith("--relation="): + relations.append(args[i].split("=", 1)[1]) + i += 1 + else: + i += 1 + gp = Path(graph_path).resolve() + if not gp.exists(): + print(f"error: graph file not found: {gp}", file=sys.stderr) + sys.exit(1) + if not gp.suffix == ".json": + print("error: graph file must be a .json file", file=sys.stderr) + sys.exit(1) + try: + graph = load_graph(gp) + except Exception as exc: + print(f"error: could not load graph: {exc}", file=sys.stderr) + sys.exit(1) + print( + format_affected( + graph, + query, + relations=relations or DEFAULT_AFFECTED_RELATIONS, + depth=depth, + ) + ) + elif cmd == "save-result": + # graphify save-result --question Q --answer A [--type T] [--nodes N1 N2 ...] + # [--outcome useful|dead_end|corrected] [--correction TEXT] + import argparse as _ap + + p = _ap.ArgumentParser(prog="graphify save-result") + p.add_argument("--question", required=True) + p.add_argument("--answer", default=None) + p.add_argument("--answer-file", dest="answer_file", default=None) + p.add_argument("--type", dest="query_type", default="query") + p.add_argument("--nodes", nargs="*", default=[]) + p.add_argument("--outcome", choices=("useful", "dead_end", "corrected"), default=None) + p.add_argument("--correction", default=None) + p.add_argument("--memory-dir", default=str(Path(_GRAPHIFY_OUT) / "memory")) + opts = p.parse_args(sys.argv[2:]) + if opts.answer_file: + opts.answer = Path(opts.answer_file).read_text(encoding="utf-8").strip() + elif not opts.answer: + p.error("--answer or --answer-file is required") + from graphify.ingest import save_query_result as _sqr + + out = _sqr( + question=opts.question, + answer=opts.answer, + memory_dir=Path(opts.memory_dir), + query_type=opts.query_type, + source_nodes=opts.nodes or None, + outcome=opts.outcome, + correction=opts.correction, + ) + print(f"Saved to {out}") + elif cmd == "reflect": + import argparse as _ap + + p = _ap.ArgumentParser(prog="graphify reflect") + p.add_argument("--memory-dir", default=str(Path(_GRAPHIFY_OUT) / "memory")) + p.add_argument( + "--out", + default=str(Path(_GRAPHIFY_OUT) / "reflections" / "LESSONS.md"), + ) + p.add_argument("--graph", default=None) + p.add_argument("--analysis", default=None) + p.add_argument("--labels", default=None) + p.add_argument("--half-life-days", type=float, default=30.0, + help="signal weight halves every N days (default 30)") + p.add_argument("--min-corroboration", type=int, default=2, + help="distinct useful results to promote a node to preferred (default 2)") + p.add_argument("--if-stale", action="store_true", + help="skip when LESSONS.md is already newer than every input " + "(e.g. the git hook just refreshed it)") + opts = p.parse_args(sys.argv[2:]) + from graphify.reflect import reflect as _reflect, lessons_fresh as _lessons_fresh + + graph_arg = opts.graph + if graph_arg is None: + default_graph = Path(_GRAPHIFY_OUT) / "graph.json" + if default_graph.exists(): + graph_arg = str(default_graph) + + _gp = Path(graph_arg) if graph_arg else None + _analysis_path = None + _labels_path = None + if _gp is not None: + _analysis_path = Path(opts.analysis) if opts.analysis else ( + _gp.parent / ".graphify_analysis.json") + _labels_path = Path(opts.labels) if opts.labels else ( + _gp.parent / ".graphify_labels.json") + + if opts.if_stale and _lessons_fresh( + Path(opts.out), Path(opts.memory_dir), _gp, _analysis_path, _labels_path + ): + print(f"Lessons already up to date -> {opts.out} (skipped; omit --if-stale to force)") + else: + out_path, agg = _reflect( + memory_dir=Path(opts.memory_dir), + out_path=Path(opts.out), + graph_path=_gp, + analysis_path=_analysis_path, + labels_path=_labels_path, + half_life_days=opts.half_life_days, + min_corroboration=opts.min_corroboration, + ) + c = agg["counts"] + print( + f"Reflected {agg['total']} memories " + f"({c['useful']} useful, {c['dead_end']} dead ends, " + f"{c['corrected']} corrected) -> {out_path}" + ) + elif cmd == "path": + if len(sys.argv) < 4: + print( + 'Usage: graphify path "" "" [--graph path]', + file=sys.stderr, + ) + sys.exit(1) + from graphify.serve import _score_nodes + from networkx.readwrite import json_graph + import networkx as _nx + + source_label = sys.argv[2] + target_label = sys.argv[3] + graph_path = _default_graph_path() + args = sys.argv[4:] + for i, a in enumerate(args): + if a == "--graph" and i + 1 < len(args): + graph_path = args[i + 1] + gp = Path(graph_path).resolve() + if not gp.exists(): + print(f"error: graph file not found: {gp}", file=sys.stderr) + sys.exit(1) + _enforce_graph_size_cap_or_exit(gp) + _raw = json.loads(gp.read_text(encoding="utf-8")) + if "links" not in _raw and "edges" in _raw: + _raw = dict(_raw, links=_raw["edges"]) + # Force directed so the renderer can recover stored caller→callee direction. + _raw = {**_raw, "directed": True} + try: + G = json_graph.node_link_graph(_raw, edges="links") + except TypeError: + G = json_graph.node_link_graph(_raw) + src_scored = _score_nodes(G, [t.lower() for t in source_label.split()]) + tgt_scored = _score_nodes(G, [t.lower() for t in target_label.split()]) + if not src_scored: + print(f"No node matching '{source_label}' found.", file=sys.stderr) + sys.exit(1) + if not tgt_scored: + print(f"No node matching '{target_label}' found.", file=sys.stderr) + sys.exit(1) + src_nid, tgt_nid = src_scored[0][1], tgt_scored[0][1] + # Ambiguity guard: when both queries resolve to the same node, the + # shortest path is trivially zero hops, which is almost never what the + # caller wanted (see bug #828). + if src_nid == tgt_nid: + print( + f"'{source_label}' and '{target_label}' both resolved to the same " + f"node '{src_nid}'. Use a more specific label or the exact node ID.", + file=sys.stderr, + ) + sys.exit(1) + for _name, _scored in (("source", src_scored), ("target", tgt_scored)): + if len(_scored) >= 2: + _top, _runner = _scored[0][0], _scored[1][0] + if _top > 0 and (_top - _runner) / _top < 0.10: + print( + f"warning: {_name} match was ambiguous " + f"(top score {_top:g}, runner-up {_runner:g})", + file=sys.stderr, + ) + try: + path_nodes = _nx.shortest_path(G.to_undirected(as_view=True), src_nid, tgt_nid) + except (_nx.NetworkXNoPath, _nx.NodeNotFound): + print(f"No path found between '{source_label}' and '{target_label}'.") + sys.exit(0) + hops = len(path_nodes) - 1 + segments = [] + from graphify.build import edge_data + for i in range(len(path_nodes) - 1): + u, v = path_nodes[i], path_nodes[i + 1] + # Check which direction the stored edge points. + if G.has_edge(u, v): + edata = edge_data(G, u, v) + forward = True + else: + edata = edge_data(G, v, u) + forward = False + rel = edata.get("relation", "") + conf = edata.get("confidence", "") + conf_str = f" [{conf}]" if conf else "" + if i == 0: + segments.append(G.nodes[u].get("label", u)) + if forward: + segments.append(f"--{rel}{conf_str}--> {G.nodes[v].get('label', v)}") + else: + segments.append(f"<--{rel}{conf_str}-- {G.nodes[v].get('label', v)}") + print(f"Shortest path ({hops} hops):\n " + " ".join(segments)) + from graphify import querylog + querylog.log_query( + kind="path", + question=f"{sys.argv[2]} -> {sys.argv[3]}", + corpus=str(gp), + nodes_returned=hops, + ) + + elif cmd == "explain": + if len(sys.argv) < 3: + print('Usage: graphify explain "" [--graph path]', file=sys.stderr) + sys.exit(1) + from graphify.serve import _find_node + from networkx.readwrite import json_graph + + label = sys.argv[2] + graph_path = _default_graph_path() + args = sys.argv[3:] + for i, a in enumerate(args): + if a == "--graph" and i + 1 < len(args): + graph_path = args[i + 1] + gp = Path(graph_path).resolve() + if not gp.exists(): + print(f"error: graph file not found: {gp}", file=sys.stderr) + sys.exit(1) + _enforce_graph_size_cap_or_exit(gp) + _raw = json.loads(gp.read_text(encoding="utf-8")) + if "links" not in _raw and "edges" in _raw: + _raw = dict(_raw, links=_raw["edges"]) + # Force directed so the renderer can recover stored caller→callee direction. + _raw = {**_raw, "directed": True} + try: + G = json_graph.node_link_graph(_raw, edges="links") + except TypeError: + G = json_graph.node_link_graph(_raw) + matches = _find_node(G, label) + if not matches: + print(f"No node matching '{label}' found.") + sys.exit(0) + nid = matches[0] + d = G.nodes[nid] + print(f"Node: {d.get('label', nid)}") + print(f" ID: {nid}") + print( + f" Source: {d.get('source_file', '')} {d.get('source_location', '')}".rstrip() + ) + print(f" Type: {d.get('file_type', '')}") + print(f" Community: {d.get('community_name') or d.get('community', '')}") + # Work-memory overlay: a derived experiential hint from `graphify reflect`, + # merged in display-only from the .graphify_learning.json sidecar next to + # graph.json. No line when the node has no overlay entry. + try: + from graphify.reflect import load_learning_overlay as _llo + from graphify.security import sanitize_label as _sl + _overlay = _llo(gp) + _entry = _overlay.get(str(nid)) + if _entry: + _status = _sl(str(_entry.get("status", ""))) + if _status == "contested": + _line = (f" Lesson: contested (useful {_entry.get('uses', 0)} / " + f"dead-end {_entry.get('neg', 0)})") + elif _status == "preferred": + _line = (f" Lesson: preferred source (start here) — " + f"{_entry.get('uses', 0)} useful, score={_entry.get('score', 0)}") + else: + _line = (f" Lesson: {_status or 'tentative'} — " + f"{_entry.get('uses', 0)} useful, score={_entry.get('score', 0)}") + if _entry.get("stale"): + _line += " [code changed since — re-verify]" + print(_line) + except Exception: + pass + print(f" Degree: {G.degree(nid)}") + from graphify.build import edge_data + connections: list[tuple[str, str, dict]] = [] # (direction, neighbor_id, edge_data) + for nb in G.successors(nid): + connections.append(("out", nb, edge_data(G, nid, nb))) + for nb in G.predecessors(nid): + connections.append(("in", nb, edge_data(G, nb, nid))) + if connections: + print(f"\nConnections ({len(connections)}):") + connections.sort(key=lambda c: G.degree(c[1]), reverse=True) + for direction, nb, edata in connections[:20]: + rel = edata.get("relation", "") + conf = edata.get("confidence", "") + arrow = "-->" if direction == "out" else "<--" + print(f" {arrow} {G.nodes[nb].get('label', nb)} [{rel}] [{conf}]") + if len(connections) > 20: + print(f" ... and {len(connections) - 20} more") + from graphify import querylog + querylog.log_query( + kind="explain", + question=sys.argv[2], + corpus=str(gp), + nodes_returned=len(connections), + ) + + elif cmd == "diagnose": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd != "multigraph": + print( + "Usage: graphify diagnose multigraph " + "[--graph path] [--json] [--max-examples N] " + "[--directed] [--undirected] [--extract-path path]", + file=sys.stderr, + ) + sys.exit(1) + + graph_path = Path(_default_graph_path()) + max_examples = 5 + directed: bool | None = None + direction_flag: str | None = None + json_output = False + extract_path: Path | None = None + + i = 3 + while i < len(sys.argv): + arg = sys.argv[i] + if arg == "--graph": + i += 1 + if i >= len(sys.argv): + print("error: --graph requires a path", file=sys.stderr) + sys.exit(1) + graph_path = Path(sys.argv[i]) + elif arg == "--json": + json_output = True + elif arg == "--max-examples": + i += 1 + if i >= len(sys.argv): + print("error: --max-examples requires an integer", file=sys.stderr) + sys.exit(1) + try: + max_examples = int(sys.argv[i]) + except ValueError: + print("error: --max-examples requires an integer", file=sys.stderr) + sys.exit(1) + if max_examples < 0: + print("error: --max-examples must be >= 0", file=sys.stderr) + sys.exit(1) + elif arg == "--directed": + if direction_flag == "undirected": + print( + "error: --directed and --undirected are mutually exclusive", + file=sys.stderr, + ) + sys.exit(1) + direction_flag = "directed" + directed = True + elif arg == "--undirected": + if direction_flag == "directed": + print( + "error: --directed and --undirected are mutually exclusive", + file=sys.stderr, + ) + sys.exit(1) + direction_flag = "undirected" + directed = False + elif arg == "--extract-path": + i += 1 + if i >= len(sys.argv): + print("error: --extract-path requires a path", file=sys.stderr) + sys.exit(1) + extract_path = Path(sys.argv[i]) + else: + print(f"error: unknown diagnose option {arg}", file=sys.stderr) + sys.exit(1) + i += 1 + + from graphify.diagnostics import ( + diagnose_file, + format_diagnostic_json, + format_diagnostic_report, + ) + + try: + summary = diagnose_file( + graph_path, + directed=directed, + root=Path(".").resolve(), + max_examples=max_examples, + extract_path=extract_path, + ) + except Exception as exc: + print(f"error: {exc}", file=sys.stderr) + sys.exit(1) + + if json_output: + print(json.dumps(format_diagnostic_json(summary), indent=2)) + else: + print(format_diagnostic_report(summary)) + + elif cmd == "add": + if len(sys.argv) < 3: + print( + "Usage: graphify add [--author Name] [--contributor Name] [--dir ./raw]", + file=sys.stderr, + ) + sys.exit(1) + from graphify.ingest import ingest as _ingest + + url = sys.argv[2] + author: str | None = None + contributor: str | None = None + target_dir = Path("raw") + args = sys.argv[3:] + i = 0 + while i < len(args): + if args[i] == "--author" and i + 1 < len(args): + author = args[i + 1] + i += 2 + elif args[i] == "--contributor" and i + 1 < len(args): + contributor = args[i + 1] + i += 2 + elif args[i] == "--dir" and i + 1 < len(args): + target_dir = Path(args[i + 1]) + i += 2 + else: + i += 1 + try: + saved = _ingest(url, target_dir, author=author, contributor=contributor) + print(f"Saved to {saved}") + print("Run /graphify --update in your AI assistant to update the graph.") + except Exception as exc: + print(f"error: {exc}", file=sys.stderr) + sys.exit(1) + + elif cmd == "watch": + watch_path = Path(sys.argv[2]) if len(sys.argv) > 2 else Path(".") + if not watch_path.exists(): + print(f"error: path not found: {watch_path}", file=sys.stderr) + sys.exit(1) + from graphify.watch import watch as _watch + + try: + _watch(watch_path) + except ImportError as exc: + print(f"error: {exc}", file=sys.stderr) + sys.exit(1) + + elif cmd in ("cluster-only", "label"): + # `label` is `cluster-only` that always (re)generates community names with + # the configured backend, even when a .graphify_labels.json already exists. + force_relabel = cmd == "label" + # Mirror the tree/export arg-parsing pattern: walk argv so flags and + # the optional positional path can appear in any order (#724). + no_viz = "--no-viz" in sys.argv + no_label = "--no-label" in sys.argv + missing_only = "--missing-only" in sys.argv + co_timing = "--timing" in sys.argv + _backend_arg = next((a for a in sys.argv if a.startswith("--backend=")), None) + label_backend = _backend_arg.split("=", 1)[1] if _backend_arg else None + _model_arg = next((a for a in sys.argv if a.startswith("--model=")), None) + label_model = _model_arg.split("=", 1)[1] if _model_arg else None + _min_cs_arg = next((a for a in sys.argv if a.startswith("--min-community-size=")), None) + min_community_size = int(_min_cs_arg.split("=")[1]) if _min_cs_arg else 3 + args = sys.argv[2:] + watch_path: Path | None = None + graph_override: Path | None = None + co_resolution: float = 1.0 + co_exclude_hubs: float | None = None + label_max_concurrency: int = 4 + label_batch_size: int = 100 + i_arg = 0 + while i_arg < len(args): + a = args[i_arg] + if a == "--graph" and i_arg + 1 < len(args): + graph_override = Path(args[i_arg + 1]); i_arg += 2 + elif a == "--backend" and i_arg + 1 < len(args): + label_backend = args[i_arg + 1]; i_arg += 2 + elif a.startswith("--backend="): + label_backend = a.split("=", 1)[1]; i_arg += 1 + elif a == "--model" and i_arg + 1 < len(args): + label_model = args[i_arg + 1]; i_arg += 2 + elif a.startswith("--model="): + label_model = a.split("=", 1)[1]; i_arg += 1 + elif a == "--resolution" and i_arg + 1 < len(args): + co_resolution = float(args[i_arg + 1]); i_arg += 2 + elif a.startswith("--resolution="): + co_resolution = float(a.split("=", 1)[1]); i_arg += 1 + elif a == "--exclude-hubs" and i_arg + 1 < len(args): + co_exclude_hubs = float(args[i_arg + 1]); i_arg += 2 + elif a.startswith("--exclude-hubs="): + co_exclude_hubs = float(a.split("=", 1)[1]); i_arg += 1 + elif a == "--max-concurrency" and i_arg + 1 < len(args): + label_max_concurrency = int(args[i_arg + 1]); i_arg += 2 + elif a.startswith("--max-concurrency="): + label_max_concurrency = int(a.split("=", 1)[1]); i_arg += 1 + elif a == "--batch-size" and i_arg + 1 < len(args): + label_batch_size = int(args[i_arg + 1]); i_arg += 2 + elif a.startswith("--batch-size="): + label_batch_size = int(a.split("=", 1)[1]); i_arg += 1 + elif a in ("--no-viz", "--missing-only") or a.startswith("--min-community-size="): + i_arg += 1 + elif a.startswith("--"): + i_arg += 1 + elif watch_path is None: + watch_path = Path(a); i_arg += 1 + else: + i_arg += 1 + if watch_path is None: + watch_path = Path(".") + graph_json = graph_override if graph_override is not None else watch_path / _GRAPHIFY_OUT / "graph.json" + if not graph_json.exists(): + print( + f"error: no graph found at {graph_json} — run /graphify first", + file=sys.stderr, + ) + sys.exit(1) + from networkx.readwrite import json_graph as _jg + from graphify.build import build_from_json + from graphify.cluster import cluster, score_all, remap_communities_to_previous + from graphify.analyze import ( + god_nodes, + surprising_connections, + suggest_questions, + ) + from graphify.report import generate + from graphify.export import to_json, to_html + + stages = _StageTimer(co_timing) + print("Loading existing graph...") + # Solution 3 (#1019): don't hard-exit on an oversized graph.json here. + # Core outputs (graph.json + GRAPH_REPORT.md) still get written; the + # graph.html render below falls back to the community-aggregation view + # (node_limit=5000) when over the cap. + from graphify.security import check_graph_file_size_cap as _check_cap + _over_cap = False + try: + _check_cap(graph_json) + except ValueError: + _over_cap = True + try: + _over_cap_bytes = graph_json.stat().st_size + except OSError: + _over_cap_bytes = -1 + print( + f"warning: graph.json exceeds cap ({_over_cap_bytes} bytes); " + f"falling back to community-aggregation view (node_limit=5000)", + file=sys.stderr, + ) + _raw = json.loads(graph_json.read_text(encoding="utf-8")) + _directed = bool(_raw.get("directed", False)) + G = build_from_json(_raw, directed=_directed) + print(f"Graph: {G.number_of_nodes()} nodes, {G.number_of_edges()} edges") + stages.mark("load") + print("Re-clustering...") + communities = cluster(G, resolution=co_resolution, exclude_hubs_percentile=co_exclude_hubs) + # Mirror the watch/update path (#822): map new cids to prior ones by + # node-overlap so the existing .graphify_labels.json keeps attaching + # to the same conceptual community after re-clustering. Without this, + # labels follow raw cid index and become misaligned whenever the + # graph has changed between labeling and cluster-only (#1027). + previous_node_community = { + n["id"]: n["community"] + for n in _raw.get("nodes", []) + if n.get("community") is not None and n.get("id") is not None + } + if previous_node_community: + communities = remap_communities_to_previous(communities, previous_node_community) + stages.mark("cluster") + cohesion = score_all(G, communities) + gods = god_nodes(G) + surprises = surprising_connections(G, communities) + stages.mark("analyze") + out = watch_path / _GRAPHIFY_OUT + out.mkdir(parents=True, exist_ok=True) + labels_path = out / ".graphify_labels.json" + existing_labels: dict[int, str] = {} + if labels_path.exists(): + try: + existing_labels = { + int(k): v + for k, v in json.loads(labels_path.read_text(encoding="utf-8")).items() + if isinstance(v, str) + } + except Exception: + existing_labels = {} + # Accumulate token usage from the labeling LLM calls so cluster-only mode + # reports real cost instead of a hardcoded zero (#1694). Stays {0, 0} on + # the reuse / no-label paths, which make no LLM calls. + label_token_usage = {"input": 0, "output": 0} + if labels_path.exists() and not force_relabel: + # Reuse saved labels, but don't blindly trust them: the graph may have + # been re-scoped/re-clustered since labeling, in which case a cid now + # covers a DIFFERENT community and its old (LLM) name is wrong (#label-stale). + # Validate each community against the membership signature saved beside the + # labels; any community that changed (or has no saved label) is renamed by + # its current hub — deterministic and correct-by-construction — and the user + # is told to `graphify label` for fresh LLM names. Unchanged communities keep + # their saved label. When no signature sidecar exists (labels predate this), + # fall back to hub-filling only the communities missing a label. + from graphify.cluster import community_member_sigs, label_communities_by_hub + sig_path = labels_path.parent / (labels_path.name + ".sig") + saved_sigs: dict[int, str] = {} + if sig_path.exists(): + try: + saved_sigs = { + int(k): v for k, v in + json.loads(sig_path.read_text(encoding="utf-8")).items() + if isinstance(v, str) + } + except Exception: + saved_sigs = {} + cur_sigs = community_member_sigs(communities) + count_mismatch = len(existing_labels) != len(communities) + labels = {} + hub_labels: dict[int, str] | None = None + changed = 0 + for cid in communities: + have_label = cid in existing_labels + if saved_sigs: + # Precise: the membership signature tells us if this exact + # community changed since it was labeled. + fresh = have_label and saved_sigs.get(cid) == cur_sigs.get(cid) + else: + # No signature sidecar (labels predate it). A differing community + # COUNT means the labels describe a different clustering, so a cid's + # old label can't be trusted; equal count is the best "same" signal. + fresh = have_label and not count_mismatch + if fresh: + labels[cid] = existing_labels[cid] + else: + if hub_labels is None: + hub_labels = label_communities_by_hub(G, communities) + labels[cid] = hub_labels[cid] + if have_label: + changed += 1 + if changed: + print( + f"[graphify] community set changed since labeling " + f"({len(existing_labels)} saved labels, {len(communities)} communities now; " + f"renamed {changed} community(ies) by their hub). " + f"Run `graphify label` to refresh names with the LLM.", + file=sys.stderr, + ) + elif no_label and not force_relabel: + labels = {cid: f"Community {cid}" for cid in communities} + else: + # No labels file yet (or `graphify label` forced a refresh). When run + # standalone there is no orchestrating agent to do skill.md Step 5, so + # auto-name communities rather than leave "Community N" (#1097). + from graphify.cluster import label_communities_by_hub + from graphify.llm import generate_community_labels + print("Labeling communities...") + # Deterministic, LLM-free base labels: name each community after its + # highest-degree hub, so the report is readable even with no backend + # (previously bare "Community N"). A configured LLM backend overrides these + # with richer names below; its no-backend placeholder fallback does NOT. + hub_labels = label_communities_by_hub(G, communities) + label_communities_input = communities + labels = dict(hub_labels) + if missing_only: + labels = { + cid: existing_labels.get(cid, hub_labels[cid]) + for cid in communities + } + label_communities_input = { + cid: members + for cid, members in communities.items() + if cid not in existing_labels or existing_labels.get(cid) == f"Community {cid}" + } + generated_labels, _ = generate_community_labels( + G, label_communities_input, backend=label_backend, model=label_model, gods=gods, + max_concurrency=label_max_concurrency, batch_size=label_batch_size, + usage_out=label_token_usage, + ) + # Only let the LLM OVERRIDE where it produced a real name — its no-backend + # fallback returns "Community {cid}" placeholders, which must not clobber + # the deterministic hub labels. + labels.update({ + cid: v for cid, v in generated_labels.items() + if v and v != f"Community {cid}" + }) + stages.mark("label") + questions = suggest_questions(G, communities, labels) + tokens = label_token_usage + from graphify.export import _git_head as _gh + _commit = _gh() + from graphify.report import load_learning_for_report as _llfr + report = generate(G, communities, cohesion, labels, gods, surprises, + {"warning": "cluster-only mode — file stats not available"}, + tokens, str(watch_path), suggested_questions=questions, + min_community_size=min_community_size, built_at_commit=_commit, + learning=_llfr(out / "graph.json")) + (out / "GRAPH_REPORT.md").write_text(report, encoding="utf-8") + stages.mark("report") + from graphify.export import backup_if_protected as _backup + _backup(out) + analysis = { + "communities": {str(k): v for k, v in communities.items()}, + "cohesion": {str(k): v for k, v in cohesion.items()}, + "gods": gods, + "surprises": surprises, + "questions": questions, + } + (out / ".graphify_analysis.json").write_text( + json.dumps(analysis, indent=2, ensure_ascii=False), + encoding="utf-8", + ) + to_json(G, communities, str(out / "graph.json"), community_labels=labels) + labels_path.write_text(json.dumps({str(k): v for k, v in labels.items()}, ensure_ascii=False), encoding="utf-8") + # Membership signatures beside the labels so a later cluster-only can detect + # which communities changed and avoid reusing a stale label (see reuse above). + from graphify.cluster import community_member_sigs as _cms + (labels_path.parent / (labels_path.name + ".sig")).write_text( + json.dumps({str(k): v for k, v in _cms(communities).items()}), encoding="utf-8") + + # Mirror watch.py pattern: gate to_html so core outputs (graph.json + + # GRAPH_REPORT.md) always land. Honor --no-viz explicitly; otherwise + # fall back to ValueError handling so an oversized graph doesn't crash + # the CLI mid-write and leave a stale graph.html on disk. + html_target = out / "graph.html" + if no_viz: + if html_target.exists(): + html_target.unlink() + stages.mark("export"); stages.total() + print(f"Done - {len(communities)} communities. GRAPH_REPORT.md and graph.json updated (--no-viz; graph.html removed).") + else: + try: + # Over-cap fallback (#1019): force the community-aggregation + # path so an oversized graph still renders a usable graph.html. + _node_limit = 5000 if _over_cap else None + to_html(G, communities, str(html_target), community_labels=labels or None, + node_limit=_node_limit) + stages.mark("export"); stages.total() + print(f"Done - {len(communities)} communities. GRAPH_REPORT.md, graph.json and graph.html updated.") + except ValueError as viz_err: + if html_target.exists(): + html_target.unlink() + print(f"Skipped graph.html: {viz_err}") + stages.mark("export"); stages.total() + print(f"Done - {len(communities)} communities. GRAPH_REPORT.md and graph.json updated.") + + elif cmd == "update": + force = os.environ.get("GRAPHIFY_FORCE", "").lower() in ("1", "true", "yes") + no_cluster = False + cpp_deep_update = False + cpp_workers_update: int | None = None + compile_db_update: str | None = None + args = sys.argv[2:] + watch_arg: str | None = None + for a in args: + if a == "--force": + force = True + continue + if a == "--no-cluster": + no_cluster = True + continue + if a == "--cpp-deep": + cpp_deep_update = True + continue + if a.startswith("--cpp-workers="): + cpp_workers_update = int(a.split("=", 1)[1]) + continue + if a.startswith("--compile-db="): + compile_db_update = a.split("=", 1)[1] + continue + if a.startswith("-"): + print(f"error: unknown update option: {a}", file=sys.stderr) + sys.exit(2) + if watch_arg is not None: + print("error: update accepts at most one path argument", file=sys.stderr) + sys.exit(2) + watch_arg = a + + if watch_arg is not None: + watch_path = Path(watch_arg) + else: + # Try to recover the scan root saved by the last full build + saved = Path(_GRAPHIFY_OUT) / ".graphify_root" + if saved.exists(): + watch_path = Path(saved.read_text(encoding="utf-8").strip()) + else: + watch_path = Path(".") + if not watch_path.exists(): + print(f"error: path not found: {watch_path}", file=sys.stderr) + sys.exit(1) + from graphify.watch import _rebuild_code + + if cpp_deep_update: + os.environ["GRAPHIFY_CPP_DEEP"] = "1" + if compile_db_update: + os.environ["GRAPHIFY_CPP_COMPILE_DB"] = compile_db_update + if cpp_workers_update is not None and cpp_workers_update > 0: + os.environ["GRAPHIFY_CPP_WORKERS"] = str(cpp_workers_update) + print(f"Re-extracting code files in {watch_path} (cpp-deep enabled)...") + else: + print(f"Re-extracting code files in {watch_path} (no LLM needed)...") + # Interactive CLI: block on the per-repo lock rather than skip, so the + # user sees their explicit `graphify update` complete instead of + # exiting silently when a hook-driven rebuild happens to be running. + ok = _rebuild_code(watch_path, force=force, no_cluster=no_cluster, block_on_lock=True) + if ok: + print("Code graph updated. For doc/paper/image changes run /graphify --update in your AI assistant.") + if not ( + os.environ.get("GEMINI_API_KEY") + or os.environ.get("GOOGLE_API_KEY") + or os.environ.get("MOONSHOT_API_KEY") + or os.environ.get("DEEPSEEK_API_KEY") + or os.environ.get("GRAPHIFY_NO_TIPS") + ): + print("Tip: set GEMINI_API_KEY or GOOGLE_API_KEY to use Gemini for semantic extraction.") + else: + print( + "Nothing to update or rebuild failed — check output above.", + file=sys.stderr, + ) + sys.exit(1) + + elif cmd == "hook-check": + # Codex Desktop rejects hookSpecificOutput.additionalContext on PreToolUse. + # Keep this as a cross-platform no-op so installed hooks never break Bash + # tool calls. Graph guidance reaches the agent via AGENTS.md / skill instead. + sys.exit(0) + elif cmd == "hook-guard": + # Shell-agnostic Claude/Codebuddy PreToolUse guard (#522). Replaces the old + # inline-bash hooks that failed on Windows. Prints an additionalContext nudge + # toward graphify when a graph exists; always exits 0 (never blocks a tool). + _run_hook_guard(sys.argv[2] if len(sys.argv) > 2 else "") + sys.exit(0) + elif cmd == "check-update": + if len(sys.argv) < 3: + print("Usage: graphify check-update ", file=sys.stderr) + sys.exit(1) + from graphify.watch import check_update + + check_update(Path(sys.argv[2]).resolve()) + sys.exit(0) + elif cmd == "tree": + # Emit a D3 v7 collapsible-tree HTML view of graph.json: + # expand-all / collapse-all / reset-view buttons, multi-line + # wrapText labels with separately-coloured name + count, + # depth-based palette, click-to-toggle subtree, hover inspector + # showing top-K outbound edges per symbol. + from typing import Optional as _Opt + from graphify.tree_html import write_tree_html, DEFAULT_MAX_CHILDREN + graph_path = Path(_GRAPHIFY_OUT) / "graph.json" + output_path: "_Opt[Path]" = None + root: "_Opt[str]" = None + max_children = DEFAULT_MAX_CHILDREN + top_k_edges = 0 + project_label: "_Opt[str]" = None + args = sys.argv[2:] + i_arg = 0 + while i_arg < len(args): + a = args[i_arg] + if a == "--graph" and i_arg + 1 < len(args): + graph_path = Path(args[i_arg + 1]); i_arg += 2 + elif a == "--output" and i_arg + 1 < len(args): + output_path = Path(args[i_arg + 1]); i_arg += 2 + elif a == "--root" and i_arg + 1 < len(args): + root = args[i_arg + 1]; i_arg += 2 + elif a == "--max-children" and i_arg + 1 < len(args): + max_children = int(args[i_arg + 1]); i_arg += 2 + elif a == "--top-k-edges" and i_arg + 1 < len(args): + top_k_edges = int(args[i_arg + 1]); i_arg += 2 + elif a == "--label" and i_arg + 1 < len(args): + project_label = args[i_arg + 1]; i_arg += 2 + elif a in ("-h", "--help"): + print("Usage: graphify tree [--graph PATH] [--output HTML]") + print(" --graph PATH path to graph.json (default graphify-out/graph.json)") + print(" --output HTML output path (default graphify-out/GRAPH_TREE.html)") + print(" --root PATH filesystem root (default: longest common dir of all source_files)") + print(" --max-children N cap visible children per node (default 200)") + print(" --top-k-edges N pre-compute top-K outbound edges per symbol (default 12)") + print(" --label NAME project label shown in the page header") + return + else: + i_arg += 1 + if not graph_path.is_file(): + print(f"error: graph.json not found at {graph_path}", file=sys.stderr) + sys.exit(1) + _enforce_graph_size_cap_or_exit(graph_path) + if output_path is None: + output_path = graph_path.parent / "GRAPH_TREE.html" + out = write_tree_html( + graph_path=graph_path, output_path=output_path, + root=root, max_children=max_children, + top_k_edges=top_k_edges, project_label=project_label, + ) + size_kb = out.stat().st_size / 1024 + print(f"wrote {out} ({size_kb:.1f} KB)") + print(f"open with: xdg-open {out} (or file://{out.resolve()})") + sys.exit(0) + + elif cmd == "merge-driver": + # git merge driver for graph.json — takes (base, current, other) and writes + # the union of current+other nodes/edges back to current. Exits 1 on + # corrupt input so git surfaces the conflict instead of silently + # accepting a poisoned merge (see F-005). + # Usage: graphify merge-driver %O %A %B (set in .git/config merge driver) + if len(sys.argv) < 5: + print("Usage: graphify merge-driver ", file=sys.stderr) + sys.exit(1) + _base_path, _current_path, _other_path = sys.argv[2], sys.argv[3], sys.argv[4] + # Hard caps so a malicious or corrupted graph.json cannot exhaust memory + # at parse time. 50 MB / 100k nodes are well above any realistic graph + # (typical graphs are <5 MB / <50k nodes); anything larger should fail + # the merge so a human can investigate. + _MERGE_MAX_BYTES = 50 * 1024 * 1024 + _MERGE_MAX_NODES = 100_000 + import networkx as _nx + from networkx.readwrite import json_graph as _jg + def _load_graph(p: str): + path_obj = Path(p) + try: + size = path_obj.stat().st_size + except OSError as exc: + raise RuntimeError(f"cannot stat {p}: {exc}") from exc + if size > _MERGE_MAX_BYTES: + raise RuntimeError( + f"graph.json {p} is {size} bytes, exceeds {_MERGE_MAX_BYTES}-byte cap" + ) + data = json.loads(path_obj.read_text(encoding="utf-8")) + try: + return _jg.node_link_graph(data, edges="links"), data + except TypeError: + return _jg.node_link_graph(data), data + try: + G_cur, _ = _load_graph(_current_path) + G_oth, _ = _load_graph(_other_path) + except Exception as exc: + print(f"[graphify merge-driver] error loading graphs: {exc}", file=sys.stderr) + sys.exit(1) # surface the conflict so git doesn't accept a corrupt merge + merged = _nx.compose(G_cur, G_oth) + if merged.number_of_nodes() > _MERGE_MAX_NODES: + print( + f"[graphify merge-driver] merged graph has {merged.number_of_nodes()} nodes, " + f"exceeds {_MERGE_MAX_NODES}-node cap; aborting merge.", + file=sys.stderr, + ) + sys.exit(1) + try: + out_data = _jg.node_link_data(merged, edges="links") + except TypeError: + out_data = _jg.node_link_data(merged) + Path(_current_path).write_text(json.dumps(out_data, indent=2), encoding="utf-8") + sys.exit(0) + + elif cmd == "merge-graphs": + # graphify merge-graphs graph1.json graph2.json ... --out merged.json + args = sys.argv[2:] + graph_paths: list[Path] = [] + out_path = Path(_GRAPHIFY_OUT) / "merged-graph.json" + i = 0 + while i < len(args): + if args[i] == "--out" and i + 1 < len(args): + out_path = Path(args[i + 1]) + i += 2 + else: + graph_paths.append(Path(args[i])) + i += 1 + if len(graph_paths) < 2: + print( + "Usage: graphify merge-graphs [...] [--out merged.json]", + file=sys.stderr, + ) + sys.exit(1) + import networkx as _nx + from networkx.readwrite import json_graph as _jg + from graphify.build import prefix_graph_for_global as _prefix + graphs = [] + for gp in graph_paths: + if not gp.exists(): + print(f"error: not found: {gp}", file=sys.stderr) + sys.exit(1) + _enforce_graph_size_cap_or_exit(gp) + data = json.loads(gp.read_text(encoding="utf-8")) + # Normalize edges/links key before loading — graphify writes "links" + # via node_link_data but older runs may have used "edges" (#738). + if "links" not in data and "edges" in data: + data = dict(data, links=data["edges"]) + try: + G = _jg.node_link_graph(data, edges="links") + except TypeError: + G = _jg.node_link_graph(data) + graphs.append(G) + # nx.compose requires all graphs to be the same type. When input graphs + # come from different sources (e.g. an AST-only run vs a full LLM run) one + # may be a MultiGraph and another a Graph. Normalise everything to Graph + # (the graphify default) by converting MultiGraphs with nx.Graph(). + def _to_simple(g: "_nx.Graph") -> "_nx.Graph": + # nx.compose requires every graph to be the same type. Inputs may + # disagree on BOTH axes — directed vs undirected, and multi vs simple + # — because per-repo graph.json files are written by different extract + # paths at different times. Normalise everything to a plain undirected + # Graph (the merged cross-repo view is undirected anyway), which covers + # DiGraph / MultiGraph / MultiDiGraph. Without this a directed input + # crashed compose with "All graphs must be directed or undirected" (#1606). + if type(g) is not _nx.Graph: + return _nx.Graph(g) + return g + merged = _nx.Graph() + for G, gp in zip(graphs, graph_paths): + repo_tag = gp.parent.parent.name # graphify-out/../ → repo dir name + prefixed = _to_simple(_prefix(G, repo_tag)) + merged = _nx.compose(merged, prefixed) + try: + out_data = _jg.node_link_data(merged, edges="links") + except TypeError: + out_data = _jg.node_link_data(merged) + out_path.parent.mkdir(parents=True, exist_ok=True) + out_path.write_text(json.dumps(out_data, indent=2), encoding="utf-8") + print(f"Merged {len(graphs)} graphs -> {merged.number_of_nodes()} nodes, {merged.number_of_edges()} edges") + print(f"Written to: {out_path}") + + elif cmd == "clone": + if len(sys.argv) < 3: + print( + "Usage: graphify clone [--branch ] [--out ]", + file=sys.stderr, + ) + sys.exit(1) + url = sys.argv[2] + branch: str | None = None + out_dir: Path | None = None + args = sys.argv[3:] + i = 0 + while i < len(args): + if args[i] == "--branch" and i + 1 < len(args): + branch = args[i + 1] + i += 2 + elif args[i] == "--out" and i + 1 < len(args): + out_dir = Path(args[i + 1]) + i += 2 + else: + i += 1 + local_path = _clone_repo(url, branch=branch, out_dir=out_dir) + print(local_path) + + elif cmd == "export": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + if subcmd not in ("html", "callflow-html", "obsidian", "wiki", "svg", "graphml", "neo4j", "falkordb"): + print("Usage: graphify export ", file=sys.stderr) + print(" html [--graph PATH] [--labels PATH] [--node-limit N] [--no-viz]", file=sys.stderr) + print(" callflow-html [GRAPH|DIR] [--graph PATH] [--labels PATH] [--report PATH] [--sections PATH] [--output HTML]", file=sys.stderr) + print(" [--lang auto|zh-CN|en] [--max-sections N] [--diagram-scale N]", file=sys.stderr) + print(" obsidian [--graph PATH] [--labels PATH] [--dir PATH]", file=sys.stderr) + print(" wiki [--graph PATH] [--labels PATH]", file=sys.stderr) + print(" svg [--graph PATH] [--labels PATH]", file=sys.stderr) + print(" graphml [--graph PATH]", file=sys.stderr) + print(" neo4j [--graph PATH] [--push URI] [--user U] [--password P]", file=sys.stderr) + print(" (or set NEO4J_PASSWORD instead of --password to keep it off argv)", file=sys.stderr) + print(" falkordb [--graph PATH] [--push URI] [--user U] [--password P]", file=sys.stderr) + print(" (or set FALKORDB_PASSWORD instead of --password to keep it off argv)", file=sys.stderr) + sys.exit(1) + + # Parse shared args + args = sys.argv[3:] + graph_path = Path(_GRAPHIFY_OUT) / "graph.json" + graph_path_explicit = False + labels_path = Path(_GRAPHIFY_OUT) / ".graphify_labels.json" + labels_path_explicit = False + report_path = Path(_GRAPHIFY_OUT) / "GRAPH_REPORT.md" + report_path_explicit = False + sections_path: Path | None = None + callflow_output: Path | None = None + callflow_lang = "auto" + callflow_max_sections = 15 + callflow_diagram_scale = 1.0 + callflow_max_diagram_nodes = 18 + callflow_max_diagram_edges = 24 + analysis_path = Path(_GRAPHIFY_OUT) / ".graphify_analysis.json" + node_limit = 5000 + no_viz = False + obsidian_dir = Path(_GRAPHIFY_OUT) / "obsidian" + # Shared push-connection settings for the graph-database sinks (neo4j, + # falkordb), parsed from the generic --push/--user/--password flags below. + push_uri: str | None = None + push_user = "neo4j" # Neo4j default user; FalkorDB auth is optional and ignores it + # F-031: prefer an env var so the password never appears on argv (visible + # in `ps` output / shell history). The explicit --password flag still + # overrides it. Each sink reads its own var: FALKORDB_PASSWORD for falkordb, + # NEO4J_PASSWORD otherwise. + push_password: str | None = ( + os.environ.get("FALKORDB_PASSWORD") if subcmd == "falkordb" + else os.environ.get("NEO4J_PASSWORD") + ) or None + i = 0 + while i < len(args): + a = args[i] + if a == "--graph" and i + 1 < len(args): + graph_path = Path(args[i + 1]) + graph_path_explicit = True + i += 2 + elif a == "--labels" and i + 1 < len(args): + labels_path = Path(args[i + 1]) + labels_path_explicit = True + i += 2 + elif a == "--report" and i + 1 < len(args): + report_path = Path(args[i + 1]) + report_path_explicit = True + i += 2 + elif a == "--sections" and i + 1 < len(args): + sections_path = Path(args[i + 1]); i += 2 + elif a == "--output" and i + 1 < len(args): + callflow_output = Path(args[i + 1]).expanduser() + if not callflow_output.is_absolute(): + callflow_output = Path.cwd() / callflow_output + i += 2 + elif a == "--lang" and i + 1 < len(args): + callflow_lang = args[i + 1]; i += 2 + elif a == "--max-sections" and i + 1 < len(args): + callflow_max_sections = int(args[i + 1]); i += 2 + elif a == "--diagram-scale" and i + 1 < len(args): + callflow_diagram_scale = float(args[i + 1]); i += 2 + elif a == "--max-diagram-nodes" and i + 1 < len(args): + callflow_max_diagram_nodes = int(args[i + 1]); i += 2 + elif a == "--max-diagram-edges" and i + 1 < len(args): + callflow_max_diagram_edges = int(args[i + 1]); i += 2 + elif a in ("-h", "--help") and subcmd == "callflow-html": + print("Usage: graphify export callflow-html [GRAPH|DIR] [--graph PATH] [--labels PATH]") + print(" --report PATH path to GRAPH_REPORT.md") + print(" --sections PATH JSON section definitions") + print(" --output HTML output path (default graphify-out/-callflow.html)") + print(" --lang LANG auto, zh-CN, en, etc. (default auto)") + print(" --max-sections N maximum auto-derived sections (default 15)") + print(" --diagram-scale N Mermaid diagram scale (default 1.0)") + print(" --max-diagram-nodes N representative nodes per section (default 18)") + print(" --max-diagram-edges N representative edges per section (default 24)") + sys.exit(0) + elif a == "--node-limit" and i + 1 < len(args): + node_limit = int(args[i + 1]); i += 2 + elif a == "--no-viz": + no_viz = True; i += 1 + elif a == "--dir" and i + 1 < len(args): + obsidian_dir = Path(args[i + 1]); i += 2 + elif a == "--push" and i + 1 < len(args): + push_uri = args[i + 1]; i += 2 + elif a == "--user" and i + 1 < len(args): + push_user = args[i + 1]; i += 2 + elif a == "--password" and i + 1 < len(args): + push_password = args[i + 1]; i += 2 + elif subcmd == "callflow-html" and not a.startswith("-") and not graph_path_explicit: + candidate = Path(a) + if candidate.name == "graph.json" or candidate.suffix.lower() == ".json": + graph_path = candidate + elif (candidate / "graph.json").exists(): + graph_path = candidate / "graph.json" + else: + graph_path = candidate / _GRAPHIFY_OUT / "graph.json" + graph_path_explicit = True + i += 1 + else: + i += 1 + + graph_path = graph_path.expanduser() + if graph_path_explicit: + graph_out_dir = graph_path.parent + if not labels_path_explicit: + labels_path = graph_out_dir / ".graphify_labels.json" + if not report_path_explicit: + report_path = graph_out_dir / "GRAPH_REPORT.md" + labels_path = labels_path.expanduser() + report_path = report_path.expanduser() + + if not graph_path.exists(): + print(f"error: graph not found: {graph_path}. Run /graphify first.", file=sys.stderr) + sys.exit(1) + + if subcmd == "callflow-html": + from graphify.callflow_html import write_callflow_html as _write_callflow_html + out = _write_callflow_html( + graph=graph_path, + report=report_path, + labels=labels_path, + sections=sections_path, + output=callflow_output, + lang=callflow_lang, + max_sections=callflow_max_sections, + diagram_scale=callflow_diagram_scale, + max_diagram_nodes=callflow_max_diagram_nodes, + max_diagram_edges=callflow_max_diagram_edges, + verbose=True, + ) + print(f"callflow HTML written - open in any browser: {out}") + sys.exit(0) + + from networkx.readwrite import json_graph as _jg + from graphify.build import build_from_json as _bfj + from graphify.security import check_graph_file_size_cap as _check_cap + + # Solution 3 (#1019): for the HTML view, an oversized graph.json should + # not be a hard error. Detect the over-cap condition here and fall back + # to the community-aggregation view (node_limit=5000) below instead of + # exiting 1. All other subcommands keep the hard cap. + _over_cap = False + try: + _check_cap(graph_path) + except ValueError as _cap_err: + if subcmd == "html": + _over_cap = True + try: + _over_cap_bytes = graph_path.stat().st_size + except OSError: + _over_cap_bytes = -1 + print( + f"warning: graph.json exceeds cap ({_over_cap_bytes} bytes); " + f"falling back to community-aggregation view (node_limit=5000)", + file=sys.stderr, + ) + else: + print(f"error: {_cap_err}", file=sys.stderr) + sys.exit(1) + _raw = json.loads(graph_path.read_text(encoding="utf-8")) + if "links" not in _raw and "edges" in _raw: + _raw = dict(_raw, links=_raw["edges"]) + try: + G = _jg.node_link_graph(_raw, edges="links") + except TypeError: + G = _jg.node_link_graph(_raw) + + # Load optional analysis/labels + communities: dict[int, list[str]] = {} + if analysis_path.exists(): + _an = json.loads(analysis_path.read_text(encoding="utf-8")) + communities = {int(k): v for k, v in _an.get("communities", {}).items()} + cohesion: dict[int, float] = {int(k): v for k, v in _an.get("cohesion", {}).items()} + gods_data = _an.get("gods", []) + else: + cohesion = {} + gods_data = [] + + # Fallback: graph.json carries the per-node community as a node attribute + # (`to_json` writes it on every node). The analysis sidecar is the + # canonical source — but the post-commit / watch rebuild path doesn't + # regenerate it, and `extract` may have its temp files cleaned up. When + # that happens, `graphify export html` previously bailed with + # "Single community - aggregated view not useful." even though the + # per-node attribute had the right data all along. Reconstruct from + # the graph itself so downstream subcommands (html, obsidian, wiki, + # svg, graphml, neo4j) don't silently produce a degraded artifact. + if not communities: + reconstructed: dict[int, list[str]] = {} + for node_id, data in G.nodes(data=True): + cid_raw = data.get("community") + if cid_raw is None: + continue + try: + cid = int(cid_raw) + except (TypeError, ValueError): + continue + reconstructed.setdefault(cid, []).append(str(node_id)) + if reconstructed: + communities = reconstructed + + labels: dict[int, str] = {} + if labels_path.exists(): + labels = {int(k): v for k, v in json.loads(labels_path.read_text(encoding="utf-8")).items()} + + out_dir = graph_path.parent + + if subcmd == "html": + from graphify.export import to_html as _to_html + if no_viz: + html_target = out_dir / "graph.html" + if html_target.exists(): + html_target.unlink() + print("--no-viz: skipped graph.html") + else: + # Over-cap fallback (#1019): force the community-aggregation + # path so the oversized graph still renders a usable artifact. + _effective_node_limit = 5000 if _over_cap else node_limit + _to_html(G, communities, str(out_dir / "graph.html"), + community_labels=labels or None, node_limit=_effective_node_limit) + if G.number_of_nodes() <= _effective_node_limit: + print(f"graph.html written - open in any browser, no server needed") + if _over_cap: + sys.exit(0) + + elif subcmd == "obsidian": + from graphify.export import to_obsidian as _to_obsidian, to_canvas as _to_canvas + n = _to_obsidian(G, communities, str(obsidian_dir), + community_labels=labels or None, cohesion=cohesion or None) + print(f"Obsidian vault: {n} notes in {obsidian_dir}/") + _to_canvas(G, communities, str(obsidian_dir / "graph.canvas"), + community_labels=labels or None) + print(f"Canvas: {obsidian_dir}/graph.canvas") + print(f"Open {obsidian_dir}/ as a vault in Obsidian.") + + elif subcmd == "wiki": + from graphify.wiki import to_wiki as _to_wiki + from graphify.analyze import god_nodes as _god_nodes + if not communities: + print( + "error: .graphify_analysis.json is missing or empty — refusing to export wiki to prevent data loss.\n" + "Run `graphify extract .` (or `graphify cluster-only .`) to regenerate community data first.", + file=sys.stderr, + ) + sys.exit(1) + if not gods_data: + gods_data = _god_nodes(G) + n = _to_wiki(G, communities, str(out_dir / "wiki"), + community_labels=labels or None, cohesion=cohesion or None, + god_nodes_data=gods_data) + print(f"Wiki: {n} articles written to {out_dir}/wiki/") + print(f" {out_dir}/wiki/index.md -> agent entry point") + + elif subcmd == "svg": + from graphify.export import to_svg as _to_svg + _to_svg(G, communities, str(out_dir / "graph.svg"), + community_labels=labels or None) + print(f"graph.svg written - embeds in Obsidian, Notion, GitHub READMEs") + + elif subcmd == "graphml": + from graphify.export import to_graphml as _to_graphml + _to_graphml(G, communities, str(out_dir / "graph.graphml")) + print(f"graph.graphml written - open in Gephi, yEd, or any GraphML tool") + + elif subcmd == "neo4j": + if push_uri: + from graphify.export import push_to_neo4j as _push + if push_password is None: + print("error: --password required for --push", file=sys.stderr) + sys.exit(1) + result = _push(G, uri=push_uri, user=push_user, + password=push_password, communities=communities) + print(f"Pushed to Neo4j: {result['nodes']} nodes, {result['edges']} edges") + else: + from graphify.export import to_cypher as _to_cypher + _to_cypher(G, str(out_dir / "cypher.txt")) + print(f"cypher.txt written - import with: cypher-shell < {out_dir}/cypher.txt") + + elif subcmd == "falkordb": + if push_uri: + from graphify.export import push_to_falkordb as _push + result = _push(G, uri=push_uri, user=push_user, + password=push_password, communities=communities) + print(f"Pushed to FalkorDB: {result['nodes']} nodes, {result['edges']} edges") + else: + from graphify.export import to_cypher as _to_cypher + _to_cypher(G, str(out_dir / "cypher.txt")) + print(f"cypher.txt written ({out_dir}/cypher.txt) - statements are OpenCypher. " + f"FalkorDB's GRAPH.QUERY runs one statement at a time (no bulk script " + f"import), so load a graph with: graphify export falkordb --push " + f"falkordb://localhost:6379") + + elif cmd == "benchmark": + from graphify.benchmark import run_benchmark, print_benchmark + + graph_path = sys.argv[2] if len(sys.argv) > 2 else _default_graph_path() + _enforce_graph_size_cap_or_exit(Path(graph_path)) + # Try to load corpus_words from detect output + corpus_words = None + detect_path = Path(".graphify_detect.json") + if detect_path.exists(): + try: + detect_data = json.loads(detect_path.read_text(encoding="utf-8")) + corpus_words = detect_data.get("total_words") + except Exception: + pass + result = run_benchmark(graph_path, corpus_words=corpus_words) + print_benchmark(result) + + elif cmd == "global": + subcmd = sys.argv[2] if len(sys.argv) > 2 else "" + from graphify.global_graph import ( + global_add as _global_add, + global_remove as _global_remove, + global_list as _global_list, + global_path as _global_path, + ) + if subcmd == "add": + # graphify global add [--as ] + args = sys.argv[3:] + source = None + tag = None + i = 0 + while i < len(args): + if args[i] == "--as" and i + 1 < len(args): + tag = args[i + 1]; i += 2 + elif not source: + source = Path(args[i]); i += 1 + else: + i += 1 + if not source: + print("Usage: graphify global add [--as ]", file=sys.stderr) + sys.exit(1) + tag = tag or source.parent.parent.name + try: + result = _global_add(source, tag) + if result["skipped"]: + print(f"'{tag}' unchanged since last add - global graph not modified.") + else: + print(f"Added '{tag}' to global graph: +{result['nodes_added']} nodes, " + f"-{result['nodes_removed']} pruned. Global: {_global_path()}") + except Exception as exc: + print(f"error: {exc}", file=sys.stderr); sys.exit(1) + elif subcmd == "remove": + tag = sys.argv[3] if len(sys.argv) > 3 else "" + if not tag: + print("Usage: graphify global remove ", file=sys.stderr); sys.exit(1) + try: + removed = _global_remove(tag) + print(f"Removed '{tag}' from global graph ({removed} nodes pruned).") + except KeyError as exc: + print(f"error: {exc}", file=sys.stderr); sys.exit(1) + elif subcmd == "list": + repos = _global_list() + if not repos: + print("Global graph is empty. Use 'graphify global add' to add a project.") + else: + print(f"Global graph: {_global_path()}") + for tag, info in repos.items(): + print(f" {tag}: {info.get('node_count', '?')} nodes, added {info.get('added_at', '?')[:10]}") + elif subcmd == "path": + print(_global_path()) + else: + print("Usage: graphify global [add|remove|list|path]", file=sys.stderr); sys.exit(1) + + elif cmd == "extract": + # Headless full-pipeline extraction for CI / scripts (#698). + # Runs detect -> AST extraction on code -> semantic LLM extraction on + # docs/papers/images -> merge -> build -> cluster -> write outputs. + # Unlike the skill.md path (which runs through Claude Code subagents), + # this calls extract_corpus_parallel directly using whichever backend + # has an API key set. + if len(sys.argv) < 3: + print( + "Usage: graphify extract [--backend gemini|kimi|claude|openai|deepseek|ollama] " + "[--model M] [--mode deep] [--out DIR] [--google-workspace] [--no-cluster] " + "[--max-workers N] [--token-budget N] [--max-concurrency N] " + "[--api-timeout S] [--postgres DSN] [--cargo] " + "[--cpp-deep] [--compile-db PATH] [--cpp-workers N] [--timing]", + file=sys.stderr, + ) + sys.exit(1) + + has_path = True + if sys.argv[2].startswith("-"): + has_path = False + target = Path(".").resolve() + else: + target = Path(sys.argv[2]).resolve() + if not target.exists(): + print(f"error: path not found: {target}", file=sys.stderr) + sys.exit(1) + + backend: str | None = None + model: str | None = None + extract_mode: str | None = None + out_dir: Path | None = None + cli_postgres_dsn: str | None = None + cli_cargo: bool = False + cli_cpp_deep: bool = False + cli_cpp_workers: int | None = None + cli_compile_db: str | None = None + _cpp_deep_fmt_stashed = None # set within if cli_cpp_deep, read below + no_cluster = False + dedup_llm = False + google_workspace = False + global_merge = False + global_repo_tag: str | None = None + # Performance/tuning knobs (issue #792). None means "use library default". + cli_max_workers: int | None = None + cli_token_budget: int | None = None + cli_max_concurrency: int | None = None + cli_api_timeout: float | None = None + # Clustering tuning knobs + cli_resolution: float = 1.0 + cli_exclude_hubs: float | None = None + cli_excludes: list[str] = [] + cli_timing: bool = False + + def _parse_int(name: str, raw: str) -> int: + try: + v = int(raw) + except ValueError: + print(f"error: {name} must be a positive integer (got {raw!r})", file=sys.stderr) + sys.exit(2) + if v <= 0: + print(f"error: {name} must be > 0 (got {v})", file=sys.stderr) + sys.exit(2) + return v + + def _parse_float(name: str, raw: str) -> float: + try: + v = float(raw) + except ValueError: + print(f"error: {name} must be a positive number (got {raw!r})", file=sys.stderr) + sys.exit(2) + if v <= 0: + print(f"error: {name} must be > 0 (got {v})", file=sys.stderr) + sys.exit(2) + return v + + args = sys.argv[3:] if has_path else sys.argv[2:] + i = 0 + while i < len(args): + a = args[i] + if a == "--backend" and i + 1 < len(args): + backend = args[i + 1]; i += 2 + elif a.startswith("--backend="): + backend = a.split("=", 1)[1]; i += 1 + elif a == "--model" and i + 1 < len(args): + model = args[i + 1]; i += 2 + elif a.startswith("--model="): + model = a.split("=", 1)[1]; i += 1 + elif a == "--mode" and i + 1 < len(args): + extract_mode = args[i + 1]; i += 2 + elif a.startswith("--mode="): + extract_mode = a.split("=", 1)[1]; i += 1 + elif a == "--out" and i + 1 < len(args): + out_dir = Path(args[i + 1]); i += 2 + elif a.startswith("--out="): + out_dir = Path(a.split("=", 1)[1]); i += 1 + elif a == "--no-cluster": + no_cluster = True; i += 1 + elif a == "--dedup-llm": + dedup_llm = True; i += 1 + elif a == "--google-workspace": + google_workspace = True; i += 1 + elif a == "--global": + global_merge = True; i += 1 + elif a == "--as" and i + 1 < len(args): + global_repo_tag = args[i + 1]; i += 2 + elif a == "--max-workers" and i + 1 < len(args): + cli_max_workers = _parse_int("--max-workers", args[i + 1]); i += 2 + elif a.startswith("--max-workers="): + cli_max_workers = _parse_int("--max-workers", a.split("=", 1)[1]); i += 1 + elif a == "--token-budget" and i + 1 < len(args): + cli_token_budget = _parse_int("--token-budget", args[i + 1]); i += 2 + elif a.startswith("--token-budget="): + cli_token_budget = _parse_int("--token-budget", a.split("=", 1)[1]); i += 1 + elif a == "--max-concurrency" and i + 1 < len(args): + cli_max_concurrency = _parse_int("--max-concurrency", args[i + 1]); i += 2 + elif a.startswith("--max-concurrency="): + cli_max_concurrency = _parse_int("--max-concurrency", a.split("=", 1)[1]); i += 1 + elif a == "--api-timeout" and i + 1 < len(args): + cli_api_timeout = _parse_float("--api-timeout", args[i + 1]); i += 2 + elif a.startswith("--api-timeout="): + cli_api_timeout = _parse_float("--api-timeout", a.split("=", 1)[1]); i += 1 + elif a == "--resolution" and i + 1 < len(args): + cli_resolution = _parse_float("--resolution", args[i + 1]); i += 2 + elif a.startswith("--resolution="): + cli_resolution = _parse_float("--resolution", a.split("=", 1)[1]); i += 1 + elif a == "--exclude-hubs" and i + 1 < len(args): + cli_exclude_hubs = float(args[i + 1]); i += 2 + elif a.startswith("--exclude-hubs="): + cli_exclude_hubs = float(a.split("=", 1)[1]); i += 1 + elif a == "--exclude" and i + 1 < len(args): + cli_excludes.append(args[i + 1]); i += 2 + elif a.startswith("--exclude="): + cli_excludes.append(a.split("=", 1)[1]); i += 1 + elif a == "--postgres" and i + 1 < len(args): + cli_postgres_dsn = args[i + 1]; i += 2 + elif a.startswith("--postgres="): + cli_postgres_dsn = a.split("=", 1)[1]; i += 1 + elif a == "--cargo": + cli_cargo = True + i += 1 + elif a == "--cpp-deep": + cli_cpp_deep = True; i += 1 + elif a == "--cpp-workers" and i + 1 < len(args): + cli_cpp_workers = _parse_int("--cpp-workers", args[i + 1]); i += 2 + elif a.startswith("--cpp-workers="): + cli_cpp_workers = _parse_int("--cpp-workers", a.split("=", 1)[1]); i += 1 + elif a == "--compile-db" and i + 1 < len(args): + cli_compile_db = args[i + 1]; i += 2 + elif a.startswith("--compile-db="): + cli_compile_db = a.split("=", 1)[1]; i += 1 + elif a == "--timing": + cli_timing = True; i += 1 + else: + i += 1 + + if not has_path and cli_postgres_dsn is None: + print("error: must specify a path to scan or a --postgres DSN", file=sys.stderr) + sys.exit(1) + + _VALID_MODES = {"deep"} + if extract_mode is not None and extract_mode not in _VALID_MODES: + print( + f"error: unknown --mode '{extract_mode}'. " + f"Available: {', '.join(sorted(_VALID_MODES))}", + file=sys.stderr, + ) + sys.exit(2) + deep_mode = extract_mode == "deep" + if deep_mode: + print("[graphify extract] deep mode enabled: richer semantic extraction") + + # Opt-in deep C/C++ pass (libclang). Surface the flag to extract() via env + # (the resolver gate reads GRAPHIFY_CPP_DEEP) and hand it the compile DB + # location if given. Availability is probed inside extract(); a missing + # libclang degrades to the tree-sitter baseline with a one-line tip. + if cli_cpp_deep: + _cpp_deep_fmt_stashed = None # set below if libclang is available + os.environ["GRAPHIFY_CPP_DEEP"] = "1" + if cli_compile_db: + os.environ["GRAPHIFY_CPP_COMPILE_DB"] = cli_compile_db + if cli_cpp_workers is not None and cli_cpp_workers > 0: + os.environ["GRAPHIFY_CPP_WORKERS"] = str(cli_cpp_workers) + from graphify.cpp_deep import libclang_status as _cpp_status + _ok, _detail = _cpp_status() + if _ok: + print(f"[graphify extract] cpp-deep enabled ({_detail})") + # Logger setup (stderr only — the file handler is added below + # once the output directory is known). + _fmt = logging.Formatter("[graphify cpp-deep] %(message)s") + _log = logging.getLogger("graphify.cpp_deep") + _log.setLevel(logging.INFO) + _log.propagate = False + _sh = logging.StreamHandler(sys.stderr) + _sh.setFormatter(_fmt) + _log.addHandler(_sh) + # Stashed for the file-handler setup below. + _cpp_deep_fmt_stashed = _fmt + else: + print( + f"[graphify extract] cpp-deep requested but libclang unavailable " + f"({_detail}); using tree-sitter for C/C++. " + f"Install with: pip install 'graphifyy[cpp]'" + ) + + # CLI flag wins over env var. Setting GRAPHIFY_API_TIMEOUT here so + # _call_openai_compat picks it up without needing a new kwarg path. + if cli_api_timeout is not None: + os.environ["GRAPHIFY_API_TIMEOUT"] = str(cli_api_timeout) + if cli_max_workers is not None: + os.environ["GRAPHIFY_MAX_WORKERS"] = str(cli_max_workers) + + # Resolve output dir. The user-facing contract is "/graphify-out/" + # so a fresh checkout writes graphify-out/ at the project root, matching + # the skill.md pipeline. + out_root = (out_dir.resolve() if out_dir else target) + graphify_out = out_root / _GRAPHIFY_OUT + graphify_out.mkdir(parents=True, exist_ok=True) + + # Attach a file handler to the cpp-deep logger now that the output dir + # is known. This is the handler that lets a `tail -f` monitor progress + # even when stderr is captured (e.g. inside a host process). + if _cpp_deep_fmt_stashed is not None: + try: + _fh = logging.FileHandler( + str(graphify_out / "cpp_deep.log"), encoding="utf-8") + _fh.setFormatter(_cpp_deep_fmt_stashed) + if hasattr(_fh.stream, "reconfigure"): + _fh.stream.reconfigure(line_buffering=True) + logging.getLogger("graphify.cpp_deep").addHandler(_fh) + print(f"[graphify extract] Progress log: {graphify_out / 'cpp_deep.log'}") + except Exception: + import traceback + _warn = logging.getLogger(__name__) + _warn.warning("Failed to create cpp-deep progress log: %s", traceback.format_exc()) + + stages = _StageTimer(cli_timing) + + from graphify.detect import ( + detect as _detect, + detect_incremental as _detect_incremental, + save_manifest as _save_manifest, + ) + manifest_path = graphify_out / "manifest.json" + existing_graph_path = graphify_out / "graph.json" + incremental_mode = manifest_path.exists() and existing_graph_path.exists() if has_path else False + + if not has_path: + code_files = [] + doc_files = [] + paper_files = [] + image_files = [] + deleted_files = [] + unchanged_total = 0 + files_by_type = {} + elif incremental_mode: + print(f"[graphify extract] incremental scan of {target}") + detection = _detect_incremental( + target, + manifest_path=str(manifest_path), + google_workspace=google_workspace or None, + extra_excludes=cli_excludes or None, + ) + files_by_type = detection.get("files", {}) + new_by_type = detection.get("new_files", {}) + code_files = [Path(p) for p in new_by_type.get("code", [])] + doc_files = [Path(p) for p in new_by_type.get("document", [])] + paper_files = [Path(p) for p in new_by_type.get("paper", [])] + image_files = [Path(p) for p in new_by_type.get("image", [])] + deleted_files = list(detection.get("deleted_files", [])) + unchanged_total = sum(len(v) for v in detection.get("unchanged_files", {}).values()) + else: + print(f"[graphify extract] scanning {target}") + detection = _detect(target, google_workspace=google_workspace or None, extra_excludes=cli_excludes or None) + files_by_type = detection.get("files", {}) + code_files = [Path(p) for p in files_by_type.get("code", [])] + doc_files = [Path(p) for p in files_by_type.get("document", [])] + paper_files = [Path(p) for p in files_by_type.get("paper", [])] + image_files = [Path(p) for p in files_by_type.get("image", [])] + deleted_files = [] + unchanged_total = 0 + + semantic_files = doc_files + paper_files + image_files + if incremental_mode: + print( + f"[graphify extract] {len(code_files)} code, {len(doc_files)} docs, " + f"{len(paper_files)} papers, {len(image_files)} images changed; " + f"{unchanged_total} unchanged; {len(deleted_files)} deleted" + ) + else: + print( + f"[graphify extract] found {len(code_files)} code, " + f"{len(doc_files)} docs, {len(paper_files)} papers, " + f"{len(image_files)} images" + ) + stages.mark("detect") + + # Resolve the LLM backend only now that we know whether the corpus + # needs one. A code-only corpus is pure local AST and must not require + # an API key; the key is enforced below only when there's LLM work. + from graphify.llm import ( + BACKENDS as _BACKENDS, + detect_backend as _detect_backend, + estimate_cost as _estimate_cost, + extract_corpus_parallel as _extract_corpus_parallel, + _format_backend_env_keys, + _get_backend_api_key, + ) + needs_llm = bool(semantic_files) or dedup_llm + if backend is None and needs_llm: + backend = _detect_backend() + if backend is not None and backend not in _BACKENDS: + print( + f"error: unknown backend '{backend}'. " + f"Available: {', '.join(sorted(_BACKENDS))}", + file=sys.stderr, + ) + sys.exit(1) + if needs_llm: + if backend is None: + reasons = [] + if semantic_files: + reasons.append( + f"{len(semantic_files)} doc/paper/image file(s) need semantic extraction" + ) + if dedup_llm: + reasons.append("--dedup-llm was passed") + print( + "error: no LLM API key found (" + "; ".join(reasons) + "). " + "Set GEMINI_API_KEY or GOOGLE_API_KEY (gemini), MOONSHOT_API_KEY " + "(kimi), ANTHROPIC_API_KEY (claude), OPENAI_API_KEY (openai), " + "DEEPSEEK_API_KEY (deepseek), or pass --backend. A code-only " + "corpus needs no key.", + file=sys.stderr, + ) + sys.exit(1) + if backend == "ollama": + from graphify.llm import _validate_ollama_base_url + _oll_url = os.environ.get("OLLAMA_BASE_URL", _BACKENDS["ollama"].get("base_url", "")) + try: + _validate_ollama_base_url(_oll_url, warn=False) + except ValueError as exc: + print(f"error: {exc}", file=sys.stderr) + sys.exit(2) + if not _get_backend_api_key(backend): + allow_no_key = False + if backend == "ollama": + from urllib.parse import urlparse + ollama_url = os.environ.get( + "OLLAMA_BASE_URL", + _BACKENDS["ollama"].get("base_url", ""), + ) + try: + host = (urlparse(ollama_url).hostname or "").lower() + except Exception: + host = "" + allow_no_key = ( + host in ("localhost", "127.0.0.1", "::1") + or host.startswith("127.") + ) + elif backend == "bedrock": + allow_no_key = bool( + os.environ.get("AWS_PROFILE") + or os.environ.get("AWS_REGION") + or os.environ.get("AWS_DEFAULT_REGION") + or os.environ.get("AWS_ACCESS_KEY_ID") + ) + elif backend == "claude-cli": + import shutil as _shutil + allow_no_key = _shutil.which("claude") is not None + if not allow_no_key: + print( + "error: backend 'claude-cli' requires the `claude` CLI on $PATH " + "(install Claude Code and run `claude` once to authenticate).", + file=sys.stderr, + ) + sys.exit(1) + if not allow_no_key: + print( + f"error: backend '{backend}' requires {_format_backend_env_keys(backend)} to be set.", + file=sys.stderr, + ) + sys.exit(1) + + # AST extraction on code files. Empty code list (docs-only corpus) is + # the issue #698 case — skip cleanly instead of crashing inside extract(). + ast_result: dict = {"nodes": [], "edges": [], "input_tokens": 0, "output_tokens": 0} + if code_files: + from graphify.extract import extract as _ast_extract + # Anchor the cache at the output root, not the scanned project: + # with --out, a /graphify-out/cache/ would leak a + # graphify-out/ dir into a project that asked for external output. + ast_kwargs: dict = {"cache_root": out_root} + if cli_max_workers is not None: + ast_kwargs["max_workers"] = cli_max_workers + print(f"[graphify extract] AST extraction on {len(code_files)} code files...") + try: + ast_result = _ast_extract(code_files, **ast_kwargs) + except Exception as exc: + print(f"[graphify extract] AST extraction failed: {exc}", file=sys.stderr) + ast_result = {"nodes": [], "edges": [], "input_tokens": 0, "output_tokens": 0} + stages.mark("AST extract") + + # Semantic extraction on docs/papers/images. Check cache first. + from graphify.cache import ( + check_semantic_cache as _check_semantic_cache, + prune_semantic_cache as _prune_semantic_cache, + save_semantic_cache as _save_semantic_cache, + ) + sem_result: dict = { + "nodes": [], "edges": [], "hyperedges": [], + "input_tokens": 0, "output_tokens": 0, + } + sem_cache_hits = 0 + sem_cache_misses = 0 + if semantic_files: + sem_paths_str = [str(p) for p in semantic_files] + cached_nodes, cached_edges, cached_hyperedges, uncached_paths = ( + _check_semantic_cache(sem_paths_str, root=out_root) + ) + sem_cache_hits = len(semantic_files) - len(uncached_paths) + sem_cache_misses = len(uncached_paths) + sem_result["nodes"].extend(cached_nodes) + sem_result["edges"].extend(cached_edges) + sem_result["hyperedges"].extend(cached_hyperedges) + if sem_cache_hits: + print(f"[graphify extract] semantic cache: {sem_cache_hits} hit / {sem_cache_misses} miss") + + if uncached_paths: + print(f"[graphify extract] semantic extraction on {len(uncached_paths)} files via {backend}...") + corpus_kwargs: dict = { + "backend": backend, + "model": model, + "root": target, + } + if deep_mode: + corpus_kwargs["deep_mode"] = True + if cli_token_budget is not None: + corpus_kwargs["token_budget"] = cli_token_budget + if cli_max_concurrency is not None: + corpus_kwargs["max_concurrency"] = cli_max_concurrency + + # Minimal progress callback so the CLI is no longer silent + # during long local-inference runs (issue #792 addendum). + # Also track per-chunk success so we can fail loudly when + # every chunk errors (e.g. missing backend SDK package). + _chunk_stats = {"total": 0, "succeeded": 0} + def _progress(idx: int, total: int, _result: dict) -> None: + _chunk_stats["total"] = total + _chunk_stats["succeeded"] += 1 + print( + f"[graphify extract] chunk {idx + 1}/{total} done", + flush=True, + ) + corpus_kwargs["on_chunk_done"] = _progress + + try: + fresh = _extract_corpus_parallel( + [Path(p) for p in uncached_paths], + **corpus_kwargs, + ) + except ImportError as exc: + print(f"error: {exc}", file=sys.stderr) + sys.exit(1) + except Exception as exc: + print( + f"[graphify extract] semantic extraction failed: {exc}", + file=sys.stderr, + ) + fresh = {"nodes": [], "edges": [], "hyperedges": [], "input_tokens": 0, "output_tokens": 0} + + # on_chunk_done only fires after a chunk succeeds. If fresh + # semantic extraction was requested and no chunks completed, + # fail instead of writing an AST-only graph with exit 0. + if uncached_paths and _chunk_stats["succeeded"] == 0: + print( + f"[graphify extract] error: all semantic chunks failed " + f"for backend '{backend}' ({len(uncached_paths)} uncached files) - " + f"see per-chunk errors above. If you see 'requires the X package', " + f"run `pip install X` and retry.", + file=sys.stderr, + ) + sys.exit(1) + try: + _save_semantic_cache( + fresh.get("nodes", []), + fresh.get("edges", []), + fresh.get("hyperedges", []), + root=out_root, + ) + except Exception as exc: + print(f"[graphify extract] warning: could not write semantic cache: {exc}", file=sys.stderr) + sem_result["nodes"].extend(fresh.get("nodes", [])) + sem_result["edges"].extend(fresh.get("edges", [])) + sem_result["hyperedges"].extend(fresh.get("hyperedges", [])) + sem_result["input_tokens"] += fresh.get("input_tokens", 0) + sem_result["output_tokens"] += fresh.get("output_tokens", 0) + + # Prune orphaned semantic cache entries. The semantic cache is + # content-hash-keyed and unversioned, so it is never swept by the AST + # version-cleanup: every content change or file deletion leaves a + # permanent orphan that accumulates unbounded (#1527). Sweep it against + # the FULL live document set (``files_by_type`` — present in both the + # incremental and full branches), NOT the incremental ``semantic_files`` + # changed-subset, which would delete every unchanged doc's valid entry. + # Best-effort: a prune failure must never break extraction. + try: + from graphify.cache import file_hash as _file_hash + _live_hashes: set[str] = set() + for _kind in ("document", "paper", "image"): + for _fp in files_by_type.get(_kind, []): + _abs = Path(_fp) + if not _abs.is_absolute(): + _abs = Path(out_root) / _abs + if not _abs.is_file(): + continue # deleted/missing — leave out so its entry is pruned + try: + _live_hashes.add(_file_hash(_abs, out_root)) + except OSError: + pass + _prune_semantic_cache(out_root, _live_hashes) + except Exception as exc: + print(f"[graphify extract] warning: could not prune semantic cache: {exc}", file=sys.stderr) + stages.mark("semantic extract") + + pg_result: dict = {"nodes": [], "edges": []} + if cli_postgres_dsn is not None: + from graphify.pg_introspect import introspect_postgres + print(f"[graphify extract] introspecting PostgreSQL schema...") + try: + pg_result = introspect_postgres(cli_postgres_dsn) + except (ConnectionError, ImportError) as exc: + print(f"error: {exc}", file=sys.stderr) + sys.exit(1) + print(f"[graphify extract] PostgreSQL: {len(pg_result['nodes'])} nodes, " + f"{len(pg_result['edges'])} edges") + + cargo_result: dict = {"nodes": [], "edges": []} + if cli_cargo: + from graphify.cargo_introspect import introspect_cargo + print("[graphify extract] introspecting Cargo workspace...") + try: + cargo_result = introspect_cargo(target) + except (ConnectionError, ImportError, OSError) as exc: + print(f"error: {exc}", file=sys.stderr) + sys.exit(1) + print(f"[graphify extract] Cargo: {len(cargo_result['nodes'])} nodes, " + f"{len(cargo_result['edges'])} edges") + + # Merge AST + semantic + pg_result + cargo_result. Order matters for deduplication: passing AST + # first means semantic node attributes win on collision (richer labels + # for symbols also referenced in docs). Hyperedges only come from the + # semantic side. + merged: dict = { + "nodes": list(ast_result.get("nodes", [])) + list(sem_result.get("nodes", [])) + list(pg_result.get("nodes", [])) + list(cargo_result.get("nodes", [])), + "edges": list(ast_result.get("edges", [])) + list(sem_result.get("edges", [])) + list(pg_result.get("edges", [])) + list(cargo_result.get("edges", [])), + "hyperedges": list(sem_result.get("hyperedges", [])), + "input_tokens": ast_result.get("input_tokens", 0) + sem_result.get("input_tokens", 0), + "output_tokens": ast_result.get("output_tokens", 0) + sem_result.get("output_tokens", 0), + } + + graph_json_path = graphify_out / "graph.json" + analysis_path = graphify_out / ".graphify_analysis.json" + + # Build a manifest-safe files dict: only stamp semantic_hash for files + # that actually produced output (cache hit or fresh extraction). Files + # whose chunk failed have no source_file entry in sem_result — leaving + # their semantic_hash empty so detect_incremental re-queues them (#933). + _sem_extracted: set[str] = { + n.get("source_file", "") for n in sem_result.get("nodes", []) + } | { + e.get("source_file", "") for e in sem_result.get("edges", []) + } + _sem_extracted.discard("") + _sem_types = {"document", "paper", "image"} + _manifest_files = { + ftype: [f for f in flist if ftype not in _sem_types or f in _sem_extracted] + for ftype, flist in files_by_type.items() + } + + if no_cluster: + # --no-cluster: dump the raw merged extraction as graph.json. + # No NetworkX, no community detection, no analysis sidecar. + # Dedupe nodes (by id) and parallel edges so the raw output matches the + # clustered path (whose DiGraph collapses both) and stays deterministic + # across modes (#1317; node dedup also collapses shared Swift module + # anchors emitted per importing file, #1327). + from graphify.build import dedupe_edges as _dedupe_edges, dedupe_nodes as _dedupe_nodes + from graphify.export import backup_if_protected as _backup + if ( + incremental_mode + and not code_files + and not semantic_files + and not deleted_files + and not pg_result.get("nodes") + and not pg_result.get("edges") + and not cargo_result.get("nodes") + and not cargo_result.get("edges") + ): + print( + "[graphify extract] no incremental changes detected " + "(--no-cluster); outputs left untouched." + ) + try: + _save_manifest(_manifest_files, manifest_path=str(manifest_path), kind="both", root=target) + except Exception as exc: + print(f"[graphify extract] warning: could not write manifest: {exc}", file=sys.stderr) + stages.total() + sys.exit(0) + + merged["nodes"] = _dedupe_nodes(merged["nodes"]) + merged["edges"] = _dedupe_edges(merged["edges"]) + # Backfill source_file from endpoint nodes — this raw path bypasses + # build_from_json's backfill, and semantic edges sometimes omit it (#1279). + _node_sf = {n.get("id"): n.get("source_file") for n in merged["nodes"]} + for _e in merged["edges"]: + if not _e.get("source_file"): + _e["source_file"] = ( + _node_sf.get(_e.get("source")) or _node_sf.get(_e.get("target")) or "" + ) + _backup(graphify_out) + graph_json_path.write_text( + json.dumps(merged, indent=2), encoding="utf-8" + ) + stages.mark("write") + cost = _estimate_cost( + backend, merged["input_tokens"], merged["output_tokens"] + ) + print( + f"[graphify extract] wrote {graph_json_path} — " + f"{len(merged['nodes'])} nodes, {len(merged['edges'])} edges " + f"(no clustering)" + ) + if merged["input_tokens"] or merged["output_tokens"]: + print( + f"[graphify extract] tokens: " + f"{merged['input_tokens']:,} in / " + f"{merged['output_tokens']:,} out, " + f"est. cost: ${cost:.4f}" + ) + try: + _save_manifest(_manifest_files, manifest_path=str(manifest_path), kind="both", root=target) + except Exception as exc: + print(f"[graphify extract] warning: could not write manifest: {exc}", file=sys.stderr) + if global_merge: + from graphify.global_graph import global_add as _global_add + _tag = global_repo_tag or target.name + try: + result = _global_add(graphify_out / "graph.json", _tag) + if result["skipped"]: + print(f"[graphify global] '{_tag}' unchanged since last add - skipped.") + else: + print(f"[graphify global] '{_tag}' merged into global graph " + f"(+{result['nodes_added']} nodes, -{result['nodes_removed']} pruned).") + except Exception as exc: + print(f"[graphify global] warning: failed to merge into global graph: {exc}", file=sys.stderr) + stages.total() + sys.exit(0) + + # Build graph + cluster + score + write. + from graphify.build import ( + build as _build, + build_from_json as _build_from_json, + build_merge as _build_merge, + ) + from graphify.cluster import cluster as _cluster, score_all as _score_all + from graphify.export import to_json as _to_json + from graphify.analyze import god_nodes as _god_nodes, surprising_connections as _surprising + dedup_backend = backend if dedup_llm else None + if incremental_mode: + G = _build_merge( + [merged], + graph_path=existing_graph_path, + prune_sources=deleted_files or None, + dedup=True, + dedup_llm_backend=dedup_backend, + root=target, + ) + else: + G = _build([merged], dedup=True, dedup_llm_backend=dedup_backend, root=target) + stages.mark("build") + if G.number_of_nodes() == 0: + print( + "[graphify extract] graph is empty — extraction produced no nodes. " + "Possible causes: all files skipped, binary-only corpus, or LLM " + "returned no edges.", + file=sys.stderr, + ) + sys.exit(1) + + communities = _cluster(G, resolution=cli_resolution, exclude_hubs_percentile=cli_exclude_hubs) + stages.mark("cluster") + cohesion = _score_all(G, communities) + try: + gods = _god_nodes(G) + except Exception: + gods = [] + try: + surprises = _surprising(G, communities) + except Exception: + surprises = [] + stages.mark("analyze") + + from graphify.export import backup_if_protected as _backup + _backup(graphify_out) + _to_json(G, communities, str(graph_json_path), force=True) + stages.mark("export") + if merged.get("output_tokens", 0) > 0: + (graphify_out / ".graphify_semantic_marker").write_text( + json.dumps({"output_tokens": merged["output_tokens"]}), encoding="utf-8" + ) + if global_merge: + from graphify.global_graph import global_add as _global_add + _tag = global_repo_tag or target.name + try: + result = _global_add(graphify_out / "graph.json", _tag) + if result["skipped"]: + print(f"[graphify global] '{_tag}' unchanged since last add - skipped.") + else: + print(f"[graphify global] '{_tag}' merged into global graph " + f"(+{result['nodes_added']} nodes, -{result['nodes_removed']} pruned).") + except Exception as exc: + print(f"[graphify global] warning: failed to merge into global graph: {exc}", file=sys.stderr) + analysis = { + "communities": {str(k): v for k, v in communities.items()}, + "cohesion": {str(k): v for k, v in cohesion.items()}, + "gods": gods, + "surprises": surprises, + "tokens": { + "input": merged["input_tokens"], + "output": merged["output_tokens"], + }, + } + analysis_path.write_text(json.dumps(analysis, indent=2), encoding="utf-8") + try: + _save_manifest(_manifest_files, manifest_path=str(manifest_path), kind="both", root=target) + except Exception as exc: + print(f"[graphify extract] warning: could not write manifest: {exc}", file=sys.stderr) + + cost = _estimate_cost(backend, merged["input_tokens"], merged["output_tokens"]) + print( + f"[graphify extract] wrote {graph_json_path}: " + f"{G.number_of_nodes()} nodes, {G.number_of_edges()} edges, " + f"{len(communities)} communities" + ) + print(f"[graphify extract] wrote {analysis_path}") + if incremental_mode: + print( + f"[graphify extract] incremental summary: " + f"{sem_cache_hits + unchanged_total} files cached/unchanged, " + f"{len(code_files) + sem_cache_misses} re-extracted, " + f"{len(deleted_files)} deleted" + ) + elif sem_cache_hits: + print(f"[graphify extract] semantic cache: {sem_cache_hits} cached, {sem_cache_misses} re-extracted") + if merged["input_tokens"] or merged["output_tokens"]: + print( + f"[graphify extract] tokens: " + f"{merged['input_tokens']:,} in / " + f"{merged['output_tokens']:,} out, " + f"est. cost (~{backend}): ${cost:.4f}" + ) + # extract intentionally stops at graph.json + analysis; the report and + # community labels are produced by `cluster-only` (or an agent's Step 5). + # Point standalone users at it so communities get named (#1097). + print( + "[graphify extract] next: run " + f"`graphify cluster-only {graphify_out.parent}` " + "to generate GRAPH_REPORT.md and name communities" + ) + stages.total() + + elif cmd == "cache-check": + # graphify cache-check [--root ] + # Reads file paths (one per line) from , checks semantic cache. + # Writes: + # graphify-out/.graphify_cached.json — already-cached nodes/edges/hyperedges + # graphify-out/.graphify_uncached.txt — paths that need extraction + # Stdout: "Cache: N hit, M miss" + from graphify.cache import check_semantic_cache + if len(sys.argv) < 3: + print("Usage: graphify cache-check [--root ]", file=sys.stderr) + sys.exit(1) + files_from = Path(sys.argv[2]) + root = Path(".") + i = 3 + while i < len(sys.argv): + if sys.argv[i] == "--root" and i + 1 < len(sys.argv): + root = Path(sys.argv[i + 1]) + i += 2 + else: + i += 1 + files = [f for f in files_from.read_text(encoding="utf-8").splitlines() if f.strip()] + cached_nodes, cached_edges, cached_hyperedges, uncached = check_semantic_cache(files, root) + out = root / _GRAPHIFY_OUT + out.mkdir(parents=True, exist_ok=True) + if cached_nodes or cached_edges or cached_hyperedges: + (out / ".graphify_cached.json").write_text( + json.dumps({"nodes": cached_nodes, "edges": cached_edges, "hyperedges": cached_hyperedges}, + ensure_ascii=False), + encoding="utf-8", + ) + (out / ".graphify_uncached.txt").write_text("\n".join(uncached), encoding="utf-8") + print(f"Cache: {len(files) - len(uncached)} hit, {len(uncached)} miss") + + elif cmd == "merge-chunks": + # graphify merge-chunks --out + # Concatenates .graphify_chunk_*.json files written by semantic subagents. + # Deduplicates nodes by id (first writer wins). Sums token counts. + import glob as _glob + if len(sys.argv) < 3: + print("Usage: graphify merge-chunks --out ", file=sys.stderr) + sys.exit(1) + out_path: Path | None = None + chunk_args: list[str] = [] + i = 2 + while i < len(sys.argv): + if sys.argv[i] == "--out" and i + 1 < len(sys.argv): + out_path = Path(sys.argv[i + 1]) + i += 2 + else: + chunk_args.append(sys.argv[i]) + i += 1 + if not out_path: + print("error: --out required", file=sys.stderr) + sys.exit(1) + chunk_files: list[str] = [] + for arg in chunk_args: + expanded = _glob.glob(arg) + chunk_files.extend(sorted(expanded) if expanded else [arg]) + merged: dict = {"nodes": [], "edges": [], "hyperedges": [], "input_tokens": 0, "output_tokens": 0} + seen_ids: set[str] = set() + for cf in chunk_files: + try: + chunk = json.loads(Path(cf).read_text(encoding="utf-8")) + except (json.JSONDecodeError, OSError) as exc: + print(f"[graphify merge-chunks] warning: skipping {cf}: {exc}", file=sys.stderr) + continue + for n in chunk.get("nodes", []): + if n.get("id") not in seen_ids: + seen_ids.add(n["id"]) + merged["nodes"].append(n) + merged["edges"].extend(chunk.get("edges", [])) + merged["hyperedges"].extend(chunk.get("hyperedges", [])) + merged["input_tokens"] += chunk.get("input_tokens", 0) + merged["output_tokens"] += chunk.get("output_tokens", 0) + out_path.parent.mkdir(parents=True, exist_ok=True) + out_path.write_text(json.dumps(merged, ensure_ascii=False), encoding="utf-8") + print( + f"Merged {len(chunk_files)} chunks: {len(merged['nodes'])} nodes, {len(merged['edges'])} edges, " + f"{merged['input_tokens']:,} in / {merged['output_tokens']:,} out tokens" + ) + + elif cmd == "merge-semantic": + # graphify merge-semantic --cached --new --out + # Merges cached semantic results with freshly-extracted chunk results. + # Deduplicates nodes by id (cached entries take priority over new ones). + if len(sys.argv) < 3: + print("Usage: graphify merge-semantic --cached --new --out ", file=sys.stderr) + sys.exit(1) + cached_path: Path | None = None + new_path: Path | None = None + out_path2: Path | None = None + i = 2 + while i < len(sys.argv): + if sys.argv[i] == "--cached" and i + 1 < len(sys.argv): + cached_path = Path(sys.argv[i + 1]); i += 2 + elif sys.argv[i] == "--new" and i + 1 < len(sys.argv): + new_path = Path(sys.argv[i + 1]); i += 2 + elif sys.argv[i] == "--out" and i + 1 < len(sys.argv): + out_path2 = Path(sys.argv[i + 1]); i += 2 + else: + i += 1 + if not out_path2: + print("error: --out required", file=sys.stderr) + sys.exit(1) + empty: dict = {"nodes": [], "edges": [], "hyperedges": []} + cached_data = json.loads(cached_path.read_text(encoding="utf-8")) if cached_path and cached_path.exists() else empty + new_data = json.loads(new_path.read_text(encoding="utf-8")) if new_path and new_path.exists() else empty + seen_ids2: set[str] = set() + all_nodes: list[dict] = [] + for n in cached_data.get("nodes", []) + new_data.get("nodes", []): + if n.get("id") not in seen_ids2: + seen_ids2.add(n["id"]) + all_nodes.append(n) + merged2 = { + "nodes": all_nodes, + "edges": cached_data.get("edges", []) + new_data.get("edges", []), + "hyperedges": cached_data.get("hyperedges", []) + new_data.get("hyperedges", []), + } + out_path2.parent.mkdir(parents=True, exist_ok=True) + out_path2.write_text(json.dumps(merged2, ensure_ascii=False), encoding="utf-8") + print(f"Merged: {len(merged2['nodes'])} nodes, {len(merged2['edges'])} edges") + + elif cmd == "optimize": + # graphify optimize --directed [--mode deep] [--out DIR] + # Post-process an existing graphify-out/graph.json without re-extraction. + # --directed set directed=true in graph.json (networkx reads either) + # --mode deep NOT SUPPORTED on existing graph — needs source re-extract + directed_opt = False + deep_opt = False + out_dir_opt: Path | None = None + i = 2 + while i < len(sys.argv): + a = sys.argv[i] + if a == "--directed": + directed_opt = True; i += 1 + elif a in ("--mode", "--mode=deep") or a.startswith("--mode="): + v = a.split("=", 1)[1] if "=" in a else (sys.argv[i + 1] if i + 1 < len(sys.argv) else "") + if v == "deep": + deep_opt = True + i += 2 if "=" not in a else 1 + elif a == "--out" and i + 1 < len(sys.argv): + out_dir_opt = Path(sys.argv[i + 1]); i += 2 + elif a.startswith("--out="): + out_dir_opt = Path(a.split("=", 1)[1]); i += 1 + else: + i += 1 + + if not directed_opt and not deep_opt: + print("Usage: graphify optimize --directed [--mode deep] [--out DIR]", file=sys.stderr) + print(" --directed set directed=true in existing graph.json", file=sys.stderr) + print(" --mode deep (needs source files — use 'graphify extract . --mode deep' instead)", file=sys.stderr) + sys.exit(1) + + from pathlib import Path as _P + gf_out = out_dir_opt if out_dir_opt is not None else _P("graphify-out") + graph_json = gf_out / "graph.json" + if not graph_json.exists(): + print(f"error: {graph_json} not found. Run 'graphify extract' first.", file=sys.stderr) + sys.exit(1) + + # ── --directed: toggle the flag in networkx node-link JSON ────────── + if directed_opt: + g_data = json.loads(graph_json.read_text(encoding="utf-8")) + links = g_data.get("links", g_data.get("edges", [])) + print(f"[graphify optimize] Setting directed=true ({len(g_data.get('nodes',[])):,} nodes, {len(links):,} edges)") + g_data["directed"] = True + graph_json.write_text(json.dumps(g_data, ensure_ascii=False), encoding="utf-8") + print(f"[graphify optimize] Done — graph.json now directed") + + # ── --mode deep: requires source files, not supported on graph.json ── + if deep_opt: + print("[graphify optimize] --mode deep requires re-running extraction on source files.", file=sys.stderr) + print("Use: graphify extract . --mode deep --cpp-deep", file=sys.stderr) + print("(This will re-run AST + semantic extraction on the full corpus)", file=sys.stderr) + + elif Path(cmd).exists() or cmd in (".", "..") or cmd.startswith(("./", "../", "/", "~")): + # User ran `graphify ` directly — treat as `graphify extract `. + # Common when following the PowerShell note in README (`graphify .`) or + # copy-pasting skill invocations without the leading slash. + sys.argv.insert(2, sys.argv[1]) + sys.argv[1] = "extract" + main() + else: + print(f"error: unknown command '{cmd}'", file=sys.stderr) + print("Run 'graphify --help' for usage.", file=sys.stderr) + sys.exit(1) if __name__ == "__main__": diff --git a/graphify/cpp_deep/__init__.py b/graphify/cpp_deep/__init__.py new file mode 100644 index 000000000..7491aa39a --- /dev/null +++ b/graphify/cpp_deep/__init__.py @@ -0,0 +1,105 @@ +"""Opt-in deep C/C++ semantic enhancement for Graphify. + +Graphify's default C/C++ path is tree-sitter only (``graphify.extract._extract_generic`` +with ``_C_CONFIG`` / ``_CPP_CONFIG``). tree-sitter is fast and dependency-free but +structurally shallow for C/C++: it never expands the preprocessor, so + + * function-like macro call sites resolve to nothing (the macro is not a node), + * macro-generated function definitions are invisible, + * ``#ifdef`` keeps *both* arms, producing duplicate/contradictory nodes, and + * cross-translation-unit / cross-DLL calls are matched only by bare name. + +This package adds a libclang-backed pass that parses the *fully preprocessed* AST +(macros instantiated, one active ``#ifdef`` arm) and uses Clang USRs (Unified Symbol +Resolutions) as translation-unit-independent symbol identities. It plugs into the +existing ``graphify.resolver_registry`` seam — the same mechanism the built-in +per-language member-call resolvers use — so nothing in ``extract()``'s body changes +and the whole feature is inert unless explicitly enabled. + +Activation is deliberately conservative and two-gated: + + 1. The caller opts in (``--cpp-deep`` flag, surfaced as ``GRAPHIFY_CPP_DEEP=1``). + 2. libclang is importable AND a usable ``libclang`` shared library is found. + +If either gate is closed the resolver never registers and the corpus falls back to +the tree-sitter baseline unchanged. Any failure inside the pass is caught and logged; +it never aborts a build. + +See ``docs/cpp-deep-enhancement-design.md`` for the full design. +""" +from __future__ import annotations + +import logging +import os + +_LOG = logging.getLogger(__name__) + +# Suffixes this enhancement claims. Mirrors the C/C++ suffixes the tree-sitter +# dispatch already routes to extract_c / extract_cpp (``.h`` is shared with ObjC +# but libclang parses it fine as C/C++; ObjC-specific handling stays on its own +# resolver). CUDA/Metal are included because clang can parse them given flags. +CPP_SUFFIXES = frozenset({ + ".c", ".h", ".cpp", ".cc", ".cxx", ".hpp", ".hh", ".hxx", ".cu", ".cuh", +}) + +# Env var the CLI sets when --cpp-deep is passed. Kept as an env toggle (not a +# function argument) so the resolver — which the registry calls with a fixed +# (per_file, all_nodes, all_edges) signature — can read it without threading a +# flag through extract()'s body. +_ENABLE_ENV = "GRAPHIFY_CPP_DEEP" + +_registered = False + + +def is_enabled() -> bool: + """True when the caller opted into the deep C/C++ pass via --cpp-deep.""" + return os.environ.get(_ENABLE_ENV, "").strip().lower() in ("1", "true", "yes", "on") + + +def libclang_status() -> tuple[bool, str]: + """Probe for a usable libclang without importing heavy machinery eagerly. + + Returns ``(available, detail)``. ``detail`` is a human-readable string suitable + for a one-line tip when unavailable (never raises). + """ + try: + from graphify.cpp_deep.clang_index import probe_libclang + except Exception as exc: # pragma: no cover - import wiring only + return False, f"cpp_deep import failed: {exc}" + return probe_libclang() + + +def maybe_register() -> bool: + """Register the deep C/C++ resolver iff opted-in and libclang is usable. + + Idempotent. Returns True if the resolver is (now or already) registered. Safe to + call unconditionally at extract-time; it self-gates and never raises. + """ + global _registered + if _registered: + return True + if not is_enabled(): + return False + + available, detail = libclang_status() + if not available: + _LOG.warning( + "cpp-deep requested but libclang is unavailable (%s); " + "falling back to tree-sitter for C/C++. Install with: " + "pip install 'graphifyy[cpp]'", + detail, + ) + return False + + try: + from graphify.resolver_registry import LanguageResolver + from graphify.resolver_registry import register as _register + from graphify.cpp_deep.resolver import resolve_cpp_deep + + _register(LanguageResolver("cpp_deep", CPP_SUFFIXES, resolve_cpp_deep)) + _registered = True + _LOG.info("cpp-deep resolver registered (%s)", detail) + return True + except Exception as exc: # pragma: no cover - defensive + _LOG.warning("cpp-deep registration failed, skipping: %s", exc) + return False diff --git a/graphify/cpp_deep/cache.py b/graphify/cpp_deep/cache.py new file mode 100644 index 000000000..7ccd1666c --- /dev/null +++ b/graphify/cpp_deep/cache.py @@ -0,0 +1,131 @@ +"""Per-TU disk cache for libclang parse results. + +Each translation unit's extracted symbols/calls/macros are serialised to a +compact JSON file in the cache directory so that: + + * Memory peaks at ~1 TU's records rather than all TUs at once. + * A crash mid-parse only loses the un-cached TUs (not the entire run). + * Future ``--update`` runs can skip TUs whose source + compile flags are + unchanged (hash-based key). + +Cache files are named ``.json`` where *hash* is the hex digest of +``(resolved_abs_path, flags_string)`` so the same file compiled with the same +flags always maps to the same cache entry. + +The cache directory lives at ``/clang_cache/`` and is emptied on +a clean build (successful completion of the whole pass). +""" +from __future__ import annotations + +import hashlib +import json +import logging +import shutil +from pathlib import Path + +from graphify.cpp_deep.records import TUResult + +_LOG = logging.getLogger(__name__) + + +def cache_dir() -> Path | None: + """Return the cache directory, creating it if necessary. + + Reads ``GRAPHIFY_OUT`` (the ``--out`` flag) then falls back to + ``graphify-out/`` under CWD, mirroring the status-file convention. + """ + import os + env = os.environ.get("GRAPHIFY_OUT", "").strip() + base = Path(env) if env else Path("graphify-out") + cd = base / "clang_cache" + cd.mkdir(parents=True, exist_ok=True) + return cd + + +def cache_key(tu_path: str, args: list[str]) -> str: + """Stable hex digest of the TU identity so same file + flags → same cache.""" + h = hashlib.sha256() + h.update(Path(tu_path).resolve().as_posix().encode()) + h.update(b"\x00") + # Preserve flag order — some flags like -include are order-dependent. + h.update(" ".join(args).encode()) + h.update(b"\x00") + # Include source-file content so editing the file invalidates the cache. + try: + h.update(Path(tu_path).read_bytes()) + except Exception: + pass + return h.hexdigest()[:16] + + +def cache_path(tu_path: str, args: list[str]) -> Path | None: + """Full path of the cache file for this TU, or None if no cache dir.""" + cd = cache_dir() + if cd is None: + return None + return cd / f"{cache_key(tu_path, args)}.json" + + +def write(result: TUResult, args: list[str]) -> None: + """Serialise a TUResult to its cache file. Fails silently (best-effort).""" + cp = cache_path(result.path, args) + if cp is None: + return + try: + payload = json.dumps(result.to_dict(), ensure_ascii=False, separators=(",", ":")) + # Atomic: write .tmp then rename so no half-written file survives a crash. + tmp = cp.with_suffix(".tmp") + tmp.write_text(payload, encoding="utf-8") + tmp.replace(cp) + except Exception: + _LOG.warning("cpp-deep cache: failed to write %s", Path(result.path).name) + + +def load(tu_path: str, args: list[str]) -> TUResult | None: + """Load a cached TUResult, or None if the file is absent/corrupt.""" + cp = cache_path(tu_path, args) + if cp is None or not cp.is_file(): + return None + try: + payload = json.loads(cp.read_text(encoding="utf-8")) + return TUResult.from_dict(payload) + except Exception: + return None + + +def exists(tu_path: str, args: list[str]) -> bool: + """True if this TU has a valid cache file.""" + try: + cp = cache_path(tu_path, args) + return cp is not None and cp.is_file() + except Exception: + return False + + +# FIXME: this function is currently unused (dead code). If resurrected, convert +# to a generator (yield instead of accumulating) so memory stays bounded as the +# docstring claims. Otherwise delete. +def load_all(unit_paths: list[str], args: list[str]) -> list[TUResult]: + """Stream-load all cached TUResults in batch order. + + Each file is loaded and immediately yielded so memory holds at most one + deserialised TUResult at a time. + """ + results: list[TUResult] = [] + for path in unit_paths: + r = load(path, args) + if r is not None: + results.append(r) + else: + _LOG.warning("cpp-deep cache: missing entry for %s", Path(path).name) + return results + + +def clear() -> None: + """Remove the entire cache directory (called after a successful merge).""" + cd = cache_dir() + if cd is not None and cd.exists(): + try: + shutil.rmtree(cd, ignore_errors=True) + except Exception: + pass diff --git a/graphify/cpp_deep/clang_index.py b/graphify/cpp_deep/clang_index.py new file mode 100644 index 000000000..d9a22f134 --- /dev/null +++ b/graphify/cpp_deep/clang_index.py @@ -0,0 +1,488 @@ +"""libclang wrapper: locate the shared library, parse translation units, and walk +Clang cursors into Graphify-shaped symbol/call/macro records. + +Everything here is import-safe even when the ``clang`` Python package or the +``libclang`` shared library is absent: ``probe_libclang`` reports availability and +the parse functions raise a clear ``ClangUnavailable`` that callers already treat as +"skip, fall back to tree-sitter". + +libclang is configured exactly once per process. The Python binding ships in two +shapes on PyPI: + + * ``libclang`` — bundles a prebuilt ``libclang`` shared library AND the ``clang`` + Python package; ``clang.cindex.Config`` finds the bundled library automatically. + * ``clang`` — only the Python binding; needs a system ``libclang.(dll|so|dylib)``. + +We support both: try the binding as-is first (covers the ``libclang`` wheel), then +fall back to searching common locations / ``GRAPHIFY_LIBCLANG`` for the ``clang`` +wheel case. +""" +from __future__ import annotations + +import logging +import os +import shutil +from pathlib import Path + +_LOG = logging.getLogger(__name__) + + +class ClangUnavailable(RuntimeError): + """Raised when a parse is attempted but libclang could not be initialized.""" + + +# Process-wide init memo: None = not attempted, True/False = result. +_INIT_OK: bool | None = None +_INIT_DETAIL: str = "not initialized" + +# Env override pointing directly at a libclang shared library file. +_LIB_ENV = "GRAPHIFY_LIBCLANG" + + +def _candidate_library_files() -> list[str]: + """Best-effort list of libclang shared-library paths to try (order matters).""" + cands: list[str] = [] + env = os.environ.get(_LIB_ENV, "").strip() + if env: + cands.append(env) + + # A `clang` / `clang-cl` on PATH usually sits beside or near the shared lib. + for exe in ("clang", "clang-cl", "clangd"): + found = shutil.which(exe) + if found: + base = Path(found).resolve().parent + # bin/ -> ../lib on unix installs; dll sits in bin/ on Windows. + for name in ("libclang.dll", "libclang.so", "libclang.dylib"): + cands.append(str(base / name)) + cands.append(str(base.parent / "lib" / name)) + + # Common fixed locations. + common = [ + r"C:\Program Files\LLVM\bin\libclang.dll", + r"C:\Program Files (x86)\LLVM\bin\libclang.dll", + "/usr/lib/llvm/lib/libclang.so", + "/usr/lib/x86_64-linux-gnu/libclang-1.so", + "/usr/local/opt/llvm/lib/libclang.dylib", + "/Library/Developer/CommandLineTools/usr/lib/libclang.dylib", + ] + cands.extend(common) + # De-dup preserving order, keep only existing files. + seen: set[str] = set() + out: list[str] = [] + for c in cands: + if c and c not in seen and Path(c).is_file(): + seen.add(c) + out.append(c) + return out + + +def _init_clang() -> tuple[bool, str]: + """Initialize clang.cindex once; return (ok, detail). Never raises.""" + global _INIT_OK, _INIT_DETAIL + if _INIT_OK is not None: + return _INIT_OK, _INIT_DETAIL + + try: + from clang import cindex + except Exception as exc: + _INIT_OK, _INIT_DETAIL = False, f"python 'clang' binding not importable: {exc}" + return _INIT_OK, _INIT_DETAIL + + # 1) Try the binding as shipped (the `libclang` wheel bundles the library and + # resolves it here without any explicit path). + try: + idx = cindex.Index.create() + del idx + _INIT_OK, _INIT_DETAIL = True, "libclang via bundled/binding default" + return _INIT_OK, _INIT_DETAIL + except Exception: + pass + + # 2) Fall back to explicitly pointing at a discovered library file. + for lib in _candidate_library_files(): + try: + cindex.Config.set_library_file(lib) + idx = cindex.Index.create() + del idx + _INIT_OK, _INIT_DETAIL = True, f"libclang at {lib}" + return _INIT_OK, _INIT_DETAIL + except Exception: + # Config is process-global; once set_library_file is called it can't be + # meaningfully reset, but a failed create means this lib was wrong. Keep + # trying — a later successful set overrides the path. + continue + + _INIT_OK, _INIT_DETAIL = False, ( + "no usable libclang shared library found " + f"(set {_LIB_ENV} or `pip install 'graphifyy[cpp]'`)" + ) + return _INIT_OK, _INIT_DETAIL + + +def probe_libclang() -> tuple[bool, str]: + """Public availability probe used by the package gate. Never raises.""" + return _init_clang() + + +def _require_clang(): + ok, detail = _init_clang() + if not ok: + raise ClangUnavailable(detail) + from clang import cindex + return cindex + + +# ── Translation-unit parsing → records ─────────────────────────────────────── + + +def parse_translation_unit(unit, root: Path, stem_of=None, file_id_of=None, + ignore_filter=None) -> "TUResult": + """Parse one TranslationUnit into a TUResult of symbols/calls/macros. + + ``unit`` is a ``compile_db.TranslationUnit``. ``root`` relativizes source paths so + node IDs match the tree-sitter side (which stores repo-relative source_file). + + ``stem_of`` (optional) maps a resolved absolute source path to the exact SYMBOL + STEM the tree-sitter baseline used for that file (``_file_stem(path)`` run through + make_id). Keying symbol ids off this guarantees byte-identical ids regardless of + the baseline's cache_root. ``file_id_of`` similarly maps to the baseline FILE-NODE + id, used as the parent of `contains` edges for newly-added nodes. When either is + absent, ids fall back to a root-relativized derivation. + + ``ignore_filter`` (optional) returns True for paths that should be excluded + (matching .graphifyignore/.gitignore patterns), preventing symbols from + ignored headers being recorded as graph nodes. + """ + cindex = _require_clang() + from graphify.cpp_deep.records import ( + CallRecord, MacroRecord, MacroUse, SymbolRecord, TUResult, + ) + from graphify.cpp_deep.usr_id import canonical_stem + from graphify.ids import make_id + + result = TUResult(path=str(unit.path), module=unit.module) + + # Resolve a source file to (symbol_stem, file_node_id) aligned to the baseline. + def _ids_for(abs_src: str) -> tuple[str, str]: + stem = "" + file_id = "" + rp = None + try: + rp = Path(abs_src).resolve() + except Exception: + rp = None + if stem_of is not None and rp is not None: + try: + stem = stem_of(rp) or "" + except Exception: + stem = "" + if file_id_of is not None and rp is not None: + try: + file_id = file_id_of(rp) or "" + except Exception: + file_id = "" + if not stem: + stem = canonical_stem(_rel(abs_src)) + if not file_id: + file_id = make_id(stem) + return stem, file_id + + def _file_node_id(abs_src: str) -> str: + return _ids_for(abs_src)[1] + + def _file_symbol_id(abs_src: str, name: str) -> str: + return make_id(_ids_for(abs_src)[0], name) + + def _type_id(abs_src: str, qualified_parts: list[str]) -> str: + return make_id(_ids_for(abs_src)[0], *qualified_parts) + + def _member_symbol_id(class_id: str, name: str) -> str: + return make_id(class_id, name) + + + + + index = cindex.Index.create() + # DETAILED_PROCESSING_RECORD is what surfaces macro definitions and + # expansions as cursors; without it the preprocessor is invisible (the exact + # tree-sitter gap we're closing). Function bodies are kept (default) because we + # need the CALL_EXPR sites inside them. + opts = ( + cindex.TranslationUnit.PARSE_DETAILED_PROCESSING_RECORD + | cindex.TranslationUnit.PARSE_INCOMPLETE + ) + try: + tu = index.parse(str(unit.path), args=unit.args, options=opts) + except Exception as exc: + result.diagnostics.append(f"parse failed: {exc}") + return result + + for diag in tu.diagnostics: + # Only keep errors; warnings are noise for our purposes. + if diag.severity >= cindex.Diagnostic.Error: + result.diagnostics.append(str(diag.spelling)) + + tu_file = str(unit.path) + + def _rel(path: str) -> str: + try: + return Path(path).resolve().relative_to(root).as_posix() + except Exception: + return Path(path).as_posix() + + def _in_corpus(cursor) -> bool: + """Only record symbols physically located in a file under the scan root + AND not excluded by .graphifyignore / .gitignore. System headers and + user-ignored generated files must NOT become nodes.""" + loc = cursor.location + if loc is None or loc.file is None: + return False + fpath = Path(loc.file.name) + try: + fpath.resolve().relative_to(root) + except Exception: + return False + if ignore_filter is not None and ignore_filter(fpath): + return False + return True + + # Track enclosing function so calls attribute to the right caller. + _CURSORKIND = cindex.CursorKind + + # (file_basename, start_line, end_line, function_node_id) for correlating + # TU-level MACRO_INSTANTIATION cursors back to the function they appear in. + func_extents: list[tuple[str, int, int, str]] = [] + + def _qualified_parents(cursor) -> list[str]: + """Namespace/class chain above a cursor, outermost first (for type ids).""" + parts: list[str] = [] + p = cursor.semantic_parent + while p is not None and p.kind in ( + _CURSORKIND.NAMESPACE, _CURSORKIND.CLASS_DECL, + _CURSORKIND.STRUCT_DECL, _CURSORKIND.CLASS_TEMPLATE, + ): + if p.spelling: + parts.append(p.spelling) + p = p.semantic_parent + parts.reverse() + return parts + + def _record_symbol(cursor, kind: str, node_id: str, parent_id: str, parent_rel: str): + loc = cursor.location + rec = SymbolRecord( + usr=cursor.get_usr() or "", + node_id=node_id, + label=_symbol_label(cursor, kind), + kind=kind, + source_file=_rel(loc.file.name) if loc and loc.file else "", + line=loc.line if loc else 0, + module=unit.module, + is_definition=cursor.is_definition(), + parent_id=parent_id, + parent_relation=parent_rel, + ) + result.symbols.append(rec) + if rec.usr: + # Prefer a definition's id for the usr mapping; keep first decl otherwise. + if rec.is_definition or rec.usr not in result.usr_to_id: + result.usr_to_id[rec.usr] = node_id + return rec + + def _symbol_label(cursor, kind: str) -> str: + name = cursor.spelling or "?" + if kind in ("function", "method"): + return f"{name}()" + return name + + def _walk(cursor, enclosing_id: str | None): + for child in cursor.get_children(): + k = child.kind + + # Class / struct. + if k in (_CURSORKIND.CLASS_DECL, _CURSORKIND.STRUCT_DECL, + _CURSORKIND.CLASS_TEMPLATE): + if _in_corpus(child) and child.spelling: + src = child.location.file.name + parents = _qualified_parents(child) + nid = _type_id(src, parents + [child.spelling]) + kind = "class" if k != _CURSORKIND.STRUCT_DECL else "struct" + file_id = _file_node_id(src) + _record_symbol(child, kind, nid, file_id, "contains") + _walk(child, nid) + else: + _walk(child, enclosing_id) + continue + + # Namespace: descend, no node (matches tree-sitter which makes no + # namespace node for C++). + if k == _CURSORKIND.NAMESPACE: + _walk(child, enclosing_id) + continue + + # Function / method. + if k in (_CURSORKIND.FUNCTION_DECL, _CURSORKIND.CXX_METHOD, + _CURSORKIND.FUNCTION_TEMPLATE, _CURSORKIND.CONSTRUCTOR, + _CURSORKIND.DESTRUCTOR): + if _in_corpus(child) and child.spelling: + src = child.location.file.name + sem_parent = child.semantic_parent + is_method = ( + sem_parent is not None + and sem_parent.kind in ( + _CURSORKIND.CLASS_DECL, _CURSORKIND.STRUCT_DECL, + _CURSORKIND.CLASS_TEMPLATE, + ) + ) + if is_method and sem_parent.spelling: + parents = _qualified_parents(sem_parent) + [sem_parent.spelling] + # The class may be declared in a header; key the method off + # the class's own declaration file for TU-independent ids. + cls_src = (sem_parent.location.file.name + if sem_parent.location and sem_parent.location.file + else src) + cls_id = _type_id(cls_src, parents) + nid = _member_symbol_id(cls_id, child.spelling) + rec = _record_symbol(child, "method", nid, cls_id, "method") + else: + nid = _file_symbol_id(src, child.spelling) + file_id = _file_node_id(src) + rec = _record_symbol(child, "function", nid, file_id, "contains") + # Walk the body for calls, attributing to this function. + _walk_calls(child, nid) + # Record extent so TU-level macro instantiations can be attributed + # back to this function (MACRO_INSTANTIATION cursors are NOT AST + # children of the function, so a body walk misses them). + ext = child.extent + if ext and ext.start.file and ext.end.file: + func_extents.append(( + Path(ext.start.file.name).resolve().as_posix(), + ext.start.line, ext.end.line, nid, + )) + continue + + # Macro definition. + if k == _CURSORKIND.MACRO_DEFINITION: + if _in_corpus(child) and child.spelling: + src = child.location.file.name + mid = _file_symbol_id(src, child.spelling) + is_fn_like = _macro_is_function_like(child) + result.macros.append(MacroRecord( + node_id=mid, + label=child.spelling, + source_file=_rel(src), + line=child.location.line, + is_function_like=is_fn_like, + module=unit.module, + file_id=_file_node_id(src), + )) + continue + + # Anything else: descend. + _walk(child, enclosing_id) + + def _walk_calls(func_cursor, caller_id: str): + """Collect CALL_EXPR and macro-expansion sites inside a function body. + + Because clang parses the POST-expansion AST, a function-like macro that + expands to a real call (``LOG(x)`` -> ``log_write(x)``) already surfaces the + expanded CALL_EXPR here — that is the macro-expansion win the tree-sitter + baseline can't produce. We ALSO record the MACRO_INSTANTIATION so the graph + keeps the macro provenance (caller -> macro node), not just the expanded call. + """ + for node in func_cursor.walk_preorder(): + if node.kind == _CURSORKIND.CALL_EXPR: + callee = node.referenced + if callee is not None: + result.calls.append(CallRecord( + caller_id=caller_id, + callee_usr=callee.get_usr() or "", + callee_name=callee.spelling or (node.spelling or ""), + source_file=_rel(node.location.file.name) if node.location and node.location.file else "", + line=node.location.line if node.location else 0, + )) + elif node.spelling: + # Unresolved (e.g. through a function pointer) — keep the name. + result.calls.append(CallRecord( + caller_id=caller_id, + callee_name=node.spelling, + source_file=_rel(node.location.file.name) if node.location and node.location.file else "", + line=node.location.line if node.location else 0, + )) + # MACRO_INSTANTIATION cursors are collected at TU level (see below), + # not here — they are not AST children of the function body. + + def _macro_is_function_like(cursor) -> bool: + # Newer bindings expose is_macro_function_like(); older ones don't, so fall + # back to token inspection: a function-like macro has '(' immediately after + # the name with NO whitespace (``FOO(x)`` vs object-like ``FOO (x)``). + is_fn = getattr(cursor, "is_macro_function_like", None) + if callable(is_fn): + try: + return bool(cursor.is_macro_function_like()) + except Exception: + pass + try: + toks = list(cursor.get_tokens()) + if len(toks) >= 2 and toks[1].spelling == "(": + name_tok, paren_tok = toks[0], toks[1] + # Adjacent (no gap) => function-like. + return (name_tok.extent.end.line == paren_tok.extent.start.line + and name_tok.extent.end.column == paren_tok.extent.start.column) + except Exception: + pass + return False + + # ── walk + macro-use pass ──────────────────────────────────────────── + # Walk the main AST first, then do a single TU-level pass for macro + # instantiations (MACRO_INSTANTIATION cursors are NOT children of + # function bodies so the per-function _walk_calls misses them). + try: + _walk(tu.cursor, None) + except Exception as exc: + result.diagnostics.append(f"walk failed: {exc}") + + known_macros = {m.label for m in result.macros} + extents_by_line = sorted(func_extents, key=lambda e: e[2] - e[1]) if func_extents else [] + + if known_macros and extents_by_line: + try: + for node in tu.cursor.walk_preorder(): + if node.kind != _CURSORKIND.MACRO_INSTANTIATION: + continue + loc = node.location + if not (loc and loc.file): + continue + if not node.spelling or node.spelling not in known_macros: + continue + fname = Path(loc.file.name).resolve().as_posix() + line = loc.line + for ext_file, start, end, fid in extents_by_line: + if ext_file == fname and start <= line <= end: + result.macro_uses.append(MacroUse( + caller_id=fid, + macro_name=node.spelling, + source_file=_rel(loc.file.name), + line=line, + )) + break + except Exception as exc: + result.diagnostics.append(f"macro-use pass failed: {exc}") + + # ── dispose the native TranslationUnit (critical for memory) ───────── + # libclang holds the entire preprocessed AST (millions of nodes for + # header-heavy TUs) in native memory via a CXTranslationUnit handle. + # Python's __del__ calls clang_disposeTranslationUnit() eventually, but + # GC of __del__ objects is slow and unreliable under memory pressure. + # Explicit dispose + dereference lets the OS reclaim pages synchronously + # before we move to the next TU. + try: + tu.cursor._tu = None + except Exception: + pass + try: + del tu + except Exception: + pass + + return result + diff --git a/graphify/cpp_deep/compile_db.py b/graphify/cpp_deep/compile_db.py new file mode 100644 index 000000000..227436bcf --- /dev/null +++ b/graphify/cpp_deep/compile_db.py @@ -0,0 +1,381 @@ +"""Obtain compile commands for C/C++ translation units. + +libclang parses a translation unit accurately only when it knows the include search +paths and preprocessor definitions that the real build uses — otherwise macros are +undefined and ``#include`` resolution fails, degrading the deep pass back toward what +tree-sitter already produces. This module produces, per TU, a +``TranslationUnit(path, args, module)`` record where ``args`` are clang command-line +flags and ``module`` is the output binary (DLL/exe/lib) the TU compiles into — the +latter drives cross-DLL call-chain grouping downstream. + +Resolution order (first that yields commands wins): + + 1. An explicit ``compile_commands.json`` (``--compile-db`` / ``GRAPHIFY_CPP_COMPILE_DB``) + or one discovered at the scan root / a ``build/`` subdir. This is the gold path: + it carries the exact flags the real build used. + 2. Visual Studio ``.sln`` / ``.vcxproj`` files: parse ``AdditionalIncludeDirectories``, + ``PreprocessorDefinitions``, language standard and the output target name. + Windows/MSVC projects rarely ship compile_commands.json, so this is the common + path for the user's corpus. + 3. Heuristic fallback: treat every directory containing a header as an include + root and parse with permissive defaults. Lower fidelity but never nothing. + +Everything is best-effort and pure-Python (no toolchain invocation): a project that +can't be understood yields an empty list and the caller falls back to tree-sitter. +""" +from __future__ import annotations + +import json +import logging +import os +import re +import shlex +import xml.etree.ElementTree as ET +from dataclasses import dataclass, field +from pathlib import Path + +_LOG = logging.getLogger(__name__) + +_C_SUFFIXES = {".c"} +_CPP_SUFFIXES = {".cpp", ".cc", ".cxx", ".cu"} +_HEADER_SUFFIXES = {".h", ".hpp", ".hh", ".hxx", ".cuh"} +_ALL_TU_SUFFIXES = _C_SUFFIXES | _CPP_SUFFIXES + +# vcxproj/MSBuild XML namespace (present on most .vcxproj files). +_MSB_NS = "{http://schemas.microsoft.com/developer/msbuild/2003}" + + +@dataclass +class TranslationUnit: + """One compilable unit and the clang flags + owning module to parse it with.""" + + path: Path + args: list[str] = field(default_factory=list) + module: str = "" # output target name (DLL/exe/lib) for cross-DLL grouping + language: str = "c++" # "c" or "c++" + + +# ── Public entry ───────────────────────────────────────────────────────────── + + +def obtain_compile_units( + paths: list[Path], + root: Path, + *, + compile_db: str | None = None, +) -> list[TranslationUnit]: + """Return TranslationUnit records for the C/C++ files in ``paths``. + + ``root`` is the scan root (used for discovery and to relativize). ``compile_db`` + overrides discovery with an explicit compile_commands.json path. Never raises. + """ + tu_paths = {p.resolve() for p in paths if p.suffix.lower() in _ALL_TU_SUFFIXES} + + try: + # 1. Explicit or discovered compile_commands.json. + db_path = _find_compile_db(compile_db, root) + if db_path is not None: + units = _from_compile_commands(db_path, tu_paths) + if units: + _LOG.info("cpp-deep: %d units from %s", len(units), db_path.name) + return units + + # 2. Visual Studio solution/projects. + units = _from_visual_studio(root, tu_paths) + if units: + _LOG.info("cpp-deep: %d units from Visual Studio projects", len(units)) + return units + + # 3. Heuristic fallback. + units = _heuristic_units(tu_paths, root) + _LOG.info("cpp-deep: %d units from heuristic fallback", len(units)) + return units + except Exception as exc: # pragma: no cover - defensive + _LOG.warning("cpp-deep: compile-unit discovery failed: %s", exc) + return _heuristic_units(tu_paths, root) + + +# ── 1. compile_commands.json ───────────────────────────────────────────────── + + +def _find_compile_db(explicit: str | None, root: Path) -> Path | None: + if explicit: + p = Path(explicit) + if p.is_dir(): + p = p / "compile_commands.json" + return p if p.is_file() else None + env = os.environ.get("GRAPHIFY_CPP_COMPILE_DB", "").strip() + if env: + p = Path(env) + if p.is_dir(): + p = p / "compile_commands.json" + if p.is_file(): + return p + for cand in (root / "compile_commands.json", root / "build" / "compile_commands.json"): + if cand.is_file(): + return cand + return None + + +def _from_compile_commands(db_path: Path, tu_paths: set[Path]) -> list[TranslationUnit]: + try: + entries = json.loads(db_path.read_text(encoding="utf-8")) + except Exception as exc: + _LOG.warning("cpp-deep: cannot read %s: %s", db_path, exc) + return [] + + units: list[TranslationUnit] = [] + for entry in entries: + try: + file_field = entry.get("file", "") + directory = Path(entry.get("directory", str(db_path.parent))) + src = (directory / file_field).resolve() if not Path(file_field).is_absolute() else Path(file_field).resolve() + # If we have a specific set of TUs, honor it; else take all. + if tu_paths and src not in tu_paths: + continue + args = _args_from_entry(entry, directory) + lang = "c" if src.suffix.lower() in _C_SUFFIXES else "c++" + units.append(TranslationUnit(path=src, args=args, module="", language=lang)) + except Exception: + continue + return units + + +def _args_from_entry(entry: dict, directory: Path) -> list[str]: + """Extract clang-usable flags (includes, defines, std) from a compile-db entry. + + We keep only flags libclang needs for parsing and drop the compiler executable, + the input/output file, and codegen-only flags that can confuse the parser. + """ + if "arguments" in entry and isinstance(entry["arguments"], list): + raw = [str(a) for a in entry["arguments"][1:]] # drop argv[0] (compiler) + else: + raw = _split_command(str(entry.get("command", "")))[1:] + return _filter_clang_args(raw, directory) + + +def _split_command(command: str) -> list[str]: + # shlex handles POSIX quoting perfectly but Windows command-line quoting + # follows different rules (MSVC runtime parsing). Use shlex on POSIX, + # fall back to the simple regex split on Windows or for malformed input. + import sys + if sys.platform != "win32": + try: + return shlex.split(command.strip()) + except ValueError: + pass + return [tok for tok in re.split(r"\s+", command.strip()) if tok] + + +def _filter_clang_args(raw: list[str], directory: Path) -> list[str]: + out: list[str] = [] + skip_next = False + for i, a in enumerate(raw): + if skip_next: + skip_next = False + continue + # Include dirs: -I, -I , /I, -isystem . + if a in ("-I", "/I", "-isystem", "-iquote"): + if i + 1 < len(raw): + out.extend(["-I", _abspath(raw[i + 1], directory)]) + skip_next = True + continue + if a.startswith("-I") or a.startswith("/I"): + out.append("-I" + _abspath(a[2:], directory)) + continue + if a.startswith("-isystem"): + out.append("-isystem" + _abspath(a[len("-isystem"):], directory)) + continue + # Defines: -D..., /D... + if a.startswith("-D") or a.startswith("/D"): + out.append("-D" + a[2:]) + continue + if a in ("-D", "/D") and i + 1 < len(raw): + out.append("-D" + raw[i + 1]) + skip_next = True + continue + # Language standard. + if a.startswith("-std=") or a.startswith("/std:"): + std = a.split("=", 1)[-1] if "=" in a else a.split(":", 1)[-1] + out.append("-std=" + std.replace("c++", "c++").lower()) + continue + # Force-include (relevant to macro-heavy code). + if a in ("-include", "/FI") and i + 1 < len(raw): + out.extend(["-include", _abspath(raw[i + 1], directory)]) + skip_next = True + continue + return out + + +def _abspath(p: str, directory: Path) -> str: + p = p.strip('"') + pp = Path(p) + return str(pp if pp.is_absolute() else (directory / pp).resolve()) + + +# ── 2. Visual Studio .sln / .vcxproj ───────────────────────────────────────── + + +def _from_visual_studio(root: Path, tu_paths: set[Path]) -> list[TranslationUnit]: + vcxprojs = list(root.rglob("*.vcxproj")) + if not vcxprojs: + return [] + units: list[TranslationUnit] = [] + for proj in vcxprojs: + try: + units.extend(_units_from_vcxproj(proj, tu_paths)) + except Exception as exc: + _LOG.debug("cpp-deep: vcxproj parse failed for %s: %s", proj, exc) + continue + return units + + +def _vcx_text(el) -> str: + return (el.text or "").strip() if el is not None else "" + + +def _units_from_vcxproj(proj: Path, tu_paths: set[Path]) -> list[TranslationUnit]: + tree = ET.parse(proj) + r = tree.getroot() + proj_dir = proj.parent + + # Output target name: or or the project stem. + target = "" + for tag in ("TargetName", "RootNamespace", "ProjectName"): + el = r.find(f".//{_MSB_NS}{tag}") + if _vcx_text(el): + target = _vcx_text(el) + break + if not target: + target = proj.stem + # Substitute the common $(ProjectName)/$(MSBuildProjectName) macro. + target = target.replace("$(ProjectName)", proj.stem).replace( + "$(MSBuildProjectName)", proj.stem + ) + + # Aggregate include dirs + defines across ItemDefinitionGroup/ClCompile blocks. + include_dirs: list[str] = [] + defines: list[str] = [] + std_flag = "" + for cl in r.findall(f".//{_MSB_NS}ClCompile"): + inc = _vcx_text(cl.find(f"{_MSB_NS}AdditionalIncludeDirectories")) + if inc: + include_dirs.extend(_split_msbuild_list(inc, proj_dir)) + dfn = _vcx_text(cl.find(f"{_MSB_NS}PreprocessorDefinitions")) + if dfn: + defines.extend(_split_defines(dfn)) + std = _vcx_text(cl.find(f"{_MSB_NS}LanguageStandard")) + if std: + std_flag = _msvc_std_to_clang(std) + + base_args = ["--driver-mode=cl"] # parse MSVC-style semantics + for d in _dedup(include_dirs): + base_args.append("-I" + d) + for d in _dedup(defines): + base_args.append("-D" + d) + + # Which source files this project compiles (). + proj_sources: list[Path] = [] + for cl in r.findall(f".//{_MSB_NS}ClCompile"): + inc_attr = cl.get("Include") + if not inc_attr: + continue + src = (proj_dir / inc_attr.replace("\\", "/")).resolve() + if src.suffix.lower() in _ALL_TU_SUFFIXES: + proj_sources.append(src) + + units: list[TranslationUnit] = [] + for src in proj_sources: + if tu_paths and src not in tu_paths: + continue + if not src.is_file(): + continue + lang = "c" if src.suffix.lower() in _C_SUFFIXES else "c++" + # The project's LanguageStandard applies to its language; a C source in a + # C++ project must not inherit a -std=c++NN (clang rejects it). Give C files + # a C default when the project std is C++ (or absent). + args = list(base_args) + if std_flag and (std_flag.startswith("-std=c++") == (lang == "c++")): + args.append(std_flag) + elif lang == "c": + args.append("-std=c11") + units.append( + TranslationUnit(path=src, args=args, module=target, language=lang) + ) + return units + + +def _split_msbuild_list(value: str, proj_dir: Path) -> list[str]: + out: list[str] = [] + for item in value.split(";"): + item = item.strip() + if not item or item.startswith("%(") or "$(" in item: + # Skip MSBuild macros we can't resolve ($(SolutionDir) etc.) and the + # %(...) self-reference — heuristic include roots backstop these. + continue + p = Path(item.replace("\\", "/")) + out.append(str(p if p.is_absolute() else (proj_dir / p).resolve())) + return out + + +def _split_defines(value: str) -> list[str]: + out: list[str] = [] + for item in value.split(";"): + item = item.strip() + if not item or item.startswith("%("): + continue + out.append(item) + return out + + +def _msvc_std_to_clang(std: str) -> str: + m = { + "stdcpp14": "-std=c++14", "stdcpp17": "-std=c++17", + "stdcpp20": "-std=c++20", "stdcpplatest": "-std=c++23", + "stdc11": "-std=c11", "stdc17": "-std=c17", + } + return m.get(std.lower(), "") + + +# ── 3. Heuristic fallback ──────────────────────────────────────────────────── + + +def _heuristic_units(tu_paths: set[Path], root: Path) -> list[TranslationUnit]: + """Parse with every header-bearing directory as an include root. + + No build metadata, so macros defined by the build are unknown; still parses + plain code, resolves in-tree includes, and — crucially — lets libclang expand + macros that ARE defined in the sources themselves (the common #define case). + """ + include_roots = _dedup( + str(p.parent.resolve()) + for p in root.rglob("*") + if p.suffix.lower() in _HEADER_SUFFIXES + ) + base_args: list[str] = [] + for d in include_roots: + base_args.append("-I" + d) + + units: list[TranslationUnit] = [] + for src in sorted(tu_paths): + if not src.is_file(): + continue + lang = "c" if src.suffix.lower() in _C_SUFFIXES else "c++" + std = "-std=c11" if lang == "c" else "-std=c++17" + # Module unknown in heuristic mode: fall back to the immediate parent dir + # name so cross-"module" grouping still has a coarse signal. + module = src.parent.name + units.append( + TranslationUnit(path=src, args=[std, *base_args], module=module, language=lang) + ) + return units + + +def _dedup(items) -> list[str]: + seen: set[str] = set() + out: list[str] = [] + for x in items: + if x and x not in seen: + seen.add(x) + out.append(x) + return out diff --git a/graphify/cpp_deep/crossmod.py b/graphify/cpp_deep/crossmod.py new file mode 100644 index 000000000..4fd11864a --- /dev/null +++ b/graphify/cpp_deep/crossmod.py @@ -0,0 +1,116 @@ +"""Cross-module / cross-DLL call resolution by USR, plus module tagging. + +The user's corpus is a set of Visual Studio projects, each compiling into its own +DLL/exe, all with source available. A function in ``AppModule`` that calls a function +defined in ``UtilLib`` is a *cross-DLL* call — the exact chain the tree-sitter +baseline can only guess at by bare name (and drops when the name is ambiguous). + +libclang already resolved every call to its callee's USR (see ``clang_index``), and +USRs are global across translation units. So this stage is a precise join: + + 1. Build ``usr -> node_id`` across ALL TUs (definitions win over declarations), + and ``node_id -> module`` from the symbols. + 2. For each call record with a ``callee_usr`` we can resolve, emit an EXTRACTED + ``calls`` edge caller_id -> callee_id. When the caller's module differs from the + callee's module, tag the edge ``cross_module`` and stamp both module names — + this is what lets a query group / traverse the call chain by DLL. + 3. Calls whose USR is unknown (function pointers, or callee outside the corpus) + fall back to a bare-name lookup against a single unambiguous definition, mirroring + the C/C++ exemption the tree-sitter cross-file pass already grants; ambiguous + names are left for that pass rather than guessed here. + +Returns a list of edge dicts; node module tags are applied to the symbol nodes by +the caller (``merge``) since it owns node creation. +""" +from __future__ import annotations + +from graphify.cpp_deep.records import TUResult + + +def build_symbol_index(results: list[TUResult]) -> tuple[dict[str, str], dict[str, str]]: + """Return (usr -> node_id, node_id -> module) across all TUs. + + Definitions take precedence over declarations for BOTH the USR mapping and the + module tag: a symbol is *declared* in a header that many modules include, but it + is *defined* in exactly one module — that defining module is its true owner and + is what makes a call from another module a genuine cross-DLL edge. A + declaration-only symbol (no definition anywhere in the corpus) falls back to its + declaration module. + """ + usr_to_id: dict[str, str] = {} + id_to_module: dict[str, str] = {} + id_has_def_module: set[str] = set() + # First pass: definitions own the id -> module mapping. + for res in results: + for s in res.symbols: + if s.usr and s.is_definition: + usr_to_id[s.usr] = s.node_id + if s.is_definition and s.module: + id_to_module[s.node_id] = s.module # definition wins, overrides decl + id_has_def_module.add(s.node_id) + # Second pass: fill USR + module ONLY where a definition didn't already claim it. + for res in results: + for s in res.symbols: + if s.usr and s.usr not in usr_to_id: + usr_to_id[s.usr] = s.node_id + if s.module and s.node_id not in id_has_def_module: + id_to_module.setdefault(s.node_id, s.module) + return usr_to_id, id_to_module + + +def resolve_cross_module_calls( + results: list[TUResult], + usr_to_id: dict[str, str], + id_to_module: dict[str, str], +) -> list[dict]: + """Emit EXTRACTED `calls` edges from USR-resolved call records, tagged by module.""" + # Bare-name fallback index: name -> set of definition node_ids. + name_to_ids: dict[str, set[str]] = {} + for res in results: + for s in res.symbols: + if s.kind in ("function", "method") and s.is_definition: + # label is "name()" — strip the parens for name matching. + nm = s.label[:-2] if s.label.endswith("()") else s.label + name_to_ids.setdefault(nm, set()).add(s.node_id) + + edges: list[dict] = [] + seen: set[tuple] = set() + + for res in results: + for c in res.calls: + target_id = "" + if c.callee_usr and c.callee_usr in usr_to_id: + target_id = usr_to_id[c.callee_usr] + elif c.callee_name and c.callee_name in name_to_ids: + ids = name_to_ids[c.callee_name] + if len(ids) == 1: # single unambiguous definition + target_id = next(iter(ids)) + if not target_id: + continue + key = (c.caller_id, target_id) + if c.caller_id == target_id or key in seen: + continue + seen.add(key) + + caller_mod = id_to_module.get(c.caller_id, "") + callee_mod = id_to_module.get(target_id, "") + cross = bool(caller_mod and callee_mod and caller_mod != callee_mod) + edge = { + "source": c.caller_id, + "target": target_id, + "relation": "calls", + "context": "cross_module_call" if cross else "call", + "confidence": "EXTRACTED", + "confidence_score": 1.0, + "source_file": c.source_file, + "source_location": f"L{c.line}" if c.line else None, + "weight": 1.0, + } + if cross: + edge["metadata"] = { + "cross_module": True, + "caller_module": caller_mod, + "callee_module": callee_mod, + } + edges.append(edge) + return edges diff --git a/graphify/cpp_deep/macro.py b/graphify/cpp_deep/macro.py new file mode 100644 index 000000000..95b35159e --- /dev/null +++ b/graphify/cpp_deep/macro.py @@ -0,0 +1,108 @@ +"""Turn macro records into graph nodes and edges (Graphify node/edge dicts). + +The tree-sitter baseline produces NO macro nodes at all and drops function-like +macro call sites. This stage adds, for each TU's macros: + + * a macro NODE (``file_type: "code"``, label = macro name) so a macro is a + first-class, queryable entity in the graph; + * a ``contains`` edge from the defining file to the macro; + * a ``uses`` edge from each function that expands the macro to the macro node, + recording provenance (``caller -> LOG``) — the *expanded* call + (``caller -> log_write``) is emitted separately by the call stage, so the graph + shows both "went through macro LOG" and "which really landed on log_write". + +Node/edge dicts follow the schema the rest of graphify uses (see +``graphify/build.py`` / ``validate.py``): nodes need ``id, label, file_type, +source_file``; edges need ``source, target, relation, confidence, source_file``. +All macro edges are EXTRACTED — clang saw the real definition and expansion. +""" +from __future__ import annotations + +from graphify.cpp_deep.records import TUResult +from graphify.cpp_deep.usr_id import file_node_id + + +def macro_nodes_and_edges(results: list[TUResult]) -> tuple[list[dict], list[dict]]: + """Build macro nodes + (contains, uses) edges from parsed TU results. + + De-duplicates macro nodes by id (the same header macro appears in every TU that + includes it) so a macro defined once but expanded across many files is ONE node — + the same header-dedup discipline the whole pass follows. + """ + nodes: dict[str, dict] = {} + edges: list[dict] = [] + seen_edges: set[tuple] = set() + + def _add_edge(src: str, tgt: str, relation: str, source_file: str, line: int, context: str) -> None: + key = (src, tgt, relation) + if src == tgt or key in seen_edges: + return + seen_edges.add(key) + edges.append({ + "source": src, + "target": tgt, + "relation": relation, + "context": context, + "confidence": "EXTRACTED", + "confidence_score": 1.0, + "source_file": source_file, + "source_location": f"L{line}" if line else None, + "weight": 1.0, + }) + + for res in results: + for m in res.macros: + if m.node_id not in nodes: + nodes[m.node_id] = { + "id": m.node_id, + "label": m.label, + "file_type": "code", + "source_file": m.source_file, + "source_location": f"L{m.line}" if m.line else None, + "type": "macro", + "metadata": {"function_like": m.is_function_like, "module": m.module}, + "_origin": "cpp_deep", + } + # File -> macro `contains`, mirroring how functions attach to files. + # Use the baseline-aligned file id the parser stamped, not a + # recomputed one (which could miss the baseline's root convention). + file_id = m.file_id or file_node_id(m.source_file) + _add_edge(file_id, m.node_id, "contains", m.source_file, m.line, "macro_def") + + for use in res.macro_uses: + # caller -> macro node. The macro node id is derived the same way as the + # definition id, so this resolves even across TUs. + # Find the macro's defining source_file via the node registry when known; + # fall back to the use site's file for the edge's source_file field. + _add_edge( + use.caller_id, + _macro_id_for(use.macro_name, res), + "uses", + use.source_file, + use.line, + "macro_expansion", + ) + + # Only keep `uses` edges whose target is a real macro node we created; a use of a + # macro defined outside the corpus (system header) has no node and would dangle. + valid = set(nodes) + kept_edges = [ + e for e in edges + if e["relation"] != "uses" or e["target"] in valid + ] + return list(nodes.values()), kept_edges + + +def _macro_id_for(macro_name: str, res: TUResult) -> str: + """Resolve a macro name to its node id using the macros seen in this TU. + + Macro uses and defs are correlated within a translation unit (the use only + exists because the def was visible via an include in the same TU), so the TU's + own macro list is the correct scope to resolve the name against. + """ + for m in res.macros: + if m.label == macro_name: + return m.node_id + # Unknown (shouldn't happen — filtered by known_macros upstream); return a + # sentinel that won't match any node so the edge is dropped. + return f"__unknown_macro__{macro_name}" diff --git a/graphify/cpp_deep/merge.py b/graphify/cpp_deep/merge.py new file mode 100644 index 000000000..b7d351275 --- /dev/null +++ b/graphify/cpp_deep/merge.py @@ -0,0 +1,179 @@ +"""Fold libclang-derived symbols/calls/macros into the tree-sitter graph, in place. + +By the time this runs, ``all_nodes`` / ``all_edges`` already hold the tree-sitter +baseline. The deep pass reinforces it: + + * Symbol nodes: clang node IDs are aligned (via ``usr_id``) to the tree-sitter IDs, + so a clang function/method/class lands on the SAME node the baseline created. We + therefore only ADD nodes the baseline missed (e.g. macro-generated functions, + macros themselves) and enrich existing ones with a ``module`` tag — never + duplicate. This is the header-dedup win: one symbol, one node, regardless of how + many TUs included its header. + + * Call edges: clang calls are EXTRACTED and precise (USR-resolved, cross-module + aware). We add them and prune the baseline's now-redundant INFERRED ``calls`` + edges between the same endpoints, so the graph keeps the stronger evidence and + the community structure tightens around real call density. + + * Macro nodes/edges from ``macro.py``. + +Everything is additive-or-dedup; nothing the baseline got right is removed. The whole +function is wrapped by the caller in try/except so any failure leaves the baseline +intact. +""" +from __future__ import annotations + +import logging + +from graphify.cpp_deep.crossmod import build_symbol_index, resolve_cross_module_calls +from graphify.cpp_deep.macro import macro_nodes_and_edges +from graphify.cpp_deep.records import TUResult + +_LOG = logging.getLogger(__name__) + + +def merge_into_graph( + results: list[TUResult], + all_nodes: list[dict], + all_edges: list[dict], +) -> dict: + """Merge parsed TU results into all_nodes/all_edges in place. Returns a stats dict.""" + existing_ids = {n.get("id") for n in all_nodes if isinstance(n, dict)} + node_by_id = {n.get("id"): n for n in all_nodes if isinstance(n, dict)} + + usr_to_id, id_to_module = build_symbol_index(results) + + added_nodes = 0 + enriched = 0 + + # 1. Symbol nodes: add missing, enrich existing with module tag. Module comes + # from id_to_module (definition-owned), NOT the per-symbol declaration module, so + # a header-declared symbol is tagged with the module that actually defines it. + # + # Canonicalize by USR first: the same symbol is seen once per TU that includes + # its header, and with an include/ + src/ split the declaration file and the + # definition file have DIFFERENT stems (``_file_stem`` only unifies foo.h/foo.c in + # the same dir). Clang gives every occurrence the same USR, so we fold each + # symbol onto the node id its USR resolves to (the definition's, preferred by + # build_symbol_index) — collapsing the cross-directory / cross-project duplicates + # that stem-alignment alone can't. Only when a USR has no definition anywhere do + # we keep the symbol's own id. + for res in results: + for s in res.symbols: + canonical_id = usr_to_id.get(s.usr, s.node_id) if s.usr else s.node_id + sym_module = id_to_module.get(canonical_id) or id_to_module.get(s.node_id) or s.module + if canonical_id in existing_ids: + # Enrich: stamp module for cross-DLL grouping without disturbing the + # baseline node's label/source_file. + n = node_by_id.get(canonical_id) + if n is not None and sym_module: + meta = n.setdefault("metadata", {}) + if isinstance(meta, dict) and not meta.get("module"): + meta["module"] = sym_module + enriched += 1 + continue + # New node the tree-sitter baseline missed (macro-generated defs, or a + # symbol only clang's preprocessed view sees). Use the canonical id. + new_node = { + "id": canonical_id, + "label": s.label, + "file_type": "code", + "source_file": s.source_file, + "source_location": f"L{s.line}" if s.line else None, + "_origin": "cpp_deep", + } + if sym_module: + new_node["metadata"] = {"module": sym_module} + all_nodes.append(new_node) + node_by_id[canonical_id] = new_node + existing_ids.add(canonical_id) + added_nodes += 1 + # contains/method edge to its parent, if the parent node exists. + # Use canonical_id for the target: when USR folds this symbol onto a + # definition in a different file (e.g. declaration in include/foo.h → + # definition in src/foo.cpp), s.node_id reflects the declaration's stem + # while canonical_id is the definition's — the node we just created. + if s.parent_id and s.parent_id in existing_ids: + all_edges.append({ + "source": s.parent_id, + "target": canonical_id, + "relation": s.parent_relation, + "context": "cpp_deep", + "confidence": "EXTRACTED", + "confidence_score": 1.0, + "source_file": s.source_file, + "source_location": f"L{s.line}" if s.line else None, + "weight": 1.0, + }) + + # 2. Cross-module (and same-module) precise call edges. + call_edges = resolve_cross_module_calls(results, usr_to_id, id_to_module) + # Only keep edges whose endpoints are real nodes (a callee outside the corpus + # has no node — leave that to nothing, don't dangle). + valid = set(existing_ids) + call_edges = [e for e in call_edges if e["source"] in valid and e["target"] in valid] + + # Prune baseline INFERRED calls superseded by an EXTRACTED clang call between the + # same endpoints, so we don't double-count and the stronger evidence wins. + precise_pairs = {(e["source"], e["target"]) for e in call_edges} + before = len(all_edges) + all_edges[:] = [ + e for e in all_edges + if not ( + e.get("relation") == "calls" + and e.get("confidence") == "INFERRED" + and (e.get("source"), e.get("target")) in precise_pairs + ) + ] + pruned = before - len(all_edges) + + # De-dup call edges against surviving baseline EXTRACTED calls (same endpoints). + existing_call_pairs = { + (e.get("source"), e.get("target")) + for e in all_edges if e.get("relation") == "calls" + } + added_calls = 0 + cross_calls = 0 + for e in call_edges: + if (e["source"], e["target"]) in existing_call_pairs: + continue + all_edges.append(e) + existing_call_pairs.add((e["source"], e["target"])) + added_calls += 1 + if e.get("context") == "cross_module_call": + cross_calls += 1 + + # 3. Macro nodes + (contains, uses) edges. + macro_ns, macro_es = macro_nodes_and_edges(results) + added_macros = 0 + for mn in macro_ns: + if mn["id"] in existing_ids: + continue + all_nodes.append(mn) + existing_ids.add(mn["id"]) + node_by_id[mn["id"]] = mn + added_macros += 1 + valid = set(existing_ids) + macro_edge_pairs = {(e.get("source"), e.get("target"), e.get("relation")) for e in all_edges} + added_macro_edges = 0 + for e in macro_es: + if e["source"] not in valid or e["target"] not in valid: + continue + k = (e["source"], e["target"], e["relation"]) + if k in macro_edge_pairs: + continue + all_edges.append(e) + macro_edge_pairs.add(k) + added_macro_edges += 1 + + stats = { + "added_nodes": added_nodes, + "enriched_nodes": enriched, + "added_calls": added_calls, + "cross_module_calls": cross_calls, + "pruned_inferred_calls": pruned, + "added_macros": added_macros, + "added_macro_edges": added_macro_edges, + } + _LOG.info("cpp-deep merge: %s", stats) + return stats diff --git a/graphify/cpp_deep/pipeline.py b/graphify/cpp_deep/pipeline.py new file mode 100644 index 000000000..132034a89 --- /dev/null +++ b/graphify/cpp_deep/pipeline.py @@ -0,0 +1,432 @@ +"""Stage orchestration for the deep C/C++ pass. + +Runs as a registered ``LanguageResolver`` (see ``resolver.py``), so its inputs are +``(per_file, all_nodes, all_edges)`` — the tree-sitter baseline graph, mutated in +place. It has no direct access to the scan paths/root, but at resolver time (before +extract() relativizes them) node ``source_file`` fields are still ABSOLUTE, so we +reconstruct the C/C++ translation-unit set and the scan root from the baseline nodes. + +Flow: + 1. Collect C/C++ source files + infer the scan root from baseline node source_files. + 2. compile_db: obtain per-TU compile flags + owning module. + 3. clang_index: parse each TU into symbol/call/macro records (PARALLEL). + 4. merge: fold records into all_nodes/all_edges (dedup by aligned id, add precise + EXTRACTED cross-module calls, prune superseded INFERRED calls, add macro nodes). + +Any failure raises out to ``resolver.resolve_cpp_deep`` which keeps the baseline. +""" +from __future__ import annotations + +import gc +import json as _json +import logging +import os +import sys +import time +from concurrent.futures import ThreadPoolExecutor, as_completed +from pathlib import Path + +_LOG = logging.getLogger(__name__) + +_CPP_TU_SUFFIXES = {".c", ".cpp", ".cc", ".cxx", ".cu"} +_CPP_ANY_SUFFIXES = _CPP_TU_SUFFIXES | {".h", ".hpp", ".hh", ".hxx", ".cuh"} + + +def _status_file() -> Path | None: + """Locate the graphify output directory for the progress status file.""" + env = os.environ.get("GRAPHIFY_OUT", "").strip() + base = Path(env) if env else Path("graphify-out") + base.mkdir(parents=True, exist_ok=True) + return base / ".cpp_deep_status.json" + + +def _write_status(phase: str, total: int, completed: int, current: str, + elapsed_s: float, workers: int, avg_s_per_tu: float = 0.0) -> None: + """Atomically write a progress status file. Fails silently on any error.""" + try: + sf = _status_file() + if sf is None: + return + est = 0.0 + if completed > 0 and avg_s_per_tu > 0: + est = (total - completed) * avg_s_per_tu + payload = _json.dumps({ + "phase": phase, + "total": total, + "completed": completed, + "current": current, + "elapsed_s": round(elapsed_s, 1), + "workers": workers, + "avg_s_per_tu": round(avg_s_per_tu, 2), + "est_remaining_s": round(est, 1), + }, ensure_ascii=False) + tmp = sf.with_suffix(".tmp") + tmp.write_text(payload, encoding="utf-8") + tmp.replace(sf) + except Exception: + pass # status file is best-effort; never block the parse on it + + +def _ensure_logging() -> None: + """Guarantee that cpp-deep progress messages reach a log file and stderr. + + Python's root logger defaults to WARNING — all INFO messages are silently + dropped without a configured handler. This function is called as the very + first thing inside run_pipeline() so progress is ALWAYS visible, regardless + of how graphify CLI / the skill / programmatic extract() was invoked. + + Idempotent: if handlers already exist, nothing is added. + """ + root_logger = logging.getLogger("graphify.cpp_deep") + if root_logger.handlers: + return # already configured (e.g. by __main__.py) + + root_logger.setLevel(logging.INFO) + root_logger.propagate = False + + # stderr (direct terminal use) + sh = logging.StreamHandler(sys.stderr) + sh.setLevel(logging.INFO) + sh.setFormatter(logging.Formatter("[graphify cpp-deep] %(message)s")) + root_logger.addHandler(sh) + + # log file (survives stdout/stderr capture by host processes) + try: + _out = Path(os.environ.get("GRAPHIFY_OUT", "").strip() or "graphify-out") + _out.mkdir(parents=True, exist_ok=True) + fh = logging.FileHandler(str(_out / "cpp_deep.log"), encoding="utf-8") + fh.setLevel(logging.INFO) + fh.setFormatter(logging.Formatter("[graphify cpp-deep] %(message)s")) + if hasattr(fh.stream, "reconfigure"): + fh.stream.reconfigure(line_buffering=True) + root_logger.addHandler(fh) + except Exception: + pass # file handler is best-effort; stderr is the fallback + + +def run_pipeline(per_file: list[dict], all_nodes: list[dict], all_edges: list[dict]) -> None: + """Enhance the graph in place with the libclang-derived C/C++ semantics.""" + _ensure_logging() + + from graphify.cpp_deep.compile_db import obtain_compile_units + from graphify.cpp_deep.clang_index import parse_translation_unit + from graphify.cpp_deep.merge import merge_into_graph + + cpp_files, root = _collect_cpp_files(all_nodes) + + # Honour .graphifyignore (+ .gitignore): clang may discover symbols inside + # headers that the user marked as ignored (e.g. generated .ph.h, precompiled + # headers). We filter BEFORE the TU list is built so ignored directories + # don't even enter the parse queue. + ignore_patterns = _load_ignore_patterns(root) + if ignore_patterns: + before = len(cpp_files) + cpp_files = [f for f in cpp_files if not _is_ignored_path(f, root, ignore_patterns)] + skipped_ignored = before - len(cpp_files) + if skipped_ignored: + _LOG.info("cpp-deep: skipped %d file(s) matching .graphifyignore", skipped_ignored) + + if not cpp_files: + _LOG.info("cpp-deep: no C/C++ source files found in baseline; nothing to do") + _write_status("skipped", 0, 0, "no C/C++ source files in corpus", 0, 0, 0) + return + + tu_files = [p for p in cpp_files if p.suffix.lower() in _CPP_TU_SUFFIXES] + if not tu_files: + _LOG.info("cpp-deep: only headers present (%d header files), no translation units to parse", + len(cpp_files)) + _write_status("skipped", 0, 0, "only headers, no .c/.cpp files", 0, 0, 0) + return + + compile_db = os.environ.get("GRAPHIFY_CPP_COMPILE_DB", "").strip() or None + units = obtain_compile_units(tu_files, root, compile_db=compile_db) + if not units: + _LOG.warning("cpp-deep: no compile units resolved (%d C/C++ files found); keeping tree-sitter baseline", + len(tu_files)) + _write_status("skipped", len(tu_files), 0, "no compile units resolved", 0, 0, 0) + return + + _LOG.info("cpp-deep: %d translation units to parse", len(units)) + + # Map resolved abs source path -> baseline SYMBOL STEM (for symbol ids) and -> + # baseline FILE-NODE id (for `contains` parents), so cpp_deep ids key off exactly + # what the tree-sitter baseline produced — the header-dedup correctness hinge, + # robust to whatever cache_root the baseline ran with. + stem_map = _baseline_file_ids(all_nodes) + file_id_map = _baseline_file_node_ids(all_nodes) + + def _stem_of(abs_path: Path) -> str: + rp = abs_path.resolve() + return stem_map.get(rp) or stem_map.get(rp.with_suffix("")) or "" + + def _file_id_of(abs_path: Path) -> str: + rp = abs_path.resolve() + return file_id_map.get(rp, "") + + # ── parallel libclang parse ────────────────────────────────────────── + # libclang parse is I/O-bound native code (.parse calls release the GIL), so + # ThreadPoolExecutor gives near-linear speedup. Default to min(cpu_count, 4) + # since too many concurrent parses can thrash memory on header-heavy projects. + # Honour GRAPHIFY_CPP_WORKERS (--cpp-workers flag) if set. + max_w = int(os.environ.get("GRAPHIFY_CPP_WORKERS", "0") or "0") + if max_w <= 0: + cpu = os.cpu_count() or 1 + max_w = min(cpu, 4) + workers = min(max_w, len(units)) + + # Build an ignore-filter callback for clang's _in_corpus: prevents recording + # symbols discovered in user-ignored headers (generated .ph.h, precompiled, etc.) + def _ignore_filter(abs_path: Path) -> bool: + return _is_ignored_path(abs_path, root, ignore_patterns) if ignore_patterns else False + + def _parse_and_cache(unit): + """Worker: parse one TU, write its result to disk cache, return metadata only. + Skips the expensive libclang parse if a valid cache file already exists (crash + recovery / re-run).""" + from graphify.cpp_deep.cache import exists as _cache_exists, write as _cache_write + t0 = time.perf_counter() + if _cache_exists(str(unit.path), unit.args): + elapsed = 0.0 + diags: list[str] = [] + return unit, diags, elapsed, True + res = parse_translation_unit(unit, root, stem_of=_stem_of, file_id_of=_file_id_of, + ignore_filter=_ignore_filter) + elapsed = time.perf_counter() - t0 + _cache_write(res, unit.args) + return unit, res.diagnostics, elapsed, False + + _GC_INTERVAL = 20 # force GC every N TUs to reclaim disposed libclang native memory + + def _maybe_gc() -> None: + gc.collect() + + parse_errors = 0 + cache_hits = 0 + started = time.perf_counter() + _write_status("parsing", len(units), 0, "", 0, workers) + + if workers > 1 and len(units) > 1: + _LOG.info("cpp-deep: parsing %d TUs with %d workers (parallel)", len(units), workers) + with ThreadPoolExecutor(max_workers=workers) as pool: + future_to_idx = {pool.submit(_parse_and_cache, u): i for i, u in enumerate(units)} + done_count = 0 + for future in as_completed(future_to_idx): + unit, diags, elapsed, cached = future.result() + done_count += 1 + if cached: + cache_hits += 1 + total = len(units) + avg = (time.perf_counter() - started) / max(done_count, 1) + if done_count % max(total // 20, 5) == 0 or elapsed > 30: + tag = " (cached)" if cached else (" - SLOW" if elapsed > 30 else "") + _LOG.info( + "cpp-deep: [%d/%d] %s (%.1fs)%s", + done_count, total, unit.path.name, elapsed, tag, + ) + _write_status("parsing", total, done_count, unit.path.name, + time.perf_counter() - started, workers, avg) + if diags: + parse_errors += 1 + if done_count % _GC_INTERVAL == 0: + _maybe_gc() + _maybe_gc() + else: + _LOG.info("cpp-deep: parsing %d TUs sequentially (single worker)", len(units)) + for i, unit in enumerate(units): + unit, diags, elapsed, cached = _parse_and_cache(unit) + if cached: + cache_hits += 1 + total = len(units) + avg = (time.perf_counter() - started) / max(i + 1, 1) + if (i + 1) % max(total // 20, 5) == 0 or elapsed > 30: + tag = " (cached)" if cached else (" - SLOW" if elapsed > 30 else "") + _LOG.info( + "cpp-deep: [%d/%d] %s (%.1fs)%s", + i + 1, total, unit.path.name, elapsed, tag, + ) + _write_status("parsing", total, i + 1, unit.path.name, + time.perf_counter() - started, workers, avg) + if diags: + parse_errors += 1 + if (i + 1) % _GC_INTERVAL == 0: + _maybe_gc() + _maybe_gc() + + fresh = len(units) - cache_hits + total_elapsed = time.perf_counter() - started + _LOG.info( + "cpp-deep: %d/%d TUs parsed (fresh), %d cached, %.1fs total (%.1f s/TU avg, %d with diagnostics)", + fresh, len(units), cache_hits, + total_elapsed, + total_elapsed / max(fresh, 1) if fresh else 0.0, + parse_errors, + ) + _write_status("parsing_done", len(units), len(units), "", + total_elapsed, workers, + total_elapsed / max(len(units), 1)) + + # ── load cached results & merge ────────────────────────────────────── + # All TUResults are deserialised at once for the merge phase. Each is + # ~100 KB of Python records (not the native TU which is already disposed), + # so 812 TUs ≈ 80 MB — a small fraction of the original multi-GB footprint. + # One-shot loading preserves USR-based dedup across all TUs. + _write_status("merging", len(units), len(units), "", + time.perf_counter() - started, workers, + total_elapsed / max(len(units), 1)) + from graphify.cpp_deep.cache import load as _cache_load + _LOG.info("cpp-deep: loading %d cached results for merge...", len(units)) + results = [] + cache_misses = 0 + for unit in units: + r = _cache_load(str(unit.path), unit.args) + if r is not None: + results.append(r) + else: + cache_misses += 1 + _LOG.error("cpp-deep: cache miss for %s — TU will be ABSENT from graph", unit.path.name) + if cache_misses: + _LOG.error("cpp-deep: %d cache misses — %d TUs are MISSING from the graph", cache_misses, cache_misses) + stats = merge_into_graph(results, all_nodes, all_edges) + _LOG.info("cpp-deep merge: %s", stats) + + # Remove status file only (cache is preserved for incremental --update). + try: + sf = _status_file() + if sf is not None and sf.exists(): + sf.unlink(missing_ok=True) + except Exception: + pass + + +def _baseline_file_ids(all_nodes: list[dict]) -> dict[Path, str]: + """Map resolved abs source path -> baseline SYMBOL STEM for that file. + + cpp_deep must key its symbol ids off the SAME stem the tree-sitter baseline used + (``make_id(_file_stem(path), name)``) or a symbol splits into two nodes. The + file-node id is NOT a reliable source for that stem: depending on cache_root it + can be ``make_id(abs_path)`` or a doubled relativized form, and neither equals + ``make_id(_file_stem(path))``. + + Instead we recover the stem from the baseline's own SYMBOL nodes: a function node + has id ``make_id(stem, name)`` and label ``name()``. Stripping the normalized + ``_`` suffix from the id yields exactly the stem, whatever root the baseline + used. We take the first symbol per file; all symbols in a file share the stem. + """ + from graphify.ids import normalize_id + + out: dict[Path, str] = {} + for n in all_nodes: + if not isinstance(n, dict): + continue + sf = n.get("source_file") + nid = n.get("id") + label = n.get("label") + if not sf or not nid or not label: + continue + p = Path(str(sf)) + if p.suffix.lower() not in _CPP_ANY_SUFFIXES: + continue + try: + key = p.resolve() + except Exception: + continue + if key in out: + continue + if not (isinstance(label, str) and label.endswith("()")): + continue + # Skip method nodes (label starts with ".") — their id contains an + # extra class-name segment (make_id(stem, "Foo", "bar")) that we cannot + # strip cleanly. File-level free functions give us the correct stem. + if label.startswith("."): + continue + name = label[:-2] + if not name: + continue + suffix = "_" + normalize_id(name) + sid = str(nid) + if sid.endswith(suffix): + stem = sid[: -len(suffix)] + out[key] = stem + # Extension-dropped alias so a header declaration resolves to the same + # stem as its sibling source definition (baseline collapses Foo.h/Foo.c). + out.setdefault(key.with_suffix(""), stem) + return out + + +def _baseline_file_node_ids(all_nodes: list[dict]) -> dict[Path, str]: + """Map resolved abs source path -> baseline FILE-NODE id (label == basename).""" + out: dict[Path, str] = {} + for n in all_nodes: + if not isinstance(n, dict): + continue + sf = n.get("source_file") + nid = n.get("id") + if not sf or not nid or n.get("label") != Path(str(sf)).name: + continue + p = Path(str(sf)) + if p.suffix.lower() not in _CPP_ANY_SUFFIXES: + continue + try: + out[p.resolve()] = str(nid) + except Exception: + continue + return out + + +def _collect_cpp_files(all_nodes: list[dict]) -> tuple[list[Path], Path]: + """Return (unique C/C++ file paths, inferred scan root) from baseline nodes. + + At resolver time source_file is absolute; we take the set of C/C++ source files + and infer the root as their longest common directory prefix (matching how + extract() itself infers the cache root). + """ + files: set[Path] = set() + for n in all_nodes: + if not isinstance(n, dict): + continue + sf = n.get("source_file") + if not sf: + continue + p = Path(str(sf)) + if p.suffix.lower() in _CPP_ANY_SUFFIXES and p.is_absolute() and p.is_file(): + files.add(p.resolve()) + + if not files: + return [], Path(".") + + root = _common_root(files) + return sorted(files), root + + +def _common_root(files: set[Path]) -> Path: + """Longest common directory prefix of the given files.""" + parts_lists = [list(f.parts) for f in files] + common: list[str] = [] + for tup in zip(*parts_lists): + if len(set(tup)) == 1: + common.append(tup[0]) + else: + break + if not common: + return Path(next(iter(files)).anchor or ".") + root = Path(*common) + return root if root.is_dir() else root.parent + + +def _load_ignore_patterns(root: Path) -> list[tuple[Path, str]]: + """Load .graphifyignore + .gitignore patterns, matching graphify's detect.py.""" + try: + from graphify.detect import _load_graphifyignore + return _load_graphifyignore(root) + except Exception: + return [] + + +def _is_ignored_path(path: Path, root: Path, + patterns: list[tuple[Path, str]]) -> bool: + """Check if a path matches any ignore pattern, reusing graphify's logic.""" + try: + from graphify.detect import _is_ignored + return _is_ignored(path, root, patterns) + except Exception: + return False \ No newline at end of file diff --git a/graphify/cpp_deep/records.py b/graphify/cpp_deep/records.py new file mode 100644 index 000000000..88078491b --- /dev/null +++ b/graphify/cpp_deep/records.py @@ -0,0 +1,134 @@ +"""Intermediate records produced by the libclang pass, consumed by macro/crossmod/merge. + +These are plain dataclasses (no clang types leak out) so the downstream stages — +and the tests — never need libclang imported. Every symbol carries both its Clang +USR (translation-unit-independent identity, used to link a call to the exact +definition across TUs/DLLs) and its Graphify canonical ``node_id`` (aligned to the +tree-sitter extractor so the two passes reinforce instead of fragment). + +All record types support ``to_dict()`` / ``from_dict()`` for caching to disk so +large multi-TU projects can stream results through files instead of accumulating +everything in memory. +""" +from __future__ import annotations + +from dataclasses import dataclass, field, asdict +from typing import Any + + +@dataclass +class SymbolRecord: + """A defined or declared C/C++ symbol (function, method, class/struct).""" + + usr: str # Clang USR — stable cross-TU identity + node_id: str # Graphify canonical id (tree-sitter-aligned) + label: str + kind: str # "function" | "method" | "class" | "struct" | "macro" + source_file: str # repo-relative posix path + line: int + module: str = "" # owning DLL/exe/lib target + is_definition: bool = False + parent_id: str = "" # container node id (file or class) for contains/method edge + parent_relation: str = "contains" + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + @classmethod + def from_dict(cls, d: dict[str, Any]) -> SymbolRecord: + return cls(**d) + + +@dataclass +class CallRecord: + """A resolved call site: caller symbol -> callee (by USR when known).""" + + caller_id: str + callee_usr: str = "" + callee_name: str = "" + source_file: str = "" + line: int = 0 + via_macro: str = "" + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + @classmethod + def from_dict(cls, d: dict[str, Any]) -> CallRecord: + return cls(**d) + + +@dataclass +class MacroRecord: + """A macro definition and (optionally) what its expansions call.""" + + node_id: str + label: str + source_file: str + line: int + is_function_like: bool = False + module: str = "" + file_id: str = "" + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + @classmethod + def from_dict(cls, d: dict[str, Any]) -> MacroRecord: + return cls(**d) + + +@dataclass +class MacroUse: + """A macro expansion site inside a function body (caller uses macro).""" + + caller_id: str + macro_name: str + source_file: str + line: int + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + @classmethod + def from_dict(cls, d: dict[str, Any]) -> MacroUse: + return cls(**d) + + +@dataclass +class TUResult: + """Everything extracted from one translation unit.""" + + path: str + module: str = "" + symbols: list[SymbolRecord] = field(default_factory=list) + calls: list[CallRecord] = field(default_factory=list) + macros: list[MacroRecord] = field(default_factory=list) + macro_uses: list[MacroUse] = field(default_factory=list) + usr_to_id: dict[str, str] = field(default_factory=dict) + diagnostics: list[str] = field(default_factory=list) + + def to_dict(self) -> dict[str, Any]: + return { + "path": self.path, + "module": self.module, + "symbols": [s.to_dict() for s in self.symbols], + "calls": [c.to_dict() for c in self.calls], + "macros": [m.to_dict() for m in self.macros], + "macro_uses": [u.to_dict() for u in self.macro_uses], + "usr_to_id": dict(self.usr_to_id), + "diagnostics": list(self.diagnostics), + } + + @classmethod + def from_dict(cls, d: dict[str, Any]) -> TUResult: + return cls( + path=d.get("path", ""), + module=d.get("module", ""), + symbols=[SymbolRecord.from_dict(s) for s in d.get("symbols", [])], + calls=[CallRecord.from_dict(c) for c in d.get("calls", [])], + macros=[MacroRecord.from_dict(m) for m in d.get("macros", [])], + macro_uses=[MacroUse.from_dict(u) for u in d.get("macro_uses", [])], + usr_to_id=dict(d.get("usr_to_id", {})), + diagnostics=list(d.get("diagnostics", [])), + ) diff --git a/graphify/cpp_deep/resolver.py b/graphify/cpp_deep/resolver.py new file mode 100644 index 000000000..d908a1825 --- /dev/null +++ b/graphify/cpp_deep/resolver.py @@ -0,0 +1,53 @@ +"""The registered cross-file resolver entry point for the deep C/C++ pass. + +Signature matches ``graphify.resolver_registry.LanguageResolver.resolve``: +``resolve(per_file, all_nodes, all_edges) -> None``, mutating ``all_nodes`` / +``all_edges`` in place. The registry only calls this when a C/C++ suffix is present +in the corpus AND the package gate (``cpp_deep.maybe_register``) registered us, so by +the time we run libclang is known-available and the user opted in. + +Orchestration (each stage is a separate module, all failure-isolated): + + 1. compile_db — obtain compile commands (compile_commands.json, or synthesize + from .sln/.vcxproj, or heuristics). + 2. clang_index — parse each translation unit into symbol/call/type records keyed + by USR and by canonical (tree-sitter-aligned) node ID. + 3. macro — add macro nodes + expansion edges + macro-generated functions. + 4. crossmod — resolve calls across modules/DLLs by USR and tag nodes with module. + 5. merge — fold the clang records into all_nodes/all_edges, preferring + EXTRACTED clang edges over the tree-sitter INFERRED ones. + +This module stays thin: it wires the stages and guarantees the whole thing is inert +on any error. The heavy lifting lives in the sibling modules so each is testable in +isolation. +""" +from __future__ import annotations + +import logging + +_LOG = logging.getLogger(__name__) + + +def resolve_cpp_deep(per_file: list[dict], all_nodes: list[dict], all_edges: list[dict]) -> None: + """Run the deep C/C++ enhancement over the already-extracted graph, in place.""" + # Re-check the opt-in gate at RUN time, not just at registration. The registry is + # process-global and registration is sticky (idempotent), so once a --cpp-deep run + # registers us we stay registered for the life of the process. A later run in the + # same process that did NOT opt in must be a no-op — otherwise the enhancement + # leaks into unrelated extractions (e.g. back-to-back CLI calls, or tests). + from graphify.cpp_deep import is_enabled + + if not is_enabled(): + return + + try: + from graphify.cpp_deep.pipeline import run_pipeline + except Exception as exc: # pragma: no cover - import wiring + _LOG.warning("cpp-deep pipeline unavailable, skipping: %s", exc) + return + + try: + run_pipeline(per_file, all_nodes, all_edges) + except Exception as exc: + # Never abort a build. The tree-sitter baseline is already in all_nodes. + _LOG.warning("cpp-deep pipeline failed, keeping tree-sitter baseline: %s", exc) diff --git a/graphify/cpp_deep/usr_id.py b/graphify/cpp_deep/usr_id.py new file mode 100644 index 000000000..202ed614b --- /dev/null +++ b/graphify/cpp_deep/usr_id.py @@ -0,0 +1,76 @@ +"""Map Clang USRs and symbol locations to Graphify canonical node IDs. + +The whole point of the deep pass is to collapse the *header duplication* problem: +one symbol declared in ``foo.h`` and included by dozens of ``.c`` files must become +ONE node, and it must share its ID with whatever the tree-sitter baseline produced +for the same symbol so the two passes reinforce instead of fragment. + +Graphify's code-node dedup is ID-only (``dedup.py`` skips label-merge for +``file_type == "code"``), so ID agreement is the entire mechanism. The tree-sitter +extractor builds symbol IDs as ``make_id(file_stem, name)`` where ``file_stem`` is +the source file's repo-relative path with the extension dropped +(``graphify.extractors.base._file_stem``). We reproduce that exact recipe here, but +anchor it to the symbol's *canonical declaration location* rather than the +translation unit currently being compiled — that is what makes the ID +TU-independent: no matter which ``.c`` pulled in ``foo.h``, a function declared in +``foo.h`` keys to ``make_id("<...>/foo", name)``. + +For C++ methods the tree-sitter side keys members as ``make_id(class_id, name)`` +where ``class_id = make_id(stem, ...namespace..., class_name)``. We mirror that by +composing the semantic parent chain from Clang's semantic parents. +""" +from __future__ import annotations + +from pathlib import Path + +from graphify.ids import make_id + +# Reuse the exact stem recipe the tree-sitter extractor uses for node IDs, so a +# clang-derived ID for the same symbol in the same file is byte-identical. +from graphify.extractors.base import _file_stem + + +def canonical_stem(source_file: str | Path, root: Path | None = None) -> str: + """Repo-relative, extension-dropped stem for a file, matching the AST side. + + ``root`` relativizes an absolute path the same way ``extract()`` does before it + persists IDs (the AST extractor's post-pass re-derives the repo-relative form + from ``source_file``). When ``root`` is None or relativization fails, the path is + used as-is (already relative), which still matches the AST side for in-repo files. + """ + p = Path(source_file) + if root is not None and p.is_absolute(): + try: + p = p.relative_to(root) + except ValueError: + pass + return _file_stem(p) + + +def file_symbol_id(source_file: str | Path, name: str, root: Path | None = None) -> str: + """ID for a free function / file-level symbol: ``make_id(stem, name)``. + + Matches ``_extract_generic``'s free-function id (extract.py ~L4473). + """ + return make_id(canonical_stem(source_file, root), name) + + +def member_symbol_id(class_id: str, name: str) -> str: + """ID for a class/struct member, matching the AST side ``make_id(class_id, name)``.""" + return make_id(class_id, name) + + +def type_id(source_file: str | Path, qualified_parts: list[str], root: Path | None = None) -> str: + """ID for a class/struct/namespaced type. + + Mirrors the AST side ``make_id(stem, *namespace, class_name)`` (extract.py ~L3646 + for C++ class/struct nodes) so a clang type node lands on the same id as its + tree-sitter counterpart. + """ + return make_id(canonical_stem(source_file, root), *qualified_parts) + + +def file_node_id(source_file: str | Path, root: Path | None = None) -> str: + """ID for the file node itself. AST side is ``make_id(str(path))`` on the + repo-relative path (the source_file it later relativizes to).""" + return make_id(canonical_stem(source_file, root)) diff --git a/graphify/extract.py b/graphify/extract.py index 59d4f8283..cfd97bc5b 100644 --- a/graphify/extract.py +++ b/graphify/extract.py @@ -20,7 +20,6 @@ run_language_resolvers, ) from .ruby_resolution import resolve_ruby_member_calls -from .pascal_resolution import resolve_pascal_inherited_calls # --- migrated to graphify/extractors/ (see graphify/extractors/MIGRATION.md) --- from graphify.extractors.base import ( # noqa: F401 @@ -29,116 +28,17 @@ _make_id, _read_text, ) -from graphify.extractors.apex import extract_apex # noqa: F401 -from graphify.extractors.bash import extract_bash # noqa: F401 from graphify.extractors.blade import extract_blade # noqa: F401 from graphify.extractors.csharp import ( _resolve_cross_file_csharp_imports, _resolve_csharp_type_references, ) -from graphify.extractors.dart import extract_dart # noqa: F401 -from graphify.extractors.dm import extract_dm, extract_dmf, extract_dmi, extract_dmm # noqa: F401 from graphify.extractors.elixir import extract_elixir # noqa: F401 -from graphify.extractors.fortran import _cpp_preprocess, extract_fortran # noqa: F401 -from graphify.extractors.go import extract_go # noqa: F401 -from graphify.extractors.json_config import extract_json # noqa: F401 -from graphify.extractors.markdown import extract_markdown # noqa: F401 -from graphify.extractors.pascal_forms import extract_delphi_form, extract_lazarus_form # noqa: F401 -from graphify.extractors.powershell import extract_powershell, extract_powershell_manifest # noqa: F401 from graphify.extractors.razor import extract_razor # noqa: F401 -from graphify.extractors.rust import extract_rust # noqa: F401 -from graphify.extractors.sln import extract_sln # noqa: F401 -from graphify.extractors.sql import extract_sql # noqa: F401 -from graphify.extractors.terraform import extract_terraform # noqa: F401 -from graphify.extractors.verilog import extract_verilog # noqa: F401 from graphify.extractors.zig import extract_zig # noqa: F401 from graphify.security import sanitize_metadata from graphify.paths import disambiguate_ambiguous_candidates -from graphify.extractors.models import LanguageConfig, _JS_CACHE_BYPASS_SUFFIXES, _NamespaceExportFact, _StarExportFact, _SymbolAliasFact, _SymbolDeclarationFact, _SymbolExportFact, _SymbolImportFact, _SymbolResolutionFacts, _SymbolUseFact, _WORKSPACE_PACKAGE_CACHE # noqa: E402,F401 - -from graphify.extractors.resolution import ( # noqa: E402,F401 - _DECLDEF_HEADER_SUFFIXES, - _DECLDEF_IMPL_SUFFIXES, - _EXPORT_CONDITION_PRIORITY, - _JS_INDEX_FILES, - _JS_PRIMITIVE_TYPES, - _JS_RESOLVE_EXTS, - _TSCONFIG_ALIAS_CACHE, - _VUE_SCRIPT_LANG_RE, - _VUE_SCRIPT_RE, - _WORKSPACE_MANIFEST_NAMES, - _apply_symbol_resolution_facts, - _augment_symbol_resolution_edges, - _collect_js_symbol_resolution_facts, - _collect_python_symbol_resolution_facts, - _contained_in_package, - _decldef_class_stem, - _disambiguate_colliding_node_ids, - _find_workspace_root, - _is_type_like_definition, - _js_call_identifier, - _js_default_export_name, - _js_default_import_name, - _js_export_clause, - _js_export_statement_is_star, - _js_exported_declaration_names, - _js_lexical_aliases, - _js_module_specifier, - _js_named_specifiers, - _js_namespace_export_name, - _js_source_path, - _js_top_level_function_bodies, - _load_tsconfig_aliases, - _load_workspace_packages, - _match_tsconfig_alias, - _merge_decl_def_classes, - _node_disambiguation_source_key, - _package_entry_candidates, - _parse_js_tree, - _parse_python_tree, - _pascal_class_stem_cache, - _pascal_project_root, - _pascal_resolve_class, - _pascal_resolve_unit, - _pascal_unit_cache, - _pnpm_workspace_globs, - _python_call_identifier, - _python_import_from_module, - _python_imported_names, - _python_top_level_function_bodies, - _read_tsconfig_aliases, - _resolve_c_include_path, - _resolve_cross_file_imports, - _resolve_cross_file_java_imports, - _resolve_export_target, - _resolve_java_type_references, - _resolve_js_import_path, - _resolve_js_import_target, - _resolve_js_module_path, - _resolve_lua_import_target, - _resolve_python_module_path, - _resolve_tsconfig_alias, - _resolve_workspace_import, - _source_key, - _strip_jsonc, - _ts_collect_type_refs, - _ts_heritage_clause_entries, - _ts_walk_class_members, - _vue_mask_non_script, - _walk_js_tree, - _walk_python_tree, - _workspace_globs, -) - -from graphify.extractors.engine import REFERENCE_CONTEXTS, _CSHARP_TYPE_PARAMETER_SCOPE_DECLARATIONS, _C_PRIMITIVE_TYPE_NODES, _JAVA_BUILTIN_TYPES, _JAVA_TYPE_PARAMETER_SCOPE_DECLARATIONS, _JS_FUNCTION_VALUE_TYPES, _JS_SCOPE_BOUNDARY, _PYTHON_ANNOTATION_NOISE, _PYTHON_TYPE_CONTAINERS, _RUBY_CLASS_FACTORIES, _c_collect_type_refs, _cpp_collect_type_refs, _cpp_declarator_name, _cpp_local_var_types, _csharp_attribute_names, _csharp_classify_base, _csharp_collect_type_refs, _csharp_extra_walk, _csharp_member_type_table, _csharp_namespace_id, _csharp_namespace_name, _csharp_pre_scan_interfaces, _csharp_type_parameters_in_scope, _dynamic_import_js, _extract_generic, _find_body, _find_require_call, _get_cpp_func_name, _java_annotation_names, _java_collect_type_refs, _java_extra_walk, _java_type_parameters_in_scope, _js_collect_pattern_idents, _js_dispatch_value_idents, _js_extra_walk, _js_local_bound_names, _js_member_assignment_target, _js_module_bound_names, _kotlin_collect_type_refs, _kotlin_function_return_type_node, _kotlin_property_type_node, _kotlin_user_type_name, _php_collect_type_refs, _php_method_return_type_node, _php_name_text, _python_collect_assignment_targets, _python_collect_param_refs, _python_collect_type_refs, _python_local_bound_names, _python_module_bound_names, _python_param_names, _read_csharp_type_name, _require_imports_js, _ruby_const_last_name, _ruby_extra_walk, _ruby_local_class_bindings, _ruby_new_class_name, _scala_collect_type_refs, _semantic_reference_edge, _source_location, _swift_classify_base, _swift_collect_type_refs, _swift_constructor_type, _swift_declaration_keyword, _swift_extra_walk, _swift_local_var_types, _swift_pre_scan, _swift_property_name, _swift_property_type_node, _swift_receiver_name, _swift_user_type_name, _ts_decorator_name, _ts_descendant_decorators, _ts_emit_decorator_edges, _ts_extra_walk, _ts_method_name, _ts_receiver_type_table # noqa: E402,F401 - -from graphify.extractors.pascal import _PAS_BEGIN_END_TOKEN_RE, _PAS_CALL_RE, _PAS_END_SEMI_RE, _PAS_IMPL_HEADER_RE, _PAS_KEYWORDS, _PAS_METHOD_DECL_RE, _PAS_MODULE_RE, _PAS_TOKEN_RE, _PAS_TYPE_HEADER_RE, _PAS_USES_RE, _extract_pascal_regex, _pascal_find_body, _pascal_split_bases, _pascal_split_sections, _pascal_split_uses, _pascal_strip_comments, extract_pascal # noqa: E402,F401 - -from graphify.extractors.objc import _objc_local_var_types, extract_objc # noqa: E402,F401 - -from graphify.extractors.julia import extract_julia # noqa: E402,F401 - _RECURSION_LIMIT = 10_000 # Language built-in globals that AST may classify as call targets when used as @@ -167,6 +67,10 @@ def _safe_extract(extractor: Callable, path: Path) -> dict: return {"nodes": [], "edges": [], "error": f"{type(e).__name__}: {e}"} + + + + def _file_node_id(rel_path: Path) -> str: """File-level node ID matching the skill.md spec: ``{parent_dir}_{stem}`` — one parent directory level, no extension. ``rel_path`` MUST be relative to @@ -177,26 +81,791 @@ def _file_node_id(rel_path: Path) -> str: return _make_id(_file_stem(rel_path)) +def _csharp_namespace_id(dotted_name: str) -> str: + digest = hashlib.sha1(dotted_name.encode("utf-8")).hexdigest()[:16] + return f"csharp_namespace:{digest}" + + +_TSCONFIG_ALIAS_CACHE: dict[str, dict[str, list[str]]] = {} +_WORKSPACE_PACKAGE_CACHE: dict[str, dict[str, Path]] = {} +_WORKSPACE_MANIFEST_NAMES = ("pnpm-workspace.yaml", "package.json") +_JS_CACHE_BYPASS_SUFFIXES = {".js", ".jsx", ".mjs", ".ts", ".tsx", ".mts", ".cts", ".vue", ".svelte"} +_JS_RESOLVE_EXTS = (".ts", ".tsx", ".mts", ".cts", ".svelte", ".js", ".jsx", ".mjs") +_JS_INDEX_FILES = ("index.ts", "index.tsx", "index.svelte", "index.js", "index.jsx", "index.mjs") + + SEMANTIC_RELATIONS = frozenset({ "inherits", "implements", "mixes_in", "embeds", "references", "calls", "imports", "imports_from", "re_exports", "contains", "method", }) +REFERENCE_CONTEXTS = frozenset({ + "field", "parameter_type", "return_type", "generic_arg", "attribute", "value", "type", +}) + + +def _source_location(line: int | str | None) -> str | None: + if line is None: + return None + if isinstance(line, str): + return line if line.startswith("L") else f"L{line}" + return f"L{line}" + + +def _semantic_reference_edge( + source: str, + target: str, + context: str, + source_file: str, + line: int | str | None, +) -> dict: + if context not in REFERENCE_CONTEXTS: + raise ValueError(f"unknown reference context: {context}") + return { + "source": source, + "target": target, + "relation": "references", + "context": context, + "confidence": "EXTRACTED", + "source_file": source_file, + "source_location": _source_location(line), + "weight": 1.0, + } + + +def _resolve_js_import_path(candidate: Path) -> Path: + """Resolve a JS/TS/Svelte import target to a local file when it exists.""" + candidate = Path(os.path.normpath(candidate)) + if candidate.is_file(): + return candidate + + # TS ESM convention: imports often spell .js/.jsx while source is .ts/.tsx. + if candidate.suffix == ".js": + ts_candidate = candidate.with_suffix(".ts") + if ts_candidate.is_file(): + return ts_candidate + elif candidate.suffix == ".jsx": + tsx_candidate = candidate.with_suffix(".tsx") + if tsx_candidate.is_file(): + return tsx_candidate + + # Append extensions to the full filename, which covers extensionless imports, + # multi-dot helpers, and Svelte 5 rune files like Foo.svelte.ts. + for ext in _JS_RESOLVE_EXTS: + with_ext = candidate.parent / f"{candidate.name}{ext}" + if with_ext.is_file(): + return with_ext + + # Only fall back to directory indexes after file candidates lose. + if candidate.is_dir(): + for index_name in _JS_INDEX_FILES: + index_candidate = candidate / index_name + if index_candidate.is_file(): + return index_candidate + + return candidate + + +def _strip_jsonc(text: str) -> str: + """Strip // line comments, /* */ block comments, and trailing commas from JSONC. + + Preserves string contents (including // and /* inside strings) by skipping over + quoted spans first. Required for tsconfig.json files generated by SvelteKit, + NestJS, Vite, T3, Astro, etc., which use JSONC by default (#700). + """ + # Remove block and line comments while leaving string literals untouched. + pattern = re.compile( + r'"(?:\\.|[^"\\])*"' # double-quoted string (with escapes) + r"|/\*.*?\*/" # /* block comment */ + r"|//[^\n]*", # // line comment + re.DOTALL, + ) + + def _replace(match: re.Match) -> str: + token = match.group(0) + if token.startswith('"'): + return token + return "" + + stripped = pattern.sub(_replace, text) + # Remove trailing commas before } or ] (allowing whitespace between). + stripped = re.sub(r",(\s*[}\]])", r"\1", stripped) + return stripped + + +def _read_tsconfig_aliases(tsconfig: Path, base_dir: Path, seen: set) -> dict[str, list[str]]: + """Recursively read path aliases from a tsconfig, following extends chains. + + Child config paths override parent. Circular extends are detected via seen set. + npm package configs (e.g. @tsconfig/svelte) are skipped since they're not on disk. + Handles JSONC (comments + trailing commas) which is the default tsconfig format + for SvelteKit, NestJS, Vite, T3, Astro, etc. (#700). + """ + if str(tsconfig) in seen: + return {} + seen.add(str(tsconfig)) + try: + raw = tsconfig.read_text(encoding="utf-8") + except Exception as e: + print(f" warning: could not read {tsconfig} ({type(e).__name__}: {e})", file=sys.stderr, flush=True) + return {} + try: + data = json.loads(raw) + except json.JSONDecodeError: + try: + data = json.loads(_strip_jsonc(raw)) + except json.JSONDecodeError as e: + print(f" warning: failed to parse {tsconfig} as JSON/JSONC ({e.msg} at line {e.lineno} col {e.colno})", file=sys.stderr, flush=True) + return {} + except Exception as e: + print(f" warning: failed to parse {tsconfig} ({type(e).__name__}: {e})", file=sys.stderr, flush=True) + return {} + + aliases: dict[str, list[str]] = {} + # `extends` may be a string or, since TypeScript 5.0, an array of paths. + # For an array, parents are processed in order with later entries + # overriding earlier ones; the extending config (paths below) overrides + # all parents. Without the list branch, an array `extends` raised + # `AttributeError: 'list' object has no attribute 'startswith'`, which + # _safe_extract turned into a skip of the whole file. + extends = data.get("extends") + if isinstance(extends, str): + extends_list = [extends] + elif isinstance(extends, list): + extends_list = [e for e in extends if isinstance(e, str)] + else: + extends_list = [] + for ext in extends_list: + # Skip scoped npm package configs (e.g. @tsconfig/svelte) — not on disk. + if not ext or ext.startswith("@"): + continue + extended_path = (base_dir / ext).resolve() + if not extended_path.suffix: + extended_path = extended_path.with_suffix(".json") + if extended_path.exists(): + aliases.update(_read_tsconfig_aliases(extended_path, extended_path.parent, seen)) + + # tsconfig `paths` are resolved relative to `baseUrl` (itself relative to + # the tsconfig's directory), not the tsconfig directory directly. Honoring + # baseUrl is required for the common monorepo / NestJS layout where + # baseUrl points at a subdirectory, e.g. baseUrl "./src" with + # "@services/*": ["services/*"] must resolve to /src/services rather + # than /services. Defaults to "." so configs without baseUrl (paths + # relative to the tsconfig dir, the TS 4.1+ behavior) keep working. + compiler_options = data.get("compilerOptions", {}) + base_url = compiler_options.get("baseUrl") or "." + paths_base = base_dir / base_url + paths = compiler_options.get("paths", {}) + for alias, targets in paths.items(): + if not targets: + continue + # Keep ALL targets in declared order — tsc tries each until one resolves + # on disk. Discarding the fallbacks (#1531) misresolved/dropped imports + # whose file lived at a non-first target. Preserve wildcard tokens in + # both sides until the resolver substitutes the captured segment, then + # normalizes the concrete path (#927). Empty/non-string entries are skipped. + target_patterns = [ + str(paths_base / t) + for t in targets + if isinstance(t, str) and t + ] + if target_patterns: + aliases[alias] = target_patterns + + return aliases + + +def _load_tsconfig_aliases(start_dir: Path) -> dict[str, list[str]]: + """Walk up from start_dir to find tsconfig.json and return compilerOptions.paths aliases. + + Follows extends chains so SvelteKit/Nuxt/NestJS inherited aliases are included. + Returns a dict mapping alias patterns to ordered resolved target patterns; + wildcard tokens remain intact for substitution during resolution (#927). + Result is cached by tsconfig path string. + """ + current = start_dir.resolve() + for candidate in [current, *current.parents]: + tsconfig = candidate / "tsconfig.json" + if tsconfig.exists(): + key = str(tsconfig) + if key not in _TSCONFIG_ALIAS_CACHE: + _TSCONFIG_ALIAS_CACHE[key] = _read_tsconfig_aliases(tsconfig, candidate, seen=set()) + return _TSCONFIG_ALIAS_CACHE[key] + return {} + + +def _match_tsconfig_alias(raw: str, pattern: str) -> "tuple[tuple[int, int], str, bool] | None": + """Return (specificity, captured text, is_wildcard) when pattern matches raw. + + Exact aliases win first. Wildcard aliases follow TypeScript's longest-prefix + rule. The final branch preserves Graphify's existing support for treating a + non-wildcard alias as a directory prefix, but only after real wildcard matches. + """ + if "*" in pattern: + if pattern.count("*") != 1: + return None + prefix, suffix = pattern.split("*", 1) + if not raw.startswith(prefix) or not raw.endswith(suffix): + return None + end = len(raw) - len(suffix) if suffix else len(raw) + if end < len(prefix): + return None + return (1, -len(prefix)), raw[len(prefix):end], True + + if raw == pattern: + return (0, -len(pattern)), "", False + + prefix = pattern.rstrip("/") + if prefix and raw.startswith(prefix + "/"): + return (2, -len(prefix)), raw[len(prefix):].lstrip("/"), False + return None + + +def _resolve_tsconfig_alias(raw: str, aliases: dict[str, list[str]]) -> "Path | None": + """Resolve `raw` against the most specific matching tsconfig alias pattern. + + Within that pattern, try targets in declared order and return the first whose + candidate resolves to a real file. If none exist, return the first candidate + so existing phantom/external-edge behavior stays unchanged. + """ + best: "tuple[tuple[int, int], str, bool, list[str]] | None" = None + for pattern, targets in aliases.items(): + match = _match_tsconfig_alias(raw, pattern) + if match is None: + continue + specificity, captured, is_wildcard = match + if best is None or specificity < best[0]: + best = specificity, captured, is_wildcard, targets + + if best is None: + return None + + _, captured, is_wildcard, targets = best + first = None + for target in targets: + if is_wildcard: + # TypeScript substitutes only when the matched star is non-empty. + substituted = target.replace("*", captured, 1) if captured else target + cand = Path(os.path.normpath(substituted)) + else: + cand = Path(target) + if captured: + cand = Path(os.path.normpath(cand / captured)) + resolved = _resolve_js_import_path(cand) + if resolved.is_file(): + return resolved + if first is None: + first = cand + return first + + +def _find_workspace_root(start_dir: Path) -> Path | None: + current = start_dir.resolve() + for candidate in [current, *current.parents]: + if (candidate / "pnpm-workspace.yaml").exists(): + return candidate + package_json = candidate / "package.json" + if package_json.is_file(): + try: + data = json.loads(package_json.read_text(encoding="utf-8")) + except Exception: + continue + if "workspaces" in data: + return candidate + return None + + +def _pnpm_workspace_globs(workspace_file: Path) -> list[str]: + globs: list[str] = [] + in_packages = False + for raw_line in workspace_file.read_text(encoding="utf-8", errors="replace").splitlines(): + line = raw_line.strip() + if not line or line.startswith("#"): + continue + if line.startswith("packages:"): + in_packages = True + continue + if in_packages and line.startswith("-"): + value = line[1:].strip().strip("'\"") + if value and not value.startswith("!"): + globs.append(value) + continue + if in_packages and not raw_line.startswith((" ", "\t")): + break + return globs + + +def _workspace_globs(root: Path) -> list[str]: + pnpm_workspace = root / "pnpm-workspace.yaml" + if pnpm_workspace.exists(): + return _pnpm_workspace_globs(pnpm_workspace) + + package_json = root / "package.json" + try: + data = json.loads(package_json.read_text(encoding="utf-8")) + except Exception: + return [] + + workspaces = data.get("workspaces") + if isinstance(workspaces, list): + return [item for item in workspaces if isinstance(item, str) and not item.startswith("!")] + if isinstance(workspaces, dict): + packages = workspaces.get("packages") + if isinstance(packages, list): + return [item for item in packages if isinstance(item, str) and not item.startswith("!")] + return [] + + +def _load_workspace_packages(start_dir: Path) -> dict[str, Path]: + root = _find_workspace_root(start_dir) + if root is None: + return {} + manifest_mtimes = tuple( + (name, (root / name).stat().st_mtime_ns) + for name in _WORKSPACE_MANIFEST_NAMES + if (root / name).is_file() + ) + key = str((root, manifest_mtimes)) + if key in _WORKSPACE_PACKAGE_CACHE: + return _WORKSPACE_PACKAGE_CACHE[key] + + packages: dict[str, Path] = {} + for pattern in _workspace_globs(root): + package_dirs: list[Path] = [root] if pattern in (".", "./") else list(root.glob(pattern)) + for package_dir in package_dirs: + manifest = package_dir / "package.json" + if not manifest.is_file(): + continue + try: + data = json.loads(manifest.read_text(encoding="utf-8")) + except Exception: + continue + name = data.get("name") + if isinstance(name, str) and name: + packages[name] = package_dir + _WORKSPACE_PACKAGE_CACHE[key] = packages + return packages + # Condition keys consulted when resolving an `exports` target, in priority # order. `default` is Node's catch-all and must be consulted LAST so a more # specific condition (source/import/module/etc.) wins when several match. +_EXPORT_CONDITION_PRIORITY = ( + "source", "import", "module", "svelte", "types", "require", "default", +) + + +def _resolve_export_target(value: Any) -> str | None: + """Resolve an `exports` map value (string or condition object) to a + relative target string, honouring _EXPORT_CONDITION_PRIORITY for objects + and recursing into nested condition objects.""" + if isinstance(value, str): + return value + if isinstance(value, dict): + for cond in _EXPORT_CONDITION_PRIORITY: + v = value.get(cond) + if isinstance(v, str): + return v + if isinstance(v, dict): + nested = _resolve_export_target(v) + if nested: + return nested + return None + + +def _contained_in_package(resolved: Path, package_dir: Path) -> bool: + """Guard against `exports` targets that escape the package directory + (e.g. "./evil": "../../../etc/passwd"). Only accept paths that stay + within package_dir after resolution.""" + try: + return resolved.resolve().is_relative_to(package_dir.resolve()) + except ValueError: + return False + + +def _package_entry_candidates(package_dir: Path, subpath: str) -> list[Path]: + manifest = package_dir / "package.json" + manifest_data: dict[str, Any] = {} + try: + manifest_data = json.loads(manifest.read_text(encoding="utf-8")) + except Exception: + pass + + if subpath: + # Consult the package's `exports` subpath map before the bare-path + # fallback (#1308): "./browser" -> conditions -> file, plus single + # wildcard "./*" patterns. Targets that escape the package dir are + # rejected; resolution then falls through to the bare path. + exports = manifest_data.get("exports") + if isinstance(exports, dict): + subpath_key = "./" + subpath + target = _resolve_export_target(exports.get(subpath_key)) + if target: + candidate = package_dir / target + if _contained_in_package(candidate, package_dir): + return [candidate] + else: + for pattern, pattern_value in exports.items(): + if "*" in pattern and pattern.count("*") == 1: + prefix, suffix = pattern.split("*", 1) + if (subpath_key.startswith(prefix) + and (not suffix or subpath_key.endswith(suffix))): + matched = subpath_key[len(prefix):len(subpath_key) - len(suffix) if suffix else None] + resolved = _resolve_export_target(pattern_value) + if resolved and "*" in resolved: + candidate = package_dir / resolved.replace("*", matched) + if _contained_in_package(candidate, package_dir): + return [candidate] + return [package_dir / subpath] + + exports = manifest_data.get("exports") + if isinstance(exports, str): + return [package_dir / exports] + if isinstance(exports, dict): + dot_target = _resolve_export_target(exports.get(".")) + if dot_target: + return [package_dir / dot_target] + + candidates: list[Path] = [] + for key in ("svelte", "module", "main", "types"): + value = manifest_data.get(key) + if isinstance(value, str): + candidates.append(package_dir / value) + candidates.append(package_dir / "src/index") + candidates.append(package_dir / "index") + return candidates + + +def _resolve_workspace_import(raw: str, start_dir: Path) -> Path | None: + packages = _load_workspace_packages(start_dir) + for package_name, package_dir in packages.items(): + if raw == package_name: + subpath = "" + elif raw.startswith(package_name + "/"): + subpath = raw[len(package_name) + 1:] + else: + continue + for candidate in _package_entry_candidates(package_dir, subpath): + resolved = _resolve_js_import_path(candidate) + if resolved.is_file(): + return resolved + return None + + +def _resolve_js_module_path(raw: str | Path, start_dir: Path | None = None) -> Path | None: + """Resolve a JS/TS module path or specifier to a local source file. + + With a Path argument this preserves the path-based helper API used by + import-extension tests. With a string plus start_dir it resolves JS/TS + module specifiers including relative paths, tsconfig aliases, and workspace + packages. + """ + if isinstance(raw, Path): + return _resolve_js_import_path(raw) + if start_dir is None: + return _resolve_js_import_path(Path(raw)) + if raw.startswith("."): + return _resolve_js_import_path(start_dir / raw) + + aliases = _load_tsconfig_aliases(start_dir) + hit = _resolve_tsconfig_alias(raw, aliases) + if hit is not None: + return _resolve_js_import_path(hit) + + return _resolve_workspace_import(raw, start_dir) # ── LanguageConfig dataclass ───────────────────────────────────────────────── +@dataclass +class LanguageConfig: + ts_module: str # e.g. "tree_sitter_python" + ts_language_fn: str = "language" # attr to call: e.g. tslang.language() + + class_types: frozenset = frozenset() + function_types: frozenset = frozenset() + import_types: frozenset = frozenset() + call_types: frozenset = frozenset() + static_prop_types: frozenset = frozenset() + helper_fn_names: frozenset = frozenset() + container_bind_methods: frozenset = frozenset() + event_listener_properties: frozenset = frozenset() + + # Name extraction + name_field: str = "name" + name_fallback_child_types: tuple = () + + # Body detection + body_field: str = "body" + body_fallback_child_types: tuple = () # e.g. ("declaration_list", "compound_statement") + + # Call name extraction + call_function_field: str = "function" # field on call node for callee + call_accessor_node_types: frozenset = frozenset() # member/attribute nodes + call_accessor_field: str = "attribute" # field on accessor for method name + call_accessor_object_field: str = "" # field on accessor for the receiver/object + + # Stop recursion at these types in walk_calls + function_boundary_types: frozenset = frozenset() + + # Import handler: called for import nodes instead of generic handling + import_handler: Callable | None = None + + # Optional custom name resolver for functions (C, C++ declarator unwrapping) + resolve_function_name_fn: Callable | None = None + + # Extra label formatting for functions: if True, functions get "name()" label + function_label_parens: bool = True + + # Extra walk hook called after generic dispatch (for JS arrow functions, C# namespaces, etc.) + extra_walk_fn: Callable | None = None + # ── Generic helpers ─────────────────────────────────────────────────────────── + +_PYTHON_TYPE_CONTAINERS = frozenset({ + "list", "dict", "set", "tuple", "frozenset", "type", + "List", "Dict", "Set", "Tuple", "FrozenSet", "Type", + "Optional", "Union", "Sequence", "Iterable", "Mapping", "MutableMapping", + "Iterator", "Callable", "Awaitable", "AsyncIterable", "AsyncIterator", "Coroutine", + "Generator", "AsyncGenerator", "ContextManager", "AsyncContextManager", + "Annotated", "ClassVar", "Final", "Literal", "Concatenate", "ParamSpec", "TypeVar", + "None", "Ellipsis", +}) + # Scalar builtins and test-mock names that appear as type annotations but carry # no useful semantic meaning as graph nodes (#1147). Suppressed at the annotation # walker level so they are never created as nodes or emitted as edges. +_PYTHON_ANNOTATION_NOISE = frozenset({ + # scalar builtins + "str", "int", "float", "bool", "bytes", "bytearray", "complex", "object", + "True", "False", + # unittest.mock + "MagicMock", "Mock", "AsyncMock", "NonCallableMock", + "NonCallableMagicMock", "PropertyMock", "patch", "sentinel", +}) + + +def _python_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[str, str]]) -> None: + """Walk a Python type annotation; append (name, role) where role is 'type' or 'generic_arg'. + + Builtin/typing containers (list, dict, Optional, Union, …) are not emitted as refs themselves, + but their nested type arguments still count as generic_arg. + """ + if node is None: + return + t = node.type + if t == "type": + for c in node.children: + if c.is_named: + _python_collect_type_refs(c, source, generic, out) + return + if t == "identifier": + name = _read_text(node, source) + if name and name not in _PYTHON_TYPE_CONTAINERS and name not in _PYTHON_ANNOTATION_NOISE: + out.append((name, "generic_arg" if generic else "type")) + return + if t == "attribute": + tail = _read_text(node, source).rsplit(".", 1)[-1] + if tail and tail not in _PYTHON_TYPE_CONTAINERS and tail not in _PYTHON_ANNOTATION_NOISE: + out.append((tail, "generic_arg" if generic else "type")) + return + if t == "generic_type": + for c in node.children: + if c.type == "identifier": + container = _read_text(c, source) + if container and container not in _PYTHON_TYPE_CONTAINERS and container not in _PYTHON_ANNOTATION_NOISE: + out.append((container, "generic_arg" if generic else "type")) + elif c.type == "type_parameter": + for sub in c.children: + if sub.is_named: + _python_collect_type_refs(sub, source, True, out) + return + if t == "subscript": + value = node.child_by_field_name("value") + if value is not None: + _python_collect_type_refs(value, source, generic, out) + for c in node.children: + if c is value or not c.is_named: + continue + _python_collect_type_refs(c, source, True, out) + return + if node.is_named: + for c in node.children: + if c.is_named: + _python_collect_type_refs(c, source, generic, out) + + +def _csharp_pre_scan_interfaces(root_node, source: bytes) -> set[str]: + """Return names declared as `interface` in this C# compilation unit.""" + out: set[str] = set() + stack = [root_node] + while stack: + n = stack.pop() + if n.type == "interface_declaration": + name_node = n.child_by_field_name("name") + if name_node is not None: + text = _read_text(name_node, source) + if text: + out.add(text) + stack.extend(n.children) + return out + + +def _csharp_classify_base(name: str, interface_names: set[str]) -> str: + """`implements` if the base name is an interface (declared or by I-prefix convention), else `inherits`.""" + if name in interface_names: + return "implements" + if len(name) >= 2 and name[0] == "I" and name[1].isupper(): + return "implements" + return "inherits" + + +_CSHARP_TYPE_PARAMETER_SCOPE_DECLARATIONS = frozenset({ + "class_declaration", + "interface_declaration", + "record_declaration", + "struct_declaration", + "method_declaration", +}) + + +def _csharp_type_parameters_in_scope(node, source: bytes) -> frozenset[str]: + """Return C# type-parameter names visible from ``node``.""" + names: set[str] = set() + scope = node + while scope is not None: + if scope.type in _CSHARP_TYPE_PARAMETER_SCOPE_DECLARATIONS: + for child in scope.children: + if child.type != "type_parameter_list": + continue + for param in child.children: + if param.type == "type_parameter": + name_node = next( + (sub for sub in param.children if sub.type == "identifier"), + None, + ) + if name_node is not None: + name = _read_text(name_node, source) + if name: + names.add(name) + elif param.type == "identifier": + name = _read_text(param, source) + if name: + names.add(name) + scope = scope.parent + return frozenset(names) + + +def _csharp_collect_type_refs( + node, + source: bytes, + generic: bool, + out: list[tuple[str, str, bool, str]], + skip: frozenset[str] | None = None, +) -> None: + """Walk a C# type expression; append (name, role, qualified, qualifier) tuples.""" + if node is None: + return + if skip is None: + skip = _csharp_type_parameters_in_scope(node, source) + t = node.type + if t == "predefined_type": + return + if t == "identifier": + name = _read_text(node, source) + if name and name not in skip: + out.append((name, "generic_arg" if generic else "type", False, "")) + return + if t == "qualified_name": + prefix, _, text = _read_text(node, source).rpartition(".") + text = text.split("<", 1)[0] + if text and text not in skip: + out.append((text, "generic_arg" if generic else "type", True, prefix)) + return + if t == "generic_name": + name_child = node.child_by_field_name("name") + if name_child is None: + for sub in node.children: + if sub.type == "identifier": + name_child = sub + break + if name_child is not None: + qualified = name_child.type == "qualified_name" + prefix, _, name = _read_text(name_child, source).rpartition(".") + if name and name not in skip: + out.append((name, "generic_arg" if generic else "type", qualified, prefix if qualified else "")) + for sub in node.children: + if sub.type == "type_argument_list": + for arg in sub.children: + if arg.is_named: + _csharp_collect_type_refs(arg, source, True, out, skip) + return + if t in ("nullable_type", "array_type", "pointer_type", "ref_type"): + for c in node.children: + if c.is_named: + _csharp_collect_type_refs(c, source, generic, out, skip) + return + if node.is_named: + for c in node.children: + if c.is_named: + _csharp_collect_type_refs(c, source, generic, out, skip) + + +def _csharp_attribute_names(method_node, source: bytes) -> list[tuple[str, bool, str]]: + """Collect attribute names from a C# method/declaration's attribute_list children.""" + names: list[tuple[str, bool, str]] = [] + skip = _csharp_type_parameters_in_scope(method_node, source) + for child in method_node.children: + if child.type != "attribute_list": + continue + for attr in child.children: + if attr.type != "attribute": + continue + name_node = attr.child_by_field_name("name") + if name_node is None: + for sub in attr.children: + if sub.type in ("identifier", "qualified_name"): + name_node = sub + break + if name_node is not None: + qualified = name_node.type == "qualified_name" + prefix, _, text = _read_text(name_node, source).rpartition(".") + if text and text not in skip: + names.append((text, qualified, prefix if qualified else "")) + return names + + +_JAVA_TYPE_PARAMETER_SCOPE_DECLARATIONS = frozenset({ + "class_declaration", + "interface_declaration", + "record_declaration", + "method_declaration", + "constructor_declaration", +}) + + +def _java_type_parameters_in_scope(node, source: bytes) -> frozenset[str]: + """Return Java type-parameter names visible from ``node``.""" + names: set[str] = set() + scope = node + while scope is not None: + if scope.type in _JAVA_TYPE_PARAMETER_SCOPE_DECLARATIONS: + params = scope.child_by_field_name("type_parameters") + if params is not None: + for param in params.children: + if param.type != "type_parameter": + continue + name_node = next( + (child for child in param.children if child.type == "type_identifier"), + None, + ) + if name_node is not None: + names.add(_read_text(name_node, source)) + scope = scope.parent + return frozenset(names) # java.lang (auto-imported) plus the ubiquitous java.util / java.io / java.time / @@ -207,80 +876,986 @@ def _file_node_id(rel_path: Path) -> str: # type-ref walker so they are never created as nodes or emitted as edges. The # boxed-scalar/`void` primitives are already dropped by grammar node type above; # these are the class/interface names the grammar reports as identifiers. +_JAVA_BUILTIN_TYPES = frozenset({ + # java.lang — core + "Object", "String", "CharSequence", "StringBuilder", "StringBuffer", + "Number", "Byte", "Short", "Integer", "Long", "Float", "Double", + "Boolean", "Character", "Void", "Class", "Enum", "Record", "Math", + "System", "Thread", "Runnable", "Comparable", "Iterable", "Cloneable", + "AutoCloseable", "Appendable", "Readable", "Process", "ProcessBuilder", + "Runtime", "Package", "ThreadLocal", "InheritableThreadLocal", + # java.lang — throwables + "Throwable", "Exception", "RuntimeException", "Error", + "IllegalArgumentException", "IllegalStateException", "NullPointerException", + "IndexOutOfBoundsException", "ArrayIndexOutOfBoundsException", + "ClassCastException", "NumberFormatException", "ArithmeticException", + "UnsupportedOperationException", "InterruptedException", + "CloneNotSupportedException", "SecurityException", "StackOverflowError", + "OutOfMemoryError", "AssertionError", + # java.util — collections & core + "Collection", "List", "ArrayList", "LinkedList", "Vector", "Stack", + "Set", "HashSet", "LinkedHashSet", "TreeSet", "SortedSet", "NavigableSet", + "EnumSet", "Map", "HashMap", "LinkedHashMap", "TreeMap", "SortedMap", + "NavigableMap", "Hashtable", "EnumMap", "Properties", "Queue", "Deque", + "ArrayDeque", "PriorityQueue", "Iterator", "ListIterator", "Comparator", + "Optional", "OptionalInt", "OptionalLong", "OptionalDouble", "Collections", + "Arrays", "Objects", "Date", "Calendar", "Random", "UUID", "Scanner", + "StringJoiner", "StringTokenizer", "BitSet", "Spliterator", "Locale", + "NoSuchElementException", "ConcurrentModificationException", + # java.util.stream + "Stream", "IntStream", "LongStream", "DoubleStream", "Collector", + "Collectors", + # java.util.function + "Function", "BiFunction", "Consumer", "BiConsumer", "Supplier", + "Predicate", "BiPredicate", "UnaryOperator", "BinaryOperator", + "IntFunction", "ToIntFunction", "ToLongFunction", "ToDoubleFunction", + # java.util.concurrent + "Callable", "Future", "CompletableFuture", "CompletionStage", "Executor", + "ExecutorService", "Executors", "ScheduledExecutorService", "TimeUnit", + "ConcurrentHashMap", "ConcurrentMap", "CopyOnWriteArrayList", + "BlockingQueue", "CountDownLatch", "Semaphore", "CyclicBarrier", + "AtomicInteger", "AtomicLong", "AtomicBoolean", "AtomicReference", + # java.time + "Instant", "Duration", "Period", "LocalDate", "LocalTime", "LocalDateTime", + "ZonedDateTime", "OffsetDateTime", "ZoneId", "ZoneOffset", "DayOfWeek", + "Month", "Year", "Clock", "DateTimeFormatter", + # java.io / java.nio.file + "IOException", "UncheckedIOException", "FileNotFoundException", "File", + "InputStream", "OutputStream", "Reader", "Writer", "BufferedReader", + "BufferedWriter", "InputStreamReader", "OutputStreamWriter", "FileReader", + "FileWriter", "PrintStream", "PrintWriter", "ByteArrayInputStream", + "ByteArrayOutputStream", "Serializable", "Closeable", "Path", "Paths", + "Files", + # java.math + "BigDecimal", "BigInteger", +}) -# ── C / C++ type-ref helpers ───────────────────────────────────────────────── +def _java_collect_type_refs( + node, + source: bytes, + generic: bool, + out: list[tuple[str, str]], + skip: frozenset[str] | None = None, +) -> None: + """Walk a Java type expression; append (name, role) tuples.""" + if node is None: + return + if skip is None: + skip = _java_type_parameters_in_scope(node, source) + t = node.type + if t in ("integral_type", "floating_point_type", "boolean_type", "void_type"): + return + if t == "type_identifier": + name = _read_text(node, source) + if name and name not in skip and name not in _JAVA_BUILTIN_TYPES: + out.append((name, "generic_arg" if generic else "type")) + return + if t == "scoped_type_identifier": + text = _read_text(node, source).rsplit(".", 1)[-1] + if text and text not in _JAVA_BUILTIN_TYPES: + out.append((text, "generic_arg" if generic else "type")) + return + if t == "generic_type": + for c in node.children: + if c.type in ("type_identifier", "scoped_type_identifier"): + text = _read_text(c, source).rsplit(".", 1)[-1] + if ( + text + and text not in _JAVA_BUILTIN_TYPES + and (c.type == "scoped_type_identifier" or text not in skip) + ): + out.append((text, "generic_arg" if generic else "type")) + break + for c in node.children: + if c.type == "type_arguments": + for arg in c.children: + if arg.is_named: + _java_collect_type_refs(arg, source, True, out, skip) + return + if t == "array_type": + for c in node.children: + if c.is_named: + _java_collect_type_refs(c, source, generic, out, skip) + return + if node.is_named: + for c in node.children: + if c.is_named: + _java_collect_type_refs(c, source, generic, out, skip) -# ── Scala type-ref helpers ─────────────────────────────────────────────────── +def _java_annotation_names(declaration_node, source: bytes) -> list[str]: + """Collect annotation names from a Java declaration's `modifiers` child.""" + names: list[str] = [] + modifiers = None + for child in declaration_node.children: + if child.type == "modifiers": + modifiers = child + break + if modifiers is None: + return names + for anno in modifiers.children: + if anno.type not in ("marker_annotation", "annotation"): + continue + name_node = anno.child_by_field_name("name") + if name_node is None: + for sub in anno.children: + if sub.type in ("identifier", "scoped_identifier", "type_identifier"): + name_node = sub + break + if name_node is not None: + text = _read_text(name_node, source).rsplit(".", 1)[-1] + if text: + names.append(text) + return names -def _resolve_name(node, source: bytes, config: LanguageConfig) -> str | None: - """Get the name from a node using config.name_field, falling back to child types.""" - if config.resolve_function_name_fn is not None: - # For C/C++ where the name is inside a declarator - return None # caller handles this separately - n = node.child_by_field_name(config.name_field) - if n: - return _read_text(n, source) - for child in node.children: - if child.type in config.name_fallback_child_types: - return _read_text(child, source) - return None +_GO_PREDECLARED_TYPES = frozenset({ + "bool", "byte", "complex64", "complex128", "error", "float32", "float64", + "int", "int8", "int16", "int32", "int64", "rune", "string", + "uint", "uint8", "uint16", "uint32", "uint64", "uintptr", "any", "comparable", +}) -# ── Import handlers ─────────────────────────────────────────────────────────── +def _go_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[str, str]]) -> None: + """Walk a Go type expression; append (name, role) tuples.""" + if node is None: + return + t = node.type + if t == "type_identifier": + text = _read_text(node, source) + if text and text not in _GO_PREDECLARED_TYPES: + out.append((text, "generic_arg" if generic else "type")) + return + if t == "qualified_type": + text = _read_text(node, source).rsplit(".", 1)[-1] + if text and text not in _GO_PREDECLARED_TYPES: + out.append((text, "generic_arg" if generic else "type")) + return + if t == "generic_type": + type_field = node.child_by_field_name("type") + if type_field is not None: + sub: list[tuple[str, str]] = [] + _go_collect_type_refs(type_field, source, generic, sub) + out.extend(sub) + for c in node.children: + if c.type == "type_arguments": + for arg in c.children: + if arg.is_named: + _go_collect_type_refs(arg, source, True, out) + return + if t in ("pointer_type", "slice_type", "array_type", "map_type", + "channel_type", "parenthesized_type"): + for c in node.children: + if c.is_named: + _go_collect_type_refs(c, source, generic, out) + return + if node.is_named: + for c in node.children: + if c.is_named: + _go_collect_type_refs(c, source, generic, out) -def _import_python(node, source: bytes, file_nid: str, stem: str, edges: list, str_path: str, scope_stack: list[str] | None = None) -> None: + +def _rust_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[str, str]]) -> None: + """Walk a Rust type expression; append (name, role) tuples.""" + if node is None: + return t = node.type - if t == "import_statement": - for child in node.children: - if child.type in ("dotted_name", "aliased_import"): - raw = _read_text(child, source) - module_name = raw.split(" as ")[0].strip().lstrip(".") - tgt_nid = _make_id(module_name) - edges.append({ - "source": file_nid, - "target": tgt_nid, - "relation": "imports", - "context": "import", - "confidence": "EXTRACTED", - "source_file": str_path, - "source_location": f"L{node.start_point[0] + 1}", - "weight": 1.0, - }) - elif t == "import_from_statement": - module_node = node.child_by_field_name("module_name") - if module_node: - raw = _read_text(module_node, source) - if raw.startswith("."): - # Relative import - resolve to full path so IDs match file node IDs - dots = len(raw) - len(raw.lstrip(".")) - module_name = raw.lstrip(".") - base = Path(str_path).parent - for _ in range(dots - 1): - base = base.parent - rel = (module_name.replace(".", "/") + ".py") if module_name else "__init__.py" - tgt_nid = _make_id(str(base / rel)) - else: - tgt_nid = _make_id(raw) - edges.append({ - "source": file_nid, - "target": tgt_nid, - "relation": "imports_from", - "context": "import", - "confidence": "EXTRACTED", - "source_file": str_path, - "source_location": f"L{node.start_point[0] + 1}", - "weight": 1.0, - }) + if t == "primitive_type": + return + if t == "type_identifier": + text = _read_text(node, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + return + if t == "scoped_type_identifier": + text = _read_text(node, source).rsplit("::", 1)[-1] + if text: + out.append((text, "generic_arg" if generic else "type")) + return + if t == "generic_type": + name_node = node.child_by_field_name("type") + if name_node is None: + for c in node.children: + if c.type in ("type_identifier", "scoped_type_identifier"): + name_node = c + break + if name_node is not None: + text = _read_text(name_node, source).rsplit("::", 1)[-1] + if text: + out.append((text, "generic_arg" if generic else "type")) + for c in node.children: + if c.type == "type_arguments": + for arg in c.children: + if arg.is_named: + _rust_collect_type_refs(arg, source, True, out) + return + if t in ("reference_type", "pointer_type", "array_type", "tuple_type", "slice_type"): + for c in node.children: + if c.is_named: + _rust_collect_type_refs(c, source, generic, out) + return + if node.is_named: + for c in node.children: + if c.is_named: + _rust_collect_type_refs(c, source, generic, out) -def _import_js(node, source: bytes, file_nid: str, stem: str, edges: list, str_path: str, scope_stack: list[str] | None = None) -> None: - is_reexport = node.type == "export_statement" - # Only handle export_statement if it has a `from` clause (re-export). - # Pure exports like `export const x = 1` or `export { localVar }` have no source module. - if is_reexport: +def _php_name_text(node, source: bytes) -> str | None: + """Return the unqualified name text from a PHP `name`/`qualified_name` node.""" + if node is None: + return None + return _read_text(node, source).rsplit("\\", 1)[-1] or None + + +def _php_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[str, str]]) -> None: + """Walk a PHP type expression; append (name, role) tuples.""" + if node is None: + return + t = node.type + if t == "primitive_type": + return + if t == "named_type": + for c in node.children: + if c.type in ("name", "qualified_name"): + text = _php_name_text(c, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + return + return + if t in ("name", "qualified_name"): + text = _php_name_text(node, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + return + if t in ("nullable_type", "union_type", "intersection_type", "optional_type"): + for c in node.children: + if c.is_named: + _php_collect_type_refs(c, source, generic, out) + return + if node.is_named: + for c in node.children: + if c.is_named: + _php_collect_type_refs(c, source, generic, out) + + +def _php_method_return_type_node(method_node): + """Return the named_type/primitive_type node sitting after formal_parameters.""" + saw_params = False + for c in method_node.children: + if c.type == "formal_parameters": + saw_params = True + continue + if saw_params and c.is_named and c.type not in ("compound_statement",): + if c.type in ("named_type", "primitive_type", "nullable_type", + "union_type", "intersection_type", "optional_type"): + return c + return None + + +def _kotlin_user_type_name(user_type_node, source: bytes) -> str | None: + """Return the head identifier text from a Kotlin user_type node (without generics).""" + if user_type_node is None: + return None + for c in user_type_node.children: + if c.type == "type_identifier": + text = _read_text(c, source) + return text or None + if c.type == "identifier": + text = _read_text(c, source) + return text or None + if c.type == "simple_user_type": + for sub in c.children: + if sub.type in ("identifier", "type_identifier"): + text = _read_text(sub, source) + return text or None + return None + + +def _kotlin_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[str, str]]) -> None: + """Walk a Kotlin type expression; append (name, role) tuples.""" + if node is None: + return + t = node.type + if t in ("integral_literal", "boolean_literal"): + return + if t == "user_type": + for c in node.children: + if c.type in ("identifier", "type_identifier"): + text = _read_text(c, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + break + if c.type == "simple_user_type": + for sub in c.children: + if sub.type in ("identifier", "type_identifier"): + text = _read_text(sub, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + break + break + for c in node.children: + if c.type == "type_arguments": + for arg in c.children: + if arg.type == "type_projection": + for sub in arg.children: + if sub.is_named: + _kotlin_collect_type_refs(sub, source, True, out) + elif arg.is_named: + _kotlin_collect_type_refs(arg, source, True, out) + return + if t in ("identifier", "type_identifier"): + text = _read_text(node, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + return + if t in ("nullable_type", "parenthesized_type", "type_reference"): + for c in node.children: + if c.is_named: + _kotlin_collect_type_refs(c, source, generic, out) + return + if node.is_named: + for c in node.children: + if c.is_named: + _kotlin_collect_type_refs(c, source, generic, out) + + +def _kotlin_property_type_node(property_node): + """Find the user_type node within a Kotlin property_declaration.""" + for c in property_node.children: + if c.type == "variable_declaration": + for sub in c.children: + if sub.type in ("user_type", "nullable_type", "type_reference"): + return sub + if c.type in ("user_type", "nullable_type", "type_reference"): + return c + return None + + +def _kotlin_function_return_type_node(func_node): + """Find the return-type node of a Kotlin function_declaration (the type after `: ` post-params).""" + saw_params = False + saw_colon = False + for c in func_node.children: + if c.type == "function_value_parameters": + saw_params = True + continue + if saw_params and c.type == ":": + saw_colon = True + continue + if saw_colon: + if c.is_named: + return c + return None + + +def _swift_declaration_keyword(node) -> str | None: + """Return the leading kind token for a Swift class_declaration: class/struct/enum/extension/actor.""" + for c in node.children: + if not c.is_named and c.type in ("class", "struct", "enum", "extension", "actor"): + return c.type + return None + + +def _swift_pre_scan(root_node, source: bytes) -> tuple[set[str], set[str]]: + """Pre-scan a Swift compilation unit and return (protocol_names, class_like_names).""" + protocols: set[str] = set() + classes: set[str] = set() + stack = [root_node] + while stack: + n = stack.pop() + if n.type == "protocol_declaration": + name_node = n.child_by_field_name("name") + if name_node is None: + for c in n.children: + if c.type == "type_identifier": + name_node = c + break + if name_node is not None: + text = _read_text(name_node, source) + if text: + protocols.add(text) + elif n.type == "class_declaration": + kw = _swift_declaration_keyword(n) + if kw in ("class", "struct", "enum", "actor"): + name_node = n.child_by_field_name("name") + if name_node is not None: + text = _read_text(name_node, source) + if text: + classes.add(text) + stack.extend(n.children) + return protocols, classes + + +def _swift_classify_base(name: str, kind: str | None, is_first: bool, + protocols: set[str], classes: set[str]) -> str: + """Classify a Swift inheritance_specifier entry as `inherits` or `implements`.""" + if name in protocols: + return "implements" + if name in classes: + return "inherits" + # struct/enum/extension/actor cannot inherit a class — all conformances are protocols. + if kind in ("struct", "enum", "extension", "actor"): + return "implements" + # `class`: first entry is conventionally the base class; subsequent are protocols. + return "inherits" if is_first else "implements" + + +def _swift_user_type_name(user_type_node, source: bytes) -> str | None: + """Return the head type_identifier text from a Swift user_type node (without generics).""" + if user_type_node is None: + return None + for c in user_type_node.children: + if c.type == "type_identifier": + text = _read_text(c, source) + return text or None + return None + + +def _swift_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[str, str]]) -> None: + """Walk a Swift type expression; append (name, role) tuples (role 'type' or 'generic_arg').""" + if node is None: + return + t = node.type + if t == "type_annotation": + for c in node.children: + if c.is_named: + _swift_collect_type_refs(c, source, generic, out) + return + if t == "user_type": + for c in node.children: + if c.type == "type_identifier": + text = _read_text(c, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + break + for c in node.children: + if c.type == "type_arguments": + for arg in c.children: + if arg.is_named: + _swift_collect_type_refs(arg, source, True, out) + return + if t == "type_identifier": + text = _read_text(node, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + return + if t in ("optional_type", "implicitly_unwrapped_optional_type", "array_type", + "dictionary_type", "tuple_type"): + for c in node.children: + if c.is_named: + _swift_collect_type_refs(c, source, generic, out) + return + if node.is_named: + for c in node.children: + if c.is_named: + _swift_collect_type_refs(c, source, generic, out) + + +def _swift_property_type_node(property_node): + """Return the type_annotation child of a Swift property_declaration, if any.""" + for c in property_node.children: + if c.type == "type_annotation": + return c + return None + + +def _swift_property_name(property_node, source: bytes) -> str | None: + """Return the bound name of a Swift property (``let x``/``var x = ...``).""" + for c in property_node.children: + if c.type == "pattern": + for sc in c.children: + if sc.type == "simple_identifier": + return _read_text(sc, source) + if c.type == "simple_identifier": + return _read_text(c, source) + return None + + +def _swift_constructor_type(call_node, source: bytes) -> str | None: + """If a Swift call expression is a constructor (``Foo()``), return the type name. + + Only upper-cased callees are treated as types so a free-function call like + ``configure()`` in an initializer is not mistaken for a constructor. + """ + first = call_node.children[0] if call_node.children else None + if first is not None and first.type == "simple_identifier": + text = _read_text(first, source) + if text and text[:1].isupper(): + return text + return None + + +def _swift_receiver_name(recv_node, source: bytes) -> str | None: + """Return the depth-1 receiver name of a Swift member call (``recv.method()``). + + ``vm.update()`` -> ``vm``; ``Type.staticMethod()`` -> ``Type``; + ``Singleton.shared.method()`` -> ``Singleton`` (head of the chain); + ``self.svc.fetch()`` -> ``svc`` (the property the call is reached through). + Returns None for anything deeper, so resolution stays depth-1. + """ + if recv_node is None: + return None + if recv_node.type == "simple_identifier": + return _read_text(recv_node, source) + if recv_node.type == "navigation_expression": + head = recv_node.children[0] if recv_node.children else None + if head is not None and head.type == "simple_identifier": + return _read_text(head, source) + if head is not None and head.type == "self_expression": + for child in recv_node.children: + if child.type == "navigation_suffix": + for sc in child.children: + if sc.type == "simple_identifier": + return _read_text(sc, source) + return None + + +# ── C / C++ type-ref helpers ───────────────────────────────────────────────── + +_C_PRIMITIVE_TYPE_NODES = frozenset({ + "primitive_type", "sized_type_specifier", "auto", "placeholder_type_specifier", +}) + + +def _c_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[str, str]]) -> None: + """Walk a C type expression; append (name, role) tuples for user-defined types. + Skips primitive types and qualifiers; recognises type_identifier.""" + if node is None or node.type in _C_PRIMITIVE_TYPE_NODES: + return + t = node.type + if t == "type_identifier": + text = _read_text(node, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + return + if t in ("pointer_declarator", "reference_declarator", "array_declarator", + "type_qualifier", "type_descriptor", "abstract_pointer_declarator", + "abstract_reference_declarator", "abstract_array_declarator"): + for c in node.children: + if c.is_named: + _c_collect_type_refs(c, source, generic, out) + + +def _cpp_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[str, str]]) -> None: + """Walk a C++ type expression; append (name, role) tuples. + Resolves qualified_identifier tails (std::string → string) and template_type + base + arguments (std::vector → vector + HttpClient as generic_arg).""" + if node is None or node.type in _C_PRIMITIVE_TYPE_NODES: + return + t = node.type + if t == "type_identifier": + text = _read_text(node, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + return + if t == "qualified_identifier": + name_node = node.child_by_field_name("name") + if name_node is not None: + _cpp_collect_type_refs(name_node, source, generic, out) + return + if t == "template_type": + name_node = node.child_by_field_name("name") + if name_node is not None: + text = _read_text(name_node, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + args_node = node.child_by_field_name("arguments") + if args_node is not None: + for c in args_node.children: + if c.is_named: + _cpp_collect_type_refs(c, source, True, out) + return + if t in ("type_descriptor", "pointer_declarator", "reference_declarator", + "array_declarator", "type_qualifier", "abstract_pointer_declarator", + "abstract_reference_declarator", "abstract_array_declarator"): + for c in node.children: + if c.is_named: + _cpp_collect_type_refs(c, source, generic, out) + + +# ── Scala type-ref helpers ─────────────────────────────────────────────────── + +def _scala_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[str, str]]) -> None: + """Walk a Scala type expression; append (name, role) tuples. + Handles type_identifier, generic_type (List[T]), and common type wrappers.""" + if node is None: + return + t = node.type + if t == "type_identifier": + text = _read_text(node, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + return + if t == "generic_type": + base = node.child_by_field_name("type") + if base is None: + for c in node.children: + if c.type == "type_identifier": + base = c + break + if base is not None and base.type == "type_identifier": + text = _read_text(base, source) + if text: + out.append((text, "generic_arg" if generic else "type")) + for c in node.children: + if c.type == "type_arguments": + for arg in c.children: + if arg.is_named: + _scala_collect_type_refs(arg, source, True, out) + return + if t in ("compound_type", "infix_type", "function_type", "tuple_type", + "annotated_type", "projected_type"): + for c in node.children: + if c.is_named: + _scala_collect_type_refs(c, source, generic, out) + + +def _python_collect_param_refs(params_node, source: bytes) -> list[tuple[str, str]]: + """Collect type refs from each typed parameter under a `parameters` node.""" + out: list[tuple[str, str]] = [] + if params_node is None: + return out + for child in params_node.children: + if child.type in ("typed_parameter", "typed_default_parameter"): + type_node = child.child_by_field_name("type") + _python_collect_type_refs(type_node, source, False, out) + return out + + +def _python_param_names(params_node, source: bytes) -> set[str]: + """Plain parameter identifiers declared on a Python `parameters` node. + + Covers positional/keyword params plus `*args` / `**kwargs` and typed or + default forms — anything that binds a local name the function body can shadow + a module-level definition with. + """ + out: set[str] = set() + if params_node is None: + return out + for child in params_node.children: + if child.type == "identifier": + out.add(_read_text(child, source)) + elif child.type in ( + "typed_parameter", + "default_parameter", + "typed_default_parameter", + "list_splat_pattern", + "dictionary_splat_pattern", + ): + # The bound name is the first identifier child (the rest is type/default). + name_n = child.child_by_field_name("name") + if name_n is None: + name_n = next( + (c for c in child.children if c.type == "identifier"), None + ) + if name_n is not None: + out.add(_read_text(name_n, source)) + return out + + +def _python_collect_assignment_targets(node, source: bytes, out: set[str]) -> None: + """Identifiers bound as `pattern` targets under a Python AST subtree. + + Recurses through `pattern_list` / `tuple_pattern` / `list_pattern` so tuple + unpacking (`a, b = ...`, `for a, b in ...`) contributes every bound name. + """ + if node is None: + return + if node.type == "identifier": + out.add(_read_text(node, source)) + return + if node.type in ("pattern_list", "tuple_pattern", "list_pattern"): + for c in node.children: + _python_collect_assignment_targets(c, source, out) + + +def _python_local_bound_names(func_def_node, source: bytes) -> set[str]: + """Names bound LOCALLY inside a Python function: parameters plus assignment, + `for`, `with ... as`, and comprehension targets. + + Used by the indirect-dispatch guard to reject a call-argument identifier that + is a parameter or a local binding — it names a local value, not the module- + level function/class that happens to share the name. Nested `function_definition` + and `class_definition` subtrees are NOT descended into: their bindings belong + to a different scope. + """ + bound: set[str] = set() + bound |= _python_param_names(func_def_node.child_by_field_name("parameters"), source) + + def walk(n) -> None: + for child in n.children: + t = child.type + if t in ("function_definition", "class_definition", "lambda"): + continue # inner scope — its bindings are not this function's locals + if t == "assignment": + _python_collect_assignment_targets( + child.child_by_field_name("left"), source, bound + ) + elif t in ("for_statement", "for_in_clause"): + _python_collect_assignment_targets( + child.child_by_field_name("left"), source, bound + ) + elif t == "with_statement": + for item in child.children: + if item.type == "with_clause": + for wi in item.children: + if wi.type == "with_item": + alias = wi.child_by_field_name("alias") + _python_collect_assignment_targets(alias, source, bound) + elif t == "named_expression": # walrus := + _python_collect_assignment_targets( + child.child_by_field_name("name"), source, bound + ) + walk(child) + + body = func_def_node.child_by_field_name("body") + if body is not None: + walk(body) + return bound + + +def _python_module_bound_names(root, source: bytes) -> set[str]: + """Names rebound by assignment at MODULE scope (top-level `x = ...`, `for`, walrus). + + The module-scope analogue of the per-function shadow set: a dispatch-table value + whose name is reassigned to data at module level (`handler = build()`) names that + value, not a same-named function, so it must not manufacture an indirect edge. + Function and class bodies are not descended into — their bindings are local. + """ + bound: set[str] = set() + + def walk(n) -> None: + for child in n.children: + t = child.type + if t in ("function_definition", "class_definition", "lambda"): + continue # inner scope — not a module-level binding + if t == "assignment": + _python_collect_assignment_targets( + child.child_by_field_name("left"), source, bound + ) + elif t in ("for_statement", "for_in_clause"): + _python_collect_assignment_targets( + child.child_by_field_name("left"), source, bound + ) + elif t == "named_expression": # walrus := + _python_collect_assignment_targets( + child.child_by_field_name("name"), source, bound + ) + walk(child) + + walk(root) + return bound + + +_JS_SCOPE_BOUNDARY = frozenset({ + "function_declaration", "function_expression", "function", "arrow_function", + "method_definition", "class_declaration", "class", "generator_function", + "generator_function_declaration", +}) + + +def _js_collect_pattern_idents(node, source: bytes, bound: set) -> None: + """Collect binding identifier names from a JS/TS pattern (a parameter, or a + declarator LHS). Recurses through destructuring (object/array patterns, rest) + but never into the default-value side of `x = default` or a type annotation, + so only names actually bound by the pattern are collected.""" + t = node.type + if t in ("identifier", "shorthand_property_identifier_pattern"): + bound.add(_read_text(node, source)) + return + if t == "type_annotation": + return # `(h: Handler)` — Handler is a type, not a bound name + if t == "assignment_pattern": # `x = default` — only x is bound + left = node.child_by_field_name("left") + if left is not None: + _js_collect_pattern_idents(left, source, bound) + return + if t == "pair_pattern": # `{ a: localName }` — localName is bound + val = node.child_by_field_name("value") + if val is not None: + _js_collect_pattern_idents(val, source, bound) + return + for c in node.children: + if c.is_named: + _js_collect_pattern_idents(c, source, bound) + + +def _js_local_bound_names(func_node, source: bytes) -> set[str]: + """Names bound locally inside a JS/TS function: parameters plus `const`/`let`/ + `var` declarator targets. Mirrors `_python_local_bound_names`: an argument that + is a parameter or local binding names a local value, not a same-named module + function, so it must not manufacture an indirect_call edge. Nested function and + class scopes are not descended into.""" + bound: set[str] = set() + params = func_node.child_by_field_name("parameters") + if params is not None: + _js_collect_pattern_idents(params, source, bound) + + def walk(n) -> None: + for c in n.children: + if c.type in _JS_SCOPE_BOUNDARY: + continue # inner scope — its bindings are not this function's locals + if c.type == "variable_declarator": + name = c.child_by_field_name("name") + if name is not None: + _js_collect_pattern_idents(name, source, bound) + walk(c) + + body = func_node.child_by_field_name("body") + if body is not None: + walk(body) + return bound + + +def _js_module_bound_names(root, source: bytes) -> set[str]: + """Module-scope names rebound to NON-function data (`const X = {...}`, `let y = 5`). + + The JS/TS module-scope shadow set. Unlike the per-function set, a declarator + whose value is itself a function (`const cb = () => {}`) is EXCLUDED: that name + IS a callable we want dispatch tables to resolve to, not a data shadow. + """ + bound: set[str] = set() + + def walk(n) -> None: + for c in n.children: + if c.type in _JS_SCOPE_BOUNDARY: + continue + if c.type == "variable_declarator": + value = c.child_by_field_name("value") + if value is None or value.type not in _JS_FUNCTION_VALUE_TYPES: + name = c.child_by_field_name("name") + if name is not None: + _js_collect_pattern_idents(name, source, bound) + walk(c) + + walk(root) + return bound + + +def _js_dispatch_value_idents(coll_node): + """Yield identifier value-nodes of a JS/TS object/array literal that are + function-reference candidates: object property VALUES and shorthand properties + (`{ handler }`), and array elements. Keys and inline methods are not references.""" + if coll_node.type == "object": + for c in coll_node.children: + if c.type == "pair": + val = c.child_by_field_name("value") + if val is not None and val.type == "identifier": + yield val + elif c.type == "shorthand_property_identifier": + yield c + else: # array + for el in coll_node.children: + if el.type == "identifier": + yield el + + +def _resolve_name(node, source: bytes, config: LanguageConfig) -> str | None: + """Get the name from a node using config.name_field, falling back to child types.""" + if config.resolve_function_name_fn is not None: + # For C/C++ where the name is inside a declarator + return None # caller handles this separately + n = node.child_by_field_name(config.name_field) + if n: + return _read_text(n, source) + for child in node.children: + if child.type in config.name_fallback_child_types: + return _read_text(child, source) + return None + + +def _find_body(node, config: LanguageConfig): + """Find the body node using config.body_field, falling back to child types.""" + b = node.child_by_field_name(config.body_field) + if b: + return b + for child in node.children: + if child.type in config.body_fallback_child_types: + return child + return None + + +# ── Import handlers ─────────────────────────────────────────────────────────── + +def _import_python(node, source: bytes, file_nid: str, stem: str, edges: list, str_path: str, scope_stack: list[str] | None = None) -> None: + t = node.type + if t == "import_statement": + for child in node.children: + if child.type in ("dotted_name", "aliased_import"): + raw = _read_text(child, source) + module_name = raw.split(" as ")[0].strip().lstrip(".") + tgt_nid = _make_id(module_name) + edges.append({ + "source": file_nid, + "target": tgt_nid, + "relation": "imports", + "context": "import", + "confidence": "EXTRACTED", + "source_file": str_path, + "source_location": f"L{node.start_point[0] + 1}", + "weight": 1.0, + }) + elif t == "import_from_statement": + module_node = node.child_by_field_name("module_name") + if module_node: + raw = _read_text(module_node, source) + if raw.startswith("."): + # Relative import - resolve to full path so IDs match file node IDs + dots = len(raw) - len(raw.lstrip(".")) + module_name = raw.lstrip(".") + base = Path(str_path).parent + for _ in range(dots - 1): + base = base.parent + rel = (module_name.replace(".", "/") + ".py") if module_name else "__init__.py" + tgt_nid = _make_id(str(base / rel)) + else: + tgt_nid = _make_id(raw) + edges.append({ + "source": file_nid, + "target": tgt_nid, + "relation": "imports_from", + "context": "import", + "confidence": "EXTRACTED", + "source_file": str_path, + "source_location": f"L{node.start_point[0] + 1}", + "weight": 1.0, + }) + + +def _resolve_js_import_target(raw: str, str_path: str) -> "tuple[str, Path | None] | None": + """Resolve a JS/TS import path string to (target_nid, resolved_path). + + Handles relative paths, tsconfig path aliases, workspace packages, and + bare/scoped imports. + Returns None if `raw` is empty. + """ + if not raw: + return None + resolved_path = _resolve_js_module_path(raw, Path(str_path).parent) + if resolved_path is not None: + return _make_id(str(resolved_path)), resolved_path + module_name = raw.split("/")[-1] + if not module_name: + return None + # Unresolved: relative/absolute, tsconfig-alias and workspace resolution have + # all run and failed, so this is an external package (or a dangling local + # path). Namespace the id with the "ref" prefix — the J-4 convention already + # used for tsconfig `extends`/`$ref` externals — so it can NEVER collapse to + # the same _make_id as a local file/symbol node. Without it, the bare + # last-segment id (e.g. "tailwindcss/colors" -> "colors") collides with any + # unrelated local file of that stem via build.py's pre-migration alias index, + # producing a confident (EXTRACTED) cross-language phantom imports_from edge + # (#1638). The ref-namespaced target has no node, so build drops it as an + # external reference — the correct outcome for a third-party import. + return _make_id("ref", raw), None + + +def _import_js(node, source: bytes, file_nid: str, stem: str, edges: list, str_path: str, scope_stack: list[str] | None = None) -> None: + is_reexport = node.type == "export_statement" + # Only handle export_statement if it has a `from` clause (re-export). + # Pure exports like `export const x = 1` or `export { localVar }` have no source module. + if is_reexport: has_from = any(child.type == "from" or (_read_text(child, source) == "from") for child in node.children if child.type in ("from", "identifier")) if not has_from: # Check for string child (source path) as a more reliable indicator @@ -373,6 +1948,73 @@ def _import_js(node, source: bytes, file_nid: str, stem: str, edges: list, str_p }) +def _dynamic_import_js(node, source: bytes, caller_nid: str, str_path: str, edges: list, + seen_dyn_pairs: set) -> bool: + """Detect dynamic import() calls in JS/TS and emit imports_from edges. + + Handles patterns like: + await import('./foo.js') + import('./foo.js').then(...) + const m = await import(`./foo`) + + Returns True if the node was a dynamic import (caller should skip normal call handling). + """ + # Dynamic import is a call_expression whose function child is the keyword "import". + # tree-sitter-typescript parses `import('...')` as call_expression with first child + # being an "import" token (type="import"). + func_node = node.child_by_field_name("function") + if func_node is None: + # Fallback: check first child directly (some TS versions) + if node.children and _read_text(node.children[0], source) == "import": + func_node = node.children[0] + else: + return False + if _read_text(func_node, source) != "import": + return False + + # Extract the module path from the arguments + args = node.child_by_field_name("arguments") + if args is None: + return True # It's an import() but no args — skip + for arg in args.children: + if arg.type == "template_string": + # Skip dynamic template literals — path can't be statically resolved + if any(c.type == "template_substitution" for c in arg.children): + break + raw = _read_text(arg, source).strip("`") + elif arg.type == "string": + raw = _read_text(arg, source).strip("'\" ") + else: + continue + if not raw: + break + # Resolve path using the same logic as static imports. + resolved = _resolve_js_import_target(raw, str_path) + if resolved is None: + break + tgt_nid, _ = resolved + pair = (caller_nid, tgt_nid) + if pair not in seen_dyn_pairs: + seen_dyn_pairs.add(pair) + edges.append({ + "source": caller_nid, + "target": tgt_nid, + # A deferred `import(...)` is a real dependency, so keep it as an + # `imports_from` edge (visible in the graph) but mark it `deferred` + # so find_import_cycles does not treat it as a static import and + # report a phantom file cycle (#1241). + "relation": "imports_from", + "context": "import", + "deferred": True, + "confidence": "EXTRACTED", + "source_file": str_path, + "source_location": f"L{node.start_point[0] + 1}", + "weight": 1.0, + }) + break + return True + + def _import_java(node, source: bytes, file_nid: str, stem: str, edges: list, str_path: str, scope_stack: list[str] | None = None) -> None: def _walk_scoped(n) -> str: parts: list[str] = [] @@ -412,6 +2054,20 @@ def _walk_scoped(n) -> str: break +def _resolve_c_include_path(raw: str, str_path: str) -> "Path | None": + """Resolve a quoted #include path to a real file on disk. + + Searches relative to the including file's directory. Returns None for + system headers (<...>) or paths that don't exist on disk. + """ + if not raw: + return None + candidate = (Path(str_path).parent / raw).resolve() + if candidate.is_file(): + return candidate + return None + + def _import_c(node, source: bytes, file_nid: str, stem: str, edges: list, str_path: str, scope_stack: list[str] | None = None) -> None: for child in node.children: if child.type in ("string_literal", "system_lib_string", "string"): @@ -571,24 +2227,710 @@ def _get_c_func_name(node, source: bytes) -> str | None: return None -# ── JS/TS extra walk for arrow functions ────────────────────────────────────── +def _get_cpp_func_name(node, source: bytes) -> str | None: + """Recursively unwrap declarator to find the innermost identifier (C++).""" + if node.type == "identifier": + return _read_text(node, source) + if node.type in ("field_identifier", "destructor_name", "operator_name"): + return _read_text(node, source) + if node.type == "qualified_identifier": + # An out-of-class DEFINITION (`void Foo::bar() {}`) carries a + # qualified_identifier declarator. Retaining the `Foo::` qualifier makes + # _make_id(stem, "Foo::bar") normalize to the same id as the in-class + # member _make_id(class_nid, "bar"), so the decl in Foo.h and the def in + # Foo.cpp resolve to ONE method node instead of two (#1547). The full + # qualified text also handles nested scopes (`A::B::bar`). Free functions + # never have a qualified_identifier here, so their bare-name ids are + # unchanged; only qualified definitions shift onto their owning class. + return _read_text(node, source) + decl = node.child_by_field_name("declarator") + if decl: + return _get_cpp_func_name(decl, source) + for child in node.children: + if child.type == "identifier": + return _read_text(child, source) + return None -# Node types whose value is a callable, for the JS/TS assignment / class-field -# / function-expression forms below. Older tree-sitter-javascript grammars -# label a function expression `function`; current ones use `function_expression`. +def _cpp_declarator_name(node, source: bytes) -> str | None: + """Return the bare variable name from a C++ declaration declarator, unwrapping + pointer/reference/init wrappers (``*f``, ``&r``, ``f = Foo()``). Returns None + for anything that isn't a plain named local (arrays, function pointers, + structured bindings) so the type table never records a guessed receiver.""" + t = node.type + if t == "identifier": + return _read_text(node, source) + if t in ("pointer_declarator", "reference_declarator", "init_declarator"): + inner = node.child_by_field_name("declarator") + if inner is None: + for c in node.children: + if c.type in ("identifier", "pointer_declarator", + "reference_declarator"): + inner = c + break + if inner is not None: + return _cpp_declarator_name(inner, source) + return None + + +def _cpp_local_var_types(body_node, source: bytes, table: dict[str, str]) -> None: + """Collect ``var -> ClassName`` from local variable declarations in a C++ + function body, for receiver-type inference in the cross-file member-call pass + (#1547). Handles ``Foo f;``, ``Foo* f;``, ``Foo *f = ...;``, ``Foo f = Foo();``. + + Only a class-like (``type_identifier``/``qualified_identifier``) type with a + single named declarator is recorded — PRECISION over recall: a built-in type + (``int x``), an ambiguous multi-declarator line, or an un-nameable declarator + contributes nothing rather than a guess. A qualified type ``ns::Foo`` records + its simple tail ``Foo`` so it keys to the type's definition node label. + """ + stack = [body_node] + while stack: + n = stack.pop() + if n.type in ("function_definition", "lambda_expression"): + # Don't descend into a nested function/lambda: its locals are scoped + # away and would pollute this body's table. + if n is not body_node: + continue + if n.type == "declaration": + type_node = n.child_by_field_name("type") + if type_node is not None and type_node.type in ( + "type_identifier", "qualified_identifier" + ): + type_name = _read_text(type_node, source).split("::")[-1].strip() + declarators = [ + c for c in n.children + if c.type in ("identifier", "pointer_declarator", + "reference_declarator", "init_declarator") + ] + # A single declarator only: `Foo a, b;` is ambiguous to attribute + # to one receiver name cleanly, so skip multi-declarator lines. + if type_name and type_name[:1].isupper() and len(declarators) == 1: + var = _cpp_declarator_name(declarators[0], source) + if var and var not in table: + table[var] = type_name + for c in n.children: + stack.append(c) + + +def _swift_local_var_types(body_node, source: bytes, table: dict[str, str]) -> None: + """Collect ``var -> Type`` from local ``let``/``var`` bindings in a Swift + function body, so a member call on the local (``x.method()``) resolves to Type + in the cross-file member-call pass (#1604). + + Two initializer shapes are recorded, PRECISION over recall: + - a constructor call ``let x = Type()`` (``_swift_constructor_type``); + - a static-member access ``let x = Type.shared`` (a navigation_expression + with an upper-cased head) — the singleton-cached-into-a-local idiom, one + of the most common Swift call patterns and previously resolved to nothing. + Nested function declarations are not descended into (their locals are scoped + away); the first binding for a name wins, so a class property of the same name + already in the table is not overwritten. + """ + stack = [body_node] + while stack: + n = stack.pop() + if n.type == "function_declaration" and n is not body_node: + continue + if n.type == "property_declaration": + prop_type: str | None = None + for child in n.children: + if child.type == "call_expression": + prop_type = _swift_constructor_type(child, source) + break + if child.type == "navigation_expression": + head = child.children[0] if child.children else None + if head is not None and head.type == "simple_identifier": + htext = _read_text(head, source) + if htext and htext[:1].isupper(): + prop_type = htext + break + name = _swift_property_name(n, source) + if name and prop_type and name not in table: + table[name] = prop_type + for c in n.children: + stack.append(c) + + +def _csharp_member_type_table(root, source: bytes) -> dict[str, str]: + """Collect ``name -> TypeName`` for C# receiver typing (#1609): class fields, + properties, method parameters, and local variable declarations. + + File-scoped, first-binding-wins (like the C++ table): a field declared once at + class scope is visible to every method's `field.Method()`, and a param/local + shadowing the same name is a conservative approximation graphify already accepts + for receiver typing. Only a resolvable, non-`var` type name is recorded; `var` + without a `new T()` initializer, and predefined/lower-cased primitives, are + skipped (precision over recall — an untypable receiver is left for the resolver + to drop rather than guess). `var v = new T()` is typed from the object-creation. + """ + table: dict[str, str] = {} + + def _typed(type_node) -> str | None: + info = _read_csharp_type_name(type_node, source) + if not info: + return None + name = info[0] + # A genuine C# class name is Pascal-cased; skip predefined primitives + # (int/bool/string) which never own a resolvable method definition here. + return name if name and name[:1].isupper() else None + + def _decl_names(var_decl): + for c in var_decl.children: + if c.type == "variable_declarator": + nm = c.child_by_field_name("name") or next( + (g for g in c.children if g.type == "identifier"), None) + if nm is not None: + yield _read_text(nm, source), c + + def _new_type(declarator) -> str | None: + # `var v = new Server()` — recover the type from the object_creation_expression. + for g in declarator.children: + if g.type == "object_creation_expression": + return _typed(g.child_by_field_name("type")) + return None + + stack = [root] + while stack: + n = stack.pop() + t = n.type + if t in ("field_declaration", "local_declaration_statement"): + vd = next((c for c in n.children if c.type == "variable_declaration"), None) + if vd is not None: + type_node = vd.child_by_field_name("type") + declared = _typed(type_node) + for name, decl in _decl_names(vd): + resolved = declared or _new_type(decl) + if name and resolved and name not in table: + table[name] = resolved + elif t == "property_declaration": + nm = n.child_by_field_name("name") + resolved = _typed(n.child_by_field_name("type")) + if nm is not None and resolved: + pname = _read_text(nm, source) + if pname not in table: + table[pname] = resolved + elif t == "parameter": + nm = n.child_by_field_name("name") + resolved = _typed(n.child_by_field_name("type")) + if nm is not None and resolved: + pname = _read_text(nm, source) + if pname not in table: + table[pname] = resolved + for c in n.children: + stack.append(c) + return table + + +def _ts_receiver_type_table(root, source: bytes, table: dict[str, str]) -> None: + """Add TS/JS receiver bindings to ``table`` (name -> TypeName), for member-call + resolution beyond the constructor-injected `this.field` case (#1630): + + * local ``const/let/var x = new Foo()`` -> ``x: Foo`` (Pattern A); + * a type-annotated parameter ``(svc: Svc)`` -> ``svc: Svc`` (Pattern B), so a + call on the param — including inside a returned closure — resolves. + + File-scoped, first-binding-wins (merged into the constructor-injection table, + which is populated first and therefore wins on a name clash). Only a bare + ``type_identifier`` (a single class/interface name) is recorded — an array, + union, generic, qualified, or predefined type is skipped (precision over + recall, matching the receiver-typed resolvers for Swift/C#/C++).""" + def _bare_type_ident(type_annotation): + # type_annotation -> ": T"; accept only a single type_identifier child. + idents = [c for c in type_annotation.children if c.type == "type_identifier"] + others = [c for c in type_annotation.children + if c.is_named and c.type not in ("type_identifier",)] + if len(idents) == 1 and not others: + return _read_text(idents[0], source) + return None + + stack = [root] + while stack: + n = stack.pop() + t = n.type + if t == "variable_declarator": + name_n = n.child_by_field_name("name") + value = n.child_by_field_name("value") + if (name_n is not None and name_n.type == "identifier" + and value is not None and value.type == "new_expression"): + ctor = value.child_by_field_name("constructor") + if ctor is not None and ctor.type in ("identifier", "type_identifier"): + name = _read_text(name_n, source) + tname = _read_text(ctor, source) + if name and tname and name not in table: + table[name] = tname + elif t == "required_parameter" or t == "optional_parameter": + pat = n.child_by_field_name("pattern") + ann = n.child_by_field_name("type") + if pat is not None and pat.type == "identifier" and ann is not None: + tname = _bare_type_ident(ann) + name = _read_text(pat, source) + if name and tname and name not in table: + table[name] = tname + for c in n.children: + stack.append(c) + + +def _objc_local_var_types(body_node, source: bytes, table: dict[str, str]) -> None: + """Collect ``var -> ClassName`` from ObjC local declarations (``Foo *f = ...;``) + in a method body, for receiver typing in the cross-file message-send pass + (#1556). Only a capitalized ``type_identifier`` with a single named declarator + is recorded; a built-in/lower-cased type or an un-nameable declarator is skipped + (precision over recall). Reuses the C++ declarator unwrapper (identical grammar). + """ + stack = [body_node] + while stack: + n = stack.pop() + if n.type == "method_definition" and n is not body_node: + continue + if n.type == "declaration": + type_node = n.child_by_field_name("type") + if type_node is None: + for c in n.children: + if c.type == "type_identifier": + type_node = c + break + if type_node is not None and type_node.type == "type_identifier": + type_name = _read_text(type_node, source).strip() + declarators = [ + c for c in n.children + if c.type in ("identifier", "pointer_declarator", "init_declarator") + ] + if type_name and type_name[:1].isupper() and len(declarators) == 1: + var = _cpp_declarator_name(declarators[0], source) + if var and var not in table: + table[var] = type_name + for c in n.children: + stack.append(c) + + +# ── JS/TS extra walk for arrow functions ────────────────────────────────────── + +def _find_require_call(value_node): + """Return the call_expression node if `value_node` is a `require(...)` call + or `require(...).x` member access. Otherwise None.""" + if value_node is None: + return None + if value_node.type == "call_expression": + fn = value_node.child_by_field_name("function") + if fn is not None and fn.type == "identifier": + return value_node + if value_node.type == "member_expression": + obj = value_node.child_by_field_name("object") + return _find_require_call(obj) + return None + + +def _require_imports_js(node, source: bytes, file_nid: str, stem: str, edges: list, str_path: str) -> bool: + """Detect CommonJS require imports inside lexical_declaration / variable_declaration. + + Handles three patterns: + const { foo, bar } = require('./mod') → file → mod (imports_from), file → foo, file → bar + const mod = require('./mod') → file → mod (imports_from) + const x = require('./mod').y → file → mod (imports_from), file → y + + Returns True if any require import was found. + """ + if node.type not in ("lexical_declaration", "variable_declaration"): + return False + found = False + for child in node.children: + if child.type != "variable_declarator": + continue + value = child.child_by_field_name("value") + call = _find_require_call(value) + if call is None: + continue + fn = call.child_by_field_name("function") + if fn is None or _read_text(fn, source) != "require": + continue + args = call.child_by_field_name("arguments") + if args is None: + continue + raw = None + for arg in args.children: + if arg.type == "string": + raw = _read_text(arg, source).strip("'\"` ") + break + if not raw: + continue + resolved = _resolve_js_import_target(raw, str_path) + if resolved is None: + continue + tgt_nid, resolved_path = resolved + line = node.start_point[0] + 1 + edges.append({ + "source": file_nid, + "target": tgt_nid, + "relation": "imports_from", + "context": "import", + "confidence": "EXTRACTED", + "source_file": str_path, + "source_location": f"L{line}", + "weight": 1.0, + }) + found = True + + # Symbol-level edges for destructured / accessor binders. + target_stem = _file_stem(resolved_path) if resolved_path is not None else None + name_node = child.child_by_field_name("name") + sym_names: list[str] = [] + if name_node is not None and name_node.type == "object_pattern": + # `const { a, b: alias } = require('./m')` — emit edges for each property key + for prop in name_node.children: + if prop.type == "shorthand_property_identifier_pattern": + sym_names.append(_read_text(prop, source)) + elif prop.type == "pair_pattern": + key = prop.child_by_field_name("key") + if key is not None: + sym_names.append(_read_text(key, source)) + elif value is not None and value.type == "member_expression": + # `const x = require('./m').y` — symbol is the property accessed + prop = value.child_by_field_name("property") + if prop is not None: + sym_names.append(_read_text(prop, source)) + if target_stem is not None: + for sym in sym_names: + edges.append({ + "source": file_nid, + "target": _make_id(target_stem, sym), + "relation": "imports", + "context": "import", + "confidence": "EXTRACTED", + "source_file": str_path, + "source_location": f"L{line}", + "weight": 1.0, + }) + return found + + +# Node types whose value is a callable, for the JS/TS assignment / class-field +# / function-expression forms below. Older tree-sitter-javascript grammars +# label a function expression `function`; current ones use `function_expression`. +_JS_FUNCTION_VALUE_TYPES = frozenset({"arrow_function", "function_expression", "function", "generator_function"}) + + +def _js_member_assignment_target(left, source: bytes): + """Classify the symbol an `assignment_expression` LHS defines when its RHS + is a function. Returns (kind, owner_name, member_name) or None. + + this.foo = fn → ("this", None, "foo") + exports.foo = fn → ("exports", None, "foo") + module.exports.foo = fn → ("exports", None, "foo") + Foo.prototype.bar = fn → ("prototype", "Foo", "bar") + + Any other shape (an arbitrary `obj.x = fn`) returns None and is skipped — + capturing those would reintroduce the bare-named / phantom-god-node class + of bug the module-level scope guard (#1077) exists to prevent. + """ + if left is None or left.type != "member_expression": + return None + prop = left.child_by_field_name("property") + if prop is None: + return None + member_name = _read_text(prop, source) + if not member_name: + return None + obj = left.child_by_field_name("object") + if obj is None: + return None + if obj.type == "this": + return ("this", None, member_name) + if obj.type == "identifier": + if _read_text(obj, source) == "exports": + return ("exports", None, member_name) + return None + if obj.type == "member_expression": + # module.exports.X or Foo.prototype.X + inner_obj = obj.child_by_field_name("object") + inner_prop = obj.child_by_field_name("property") + if inner_obj is None or inner_prop is None: + return None + inner_prop_name = _read_text(inner_prop, source) + if inner_obj.type == "identifier": + inner_obj_name = _read_text(inner_obj, source) + if inner_obj_name == "module" and inner_prop_name == "exports": + return ("exports", None, member_name) + if inner_prop_name == "prototype": + return ("prototype", inner_obj_name, member_name) + return None + + +def _js_extra_walk(node, source: bytes, file_nid: str, stem: str, str_path: str, + nodes: list, edges: list, seen_ids: set, function_bodies: list, + parent_class_nid: str | None, add_node_fn, add_edge_fn, + callable_def_nids: set | None = None, + local_bound_names: dict | None = None) -> bool: + """Handle lexical_declaration (arrow functions, CJS requires, module-level const literals) for JS/TS. Returns True if handled.""" + # CommonJS / prototype member assignments whose value is a function: + # exports.X = () => {} → file-contained function X() + # module.exports.X = fn → file-contained function X() + # Foo.prototype.bar = fn → method bar() owned by Foo + # (`this.X = fn` lives inside a function body, which is not recursed here; + # it is captured at the enclosing function — see the function branch.) + if node.type == "expression_statement": + assign = next((c for c in node.children + if c.type == "assignment_expression"), None) + if assign is not None: + value = assign.child_by_field_name("right") + if value is not None and value.type in _JS_FUNCTION_VALUE_TYPES: + target = _js_member_assignment_target( + assign.child_by_field_name("left"), source) + if target is not None: + kind, owner_name, member_name = target + line = node.start_point[0] + 1 + handled = False + if kind == "exports": + nid = _make_id(stem, member_name) + add_node_fn(nid, f"{member_name}()", line) + add_edge_fn(file_nid, nid, "contains", line) + handled = True + elif kind == "prototype": + owner_nid = _make_id(stem, owner_name) + nid = _make_id(owner_nid, member_name) + add_node_fn(nid, f".{member_name}()", line) + add_edge_fn(owner_nid, nid, "method", line) + handled = True + if handled: + if callable_def_nids is not None: + callable_def_nids.add(nid) # CJS/prototype fn is callable + if local_bound_names is not None: + local_bound_names[nid] = _js_local_bound_names(value, source) + body = value.child_by_field_name("body") + if body: + function_bodies.append((nid, body)) + return True + + # Class fields whose value is a function: + # class C { handler = () => {} } → method handler() owned by C + # Reaches here with parent_class_nid set because class bodies are recursed + # with the class nid as parent. + if parent_class_nid and node.type in ("field_definition", "public_field_definition"): + prop = node.child_by_field_name("property") or node.child_by_field_name("name") + value = node.child_by_field_name("value") + if (prop is not None and value is not None + and value.type in _JS_FUNCTION_VALUE_TYPES): + field_name = _read_text(prop, source) + if field_name: + line = node.start_point[0] + 1 + nid = _make_id(parent_class_nid, field_name) + add_node_fn(nid, f".{field_name}()", line) + add_edge_fn(parent_class_nid, nid, "method", line) + if callable_def_nids is not None: + callable_def_nids.add(nid) # arrow class-field is callable + if local_bound_names is not None: + local_bound_names[nid] = _js_local_bound_names(value, source) + body = value.child_by_field_name("body") + if body: + function_bodies.append((nid, body)) + return True + + if node.type in ("lexical_declaration", "variable_declaration"): + # CJS require imports — emit edges, do not block other lexical_declaration handling + require_found = _require_imports_js(node, source, file_nid, stem, edges, str_path) + + # Scope guard (#1077): only emit nodes for module-level declarations. + # Without this, `const x = ...` inside an arrow callback (e.g. inside + # `describe(() => { const set = new Set(...) })`) emits a bare-named + # node, and the same name collides across unrelated files producing + # phantom god-nodes. Bodies of arrow functions are walked separately + # via function_bodies, so we never need to emit nodes for locals here. + parent = node.parent + is_module_level = parent is not None and ( + parent.type == "program" + or (parent.type == "export_statement" + and parent.parent is not None + and parent.parent.type == "program") + ) + + # Arrow function declarations and module-level const literals (lexical_declaration only) + arrow_found = False + const_found = False + if node.type == "lexical_declaration" and is_module_level: + for child in node.children: + if child.type == "variable_declarator": + value = child.child_by_field_name("value") + if value and value.type in _JS_FUNCTION_VALUE_TYPES: + # `const f = () => {}` and `const f = function(){}` + name_node = child.child_by_field_name("name") + if name_node: + func_name = _read_text(name_node, source) + line = child.start_point[0] + 1 + func_nid = _make_id(stem, func_name) + add_node_fn(func_nid, f"{func_name}()", line) + add_edge_fn(file_nid, func_nid, "contains", line) + if callable_def_nids is not None: + callable_def_nids.add(func_nid) # `const f = () =>` is callable + if local_bound_names is not None: + local_bound_names[func_nid] = _js_local_bound_names(value, source) + body = value.child_by_field_name("body") + if body: + function_bodies.append((func_nid, body)) + arrow_found = True + elif value and value.type in ( + "object", "array", "as_expression", "call_expression", "new_expression", + ): + # Module-level const with literal/object/array/factory value + name_node = child.child_by_field_name("name") + if name_node: + const_name = _read_text(name_node, source) + line = child.start_point[0] + 1 + const_nid = _make_id(stem, const_name) + add_node_fn(const_nid, const_name, line) + add_edge_fn(file_nid, const_nid, "contains", line) + const_found = True + if arrow_found: + return True + if const_found: + return True + if require_found: + return True + return False # ── TS extra walk for namespace / module declarations ───────────────────────── +def _ts_extra_walk(node, source: bytes, file_nid: str, stem: str, str_path: str, + nodes: list, edges: list, seen_ids: set, function_bodies: list, + parent_class_nid: str | None, add_node_fn, add_edge_fn, + walk_fn) -> bool: + """Emit a container node for a TS `namespace`/`module` declaration. + + `namespace Foo {}` parses as `internal_module` (with `name`/`body` fields); + `module Bar {}` and ambient `declare module "pkg" {}` parse as a named + `module` node that exposes no fields, so its name and body are found + positionally. Without this the container was never a node — its members were + still reached by the default recurse but lost their namespace context. The + members stay file-contained (parity with C#'s `_csharp_extra_walk`); the + namespace becomes a sibling marker node so it is queryable. Returns True if + handled. + + The guard requires `is_named` because the anonymous `module` keyword token + shares the `module` type string and would otherwise match here. + """ + if node.is_named and node.type in ("internal_module", "module"): + name_node = node.child_by_field_name("name") + if name_node is None: + for child in node.children: + if child.is_named and child.type in ( + "identifier", "nested_identifier", "string"): + name_node = child + break + body = node.child_by_field_name("body") + if body is None: + for child in node.children: + if child.type == "statement_block": + body = child + break + if name_node is not None: + ns_name = _read_text(name_node, source) + if name_node.type == "string": + ns_name = ns_name.strip("'\"`") + if ns_name: + ns_nid = _make_id(stem, ns_name) + line = node.start_point[0] + 1 + add_node_fn(ns_nid, ns_name, line) + add_edge_fn(file_nid, ns_nid, "contains", line) + if body is not None: + for child in body.children: + walk_fn(child, parent_class_nid) + return True + return False + # ── C# extra walk for namespace declarations ────────────────────────────────── +def _csharp_namespace_name(node, source: bytes) -> str: + name_node = node.child_by_field_name("name") + if name_node is not None: + return _read_text(name_node, source).strip() + for child in node.children: + if child.type in ("identifier", "qualified_name"): + return _read_text(child, source).strip() + return "" + + +def _csharp_extra_walk(node, source: bytes, file_nid: str, stem: str, str_path: str, + nodes: list, edges: list, seen_ids: set, function_bodies: list, + parent_class_nid: str | None, add_node_fn, add_edge_fn, + walk_fn, namespace_stack: list[str], scope_stack: list[str]) -> bool: + """Handle namespace declarations for C#. Returns True if handled.""" + if node.type == "namespace_declaration": + ns_name = _csharp_namespace_name(node, source) + pushed = False + if ns_name: + namespace_stack.append(ns_name) + scope_stack.append(f"s{node.start_byte}") + pushed = True + ns_label = ".".join(namespace_stack) + ns_nid = _csharp_namespace_id(ns_label) + line = node.start_point[0] + 1 + add_node_fn(ns_nid, ns_label, line, node_type="namespace", metadata={"kind": "csharp_namespace"}) + add_edge_fn(file_nid, ns_nid, "contains", line) + body = node.child_by_field_name("body") + if body: + try: + for child in body.children: + walk_fn(child, parent_class_nid) + finally: + if pushed: + namespace_stack.pop() + scope_stack.pop() + elif pushed: + namespace_stack.pop() + scope_stack.pop() + return True + if node.type == "file_scoped_namespace_declaration": + ns_name = _csharp_namespace_name(node, source) + if ns_name: + namespace_stack.append(ns_name) + scope_stack.append(f"s{node.start_byte}") + ns_label = ".".join(namespace_stack) + ns_nid = _csharp_namespace_id(ns_label) + line = node.start_point[0] + 1 + add_node_fn(ns_nid, ns_label, line, node_type="namespace", metadata={"kind": "csharp_namespace"}) + add_edge_fn(file_nid, ns_nid, "contains", line) + return True + return False + # ── Swift extra walk for enum cases ────────────────────────────────────────── - -# ── Java extra walk for enum constants ─────────────────────────────────────── +def _swift_extra_walk(node, source: bytes, file_nid: str, stem: str, str_path: str, + nodes: list, edges: list, seen_ids: set, function_bodies: list, + parent_class_nid: str | None, add_node_fn, add_edge_fn, + ensure_named_node_fn) -> bool: + """Handle enum_entry for Swift. Returns True if handled.""" + if node.type == "enum_entry" and parent_class_nid: + line = node.start_point[0] + 1 + for child in node.children: + if child.type == "simple_identifier": + case_name = _read_text(child, source) + case_nid = _make_id(parent_class_nid, case_name) + add_node_fn(case_nid, case_name, line) + add_edge_fn(parent_class_nid, case_nid, "case_of", line) + # Associated-value types nest as `enum_type_parameters -> user_type -> + # type_identifier` (a sibling of the case-name simple_identifier). The + # case-name loop above never descends into them, so `case started(Session)` + # used to drop the Event -> Session reference entirely. Mirror the Swift + # property/parameter emit style: collect the type refs and emit a + # `references` edge from the ENUM node to each collected type. + for child in node.children: + if child.type != "enum_type_parameters": + continue + for grand in child.children: + if not grand.is_named: + continue + refs: list[tuple[str, str]] = [] + _swift_collect_type_refs(grand, source, False, refs) + for ref_name, role in refs: + ctx = "generic_arg" if role == "generic_arg" else "type" + target_nid = ensure_named_node_fn(ref_name, line) + if target_nid != parent_class_nid: + add_edge_fn(parent_class_nid, target_nid, "references", + line, context=ctx) + return True + return False # ── Language configs ────────────────────────────────────────────────────────── @@ -773,7 +3115,7 @@ def _get_c_func_name(node, source: bytes) -> str | None: # older forks use `simple_identifier`. Accept both so the extractor # works across grammar generations. name_fallback_child_types=("simple_identifier", "identifier"), - body_fallback_child_types=("function_body", "class_body", "enum_class_body"), + body_fallback_child_types=("function_body", "class_body"), function_boundary_types=frozenset({"function_declaration"}), import_handler=_import_kotlin, ) @@ -814,6 +3156,44 @@ def _get_c_func_name(node, source: bytes) -> str | None: ) +def _resolve_lua_import_target(raw_module: str, str_path: str) -> str: + """Resolve a Lua require() module name to a node id. + + Lua module names use dots as path separators: `require("pkg.b")` looks for + `pkg/b.lua` (or `pkg/b/init.lua`) relative to a package root. We probe the + importing file's directory and walk upward looking for a matching file on + disk; if found, the returned id matches the file node id `_extract_generic` + assigns to that file (`_make_id(str(path))`), so the edge lands on a real + node. When nothing matches, fall back to `_make_id` of the full dotted + module name so cross-file resolution can still complete via the symbol + resolution pass instead of dropping the edge entirely (#1075). + """ + if not raw_module: + return "" + rel = raw_module.replace(".", "/") + try: + start_dir = Path(str_path).parent + except Exception: + start_dir = None + if start_dir is not None: + probe = start_dir + # Walk up a few levels so requires from nested files still resolve when + # the package root is above the importing file. + for _ in range(6): + for suffix in (".lua", ".luau"): + cand = probe / f"{rel}{suffix}" + if cand.is_file(): + return _make_id(str(cand)) + for suffix in (".lua", ".luau"): + cand = probe / rel / f"init{suffix}" + if cand.is_file(): + return _make_id(str(cand)) + if probe.parent == probe: + break + probe = probe.parent + return _make_id(raw_module) + + def _import_lua(node, source: bytes, file_nid: str, stem: str, edges: list, str_path: str, scope_stack: list[str] | None = None) -> None: """Extract require('module') from Lua variable_declaration nodes.""" text = _read_text(node, source) @@ -883,6 +3263,31 @@ def _import_swift(node, source: bytes, file_nid: str, stem: str, edges: list, st return modules +def _read_csharp_type_name(node, source: bytes) -> tuple[str, bool, str] | None: + """Resolve a C# type name, whether it was qualified, and its qualifier prefix.""" + if node is None: + return None + if node.type in ("identifier", "predefined_type"): + return (_read_text(node, source), False, "") + if node.type == "qualified_name": + prefix, _, tail = _read_text(node, source).rpartition(".") + tail = tail.split("<", 1)[0] + return (tail, True, prefix) + if node.type == "generic_name": + name_node = node.child_by_field_name("name") + if name_node is not None: + qualified = name_node.type == "qualified_name" + prefix, _, tail = _read_text(name_node, source).rpartition(".") + return (tail, qualified, prefix if qualified else "") + for child in node.children: + if not child.is_named: + continue + result = _read_csharp_type_name(child, source) + if result: + return result + return None + + _SWIFT_CONFIG = LanguageConfig( ts_module="tree_sitter_swift", class_types=frozenset({"class_declaration", "protocol_declaration"}), @@ -901,257 +3306,2661 @@ def _import_swift(node, source: bytes, file_nid: str, stem: str, edges: list, st # ── Ruby local type inference (for member-call resolution) ───────────────────── -# `Const = (...)` shapes that define a lightweight class named after the -# constant. tree-sitter parses each as an `assignment`, not a `class`, so the -# generic class branch never saw them (#1640). +def _ruby_new_class_name(node, source: bytes) -> str | None: + """Return ``ClassName`` if ``node`` is a ``ClassName.new(...)`` call, else None. + Only a bare capitalized constant receiver counts (``Processor.new``); + namespaced (``A::B.new``) and dynamic receivers are intentionally ignored so + the binding stays unambiguous. + """ + if node is None or node.type != "call": + return None + recv = node.child_by_field_name("receiver") + meth = node.child_by_field_name("method") + if recv is None or meth is None: + return None + if recv.type != "constant" or _read_text(meth, source) != "new": + return None + return _read_text(recv, source) -# ── Generic extractor ───────────────────────────────────────────────────────── +def _ruby_local_class_bindings(body_node, source: bytes) -> dict[str, str | None]: + """Map ``local_var -> ClassName`` for ``var = ClassName.new`` within one Ruby + method body, not descending into nested method definitions. -# ── Python rationale extraction ─────────────────────────────────────────────── + 100%-confidence contract: a variable assigned more than once, or to anything + other than a single ``Constant.new``, maps to ``None`` (ambiguous) so callers + never resolve it. Only the certain single-binding case carries a type. + """ + bindings: dict[str, str | None] = {} + boundary = {"method", "singleton_method"} + + def visit(n) -> None: + for child in n.children: + if child.type in boundary: + continue # nested method has its own scope + if child.type == "assignment": + left = child.child_by_field_name("left") + right = child.child_by_field_name("right") + if left is not None and left.type == "identifier": + var = _read_text(left, source) + cls = _ruby_new_class_name(right, source) if right is not None else None + if cls is None: + # assigned to something we can't type: poison if it was typed + if var in bindings: + bindings[var] = None + elif var in bindings: + if bindings[var] != cls: + bindings[var] = None # reassigned to a different class + else: + bindings[var] = cls + visit(child) -_RATIONALE_PREFIXES = ("# NOTE:", "# IMPORTANT:", "# HACK:", "# WHY:", "# RATIONALE:", "# TODO:", "# FIXME:") + visit(body_node) + return bindings -def _is_autogenerated_python(source: bytes) -> bool: - """Return True if this Python file is auto-generated and its module docstring is noise. +def _ruby_const_last_name(node, source: bytes) -> str: + """Last constant of a ``constant`` or ``scope_resolution`` (``A::B::C`` -> ``C``).""" + if node is None: + return "" + if node.type == "constant": + return _read_text(node, source) + if node.type == "scope_resolution": + consts = [c for c in node.children if c.type == "constant"] + if consts: + return _read_text(consts[-1], source) + return "" - Covers: Alembic/Flask-Migrate revisions, Django migrations, protobuf/gRPC/OpenAPI stubs. - Module docstrings in these files are change annotations or boilerplate, not rationale. + +# `Const = (...)` shapes that define a lightweight class named after the +# constant. tree-sitter parses each as an `assignment`, not a `class`, so the +# generic class branch never saw them (#1640). +_RUBY_CLASS_FACTORIES = frozenset({("Struct", "new"), ("Class", "new"), ("Data", "define")}) + + +def _ruby_extra_walk(node, source: bytes, file_nid: str, stem: str, str_path: str, + nodes: list, edges: list, seen_ids: set, function_bodies: list, + parent_class_nid: str | None, add_node, add_edge, walk, + callable_def_nids: set) -> bool: + """Ruby: a constant assignment whose RHS is ``Struct.new(...)``, + ``Class.new(Super)`` or ``Data.define(...)`` defines a class named after the + constant (#1640). Synthesize the class node, attach block-defined methods via + ``method`` (by recursing the block with the new node as parent), and emit an + ``inherits`` edge for ``Class.new(Super)``. Returns True if handled. """ - head = source[:2048].decode("utf-8", errors="replace") - # Generic generated-file markers (protobuf, gRPC, OpenAPI codegen, etc.) - if any(m in head for m in ("DO NOT EDIT", "@generated", "Generated by the protocol buffer")): - return True - # Alembic / Flask-Migrate revision files - if (re.search(r"^revision\s*[:=]", head, re.MULTILINE) - and "def upgrade(" in head - and "down_revision" in head): - return True - # Django migrations - if "class Migration(migrations.Migration)" in head and "operations" in head: - return True - return False + if node.type != "assignment": + return False + left = node.child_by_field_name("left") + right = node.child_by_field_name("right") + if left is None or right is None or left.type != "constant" or right.type != "call": + return False + recv = right.child_by_field_name("receiver") + meth = right.child_by_field_name("method") + if recv is None or meth is None or recv.type != "constant": + return False + if (_read_text(recv, source), _read_text(meth, source)) not in _RUBY_CLASS_FACTORIES: + return False + + const_name = _read_text(left, source) + if not const_name: + return False + line = node.start_point[0] + 1 + class_nid = _make_id(stem, const_name) + add_node(class_nid, const_name, line) + callable_def_nids.add(class_nid) # a class is callable (its constructor) + # Mirror the generic class branch: containment always hangs off the file node. + add_edge(file_nid, class_nid, "contains", line) + + # `Class.new(Super)` — the first positional constant argument is the superclass. + if _read_text(recv, source) == "Class": + args = next((c for c in right.children if c.type == "argument_list"), None) + if args is not None: + for arg in args.children: + if arg.type in ("constant", "scope_resolution"): + base = _ruby_const_last_name(arg, source) + if base: + base_nid = _make_id(stem, base) + if base_nid not in seen_ids: + base_nid = _make_id(base) + if base_nid not in seen_ids: + nodes.append({ + "id": base_nid, "label": base, + "file_type": "code", "source_file": "", + "source_location": "", + }) + seen_ids.add(base_nid) + add_edge(class_nid, base_nid, "inherits", line) + break + # Recurse the do/brace block so block-defined methods attach to the class. + # The block wraps its statements in a `body_statement` (like a class body); + # descend into it so the method handler sees parent_class_nid — otherwise the + # default recurse resets the parent to None and the method hangs off the file + # with a dot-less label. + block = next((c for c in right.children if c.type in ("do_block", "block")), None) + if block is not None: + body = next((c for c in block.children if c.type == "body_statement"), block) + for child in body.children: + walk(child, parent_class_nid=class_nid) + return True -def _extract_python_rationale(path: Path, result: dict) -> None: - """Post-pass: extract docstrings and rationale comments from Python source. - Mutates result in-place by appending to result['nodes'] and result['edges']. + +# ── Generic extractor ───────────────────────────────────────────────────────── + +def _extract_generic( + path: Path, config: LanguageConfig, *, source_override: bytes | None = None +) -> dict: + """Generic AST extractor driven by LanguageConfig. + + ``source_override`` parses the given bytes instead of reading ``path``, while + still keying nodes/edges off ``path``. Lets container formats (e.g. Vue SFCs) + mask the wrapper and parse just the embedded `` close tag + pos = m.end() + if lang is None: + lang_m = _VUE_SCRIPT_LANG_RE.search(m.group(1)) + if lang_m: + lang = lang_m.group(1).lower() + out.append(_blank(src[pos:])) + return "".join(out), lang + - def extract_vue(path: Path) -> dict: """Extract imports, symbols, and type refs from a ``.vue`` SFC. @@ -1637,6 +6480,215 @@ def extract_csharp(path: Path) -> dict: return _extract_generic(path, _CSHARP_CONFIG) +def extract_apex(path: Path) -> dict: + """Extract classes, interfaces, enums, methods, and Salesforce constructs from + Apex .cls and .trigger files using regex (no tree-sitter grammar on PyPI).""" + import re as _re + try: + source = path.read_text(encoding="utf-8", errors="replace") + except OSError: + return {"nodes": [], "edges": []} + + str_path = str(path) + stem = _file_stem(path) + file_nid = _make_id(str_path) + + nodes: list[dict] = [] + edges: list[dict] = [] + seen_ids: set[str] = set() + + def add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": label, + "file_type": "code", + "source_file": str_path, + "source_location": f"L{line}", + }) + + def add_edge(src: str, tgt: str, relation: str, line: int, + confidence: str = "EXTRACTED") -> None: + edges.append({ + "source": src, + "target": tgt, + "relation": relation, + "confidence": confidence, + "source_file": str_path, + "source_location": f"L{line}", + "weight": 1.0, + }) + + add_node(file_nid, path.name, 1) + + lines = source.splitlines() + + _ACCESS = r"(?:public|private|protected|global|webService)?" + _SHARING = r"(?:\s+(?:with|without|inherited)\s+sharing)?" + _MOD = r"(?:\s+(?:abstract|virtual|override|static|final|transient|testMethod))?" + _ANNOTATION = r"(?:\s*@\w+(?:\s*\([^)]*\))?\s*)*" + + cls_re = _re.compile( + rf"^{_ANNOTATION}\s*{_ACCESS}{_SHARING}{_MOD}\s*class\s+(\w+)" + rf"(?:\s+extends\s+(\w+))?(?:\s+implements\s+([\w,\s]+))?\s*\{{?", + _re.IGNORECASE, + ) + iface_re = _re.compile( + rf"^{_ANNOTATION}\s*{_ACCESS}{_SHARING}{_MOD}\s*interface\s+(\w+)" + rf"(?:\s+extends\s+([\w,\s]+))?\s*\{{?", + _re.IGNORECASE, + ) + enum_re = _re.compile( + rf"^{_ANNOTATION}\s*{_ACCESS}{_SHARING}{_MOD}\s*enum\s+(\w+)\s*\{{?", + _re.IGNORECASE, + ) + trigger_re = _re.compile( + r"^\s*trigger\s+(\w+)\s+on\s+(\w+)\s*\(", + _re.IGNORECASE, + ) + method_re = _re.compile( + rf"^{_ANNOTATION}\s*{_ACCESS}{_MOD}\s*(?:static\s+)?[\w<>\[\]]+\s+(\w+)\s*\([^)]*\)\s*(?:throws\s+\w+\s*)?\{{?", + _re.IGNORECASE, + ) + annotation_re = _re.compile(r"@(\w+)", _re.IGNORECASE) + soql_re = _re.compile(r"\[\s*SELECT\b[^\]]+FROM\s+(\w+)", _re.IGNORECASE) + dml_re = _re.compile(r"\b(insert|update|delete|upsert|merge|undelete)\s+\w", _re.IGNORECASE) + + _CONTROL_FLOW = frozenset({ + "if", "else", "for", "while", "do", "switch", "try", "catch", + "finally", "return", "throw", "new", "void", "null", + "true", "false", "this", "super", "class", "interface", "enum", + "trigger", "on", + }) + + current_class_nid: str | None = None + pending_annotations: list[str] = [] + + for lineno, line_text in enumerate(lines, start=1): + stripped = line_text.strip() + + if stripped.startswith("@"): + for m in annotation_re.finditer(stripped): + pending_annotations.append(m.group(1).lower()) + continue + + tm = trigger_re.match(stripped) + if tm: + trig_name, sobject = tm.group(1), tm.group(2) + trig_nid = _make_id(stem, trig_name) + add_node(trig_nid, trig_name, lineno) + add_edge(file_nid, trig_nid, "contains", lineno) + sob_nid = _make_id(sobject) + if sob_nid not in seen_ids: + add_node(sob_nid, sobject, lineno) + add_edge(trig_nid, sob_nid, "uses", lineno, confidence="INFERRED") + current_class_nid = trig_nid + pending_annotations = [] + continue + + cm = cls_re.match(stripped) + if cm: + class_name = cm.group(1) + if class_name.lower() in _CONTROL_FLOW: + pending_annotations = [] + continue + class_nid = _make_id(stem, class_name) + add_node(class_nid, class_name, lineno) + add_edge(file_nid, class_nid, "contains", lineno) + if cm.group(2): + base = cm.group(2).strip() + base_nid = _make_id(stem, base) + if base_nid not in seen_ids: + base_nid = _make_id(base) + if base_nid not in seen_ids: + add_node(base_nid, base, lineno) + add_edge(class_nid, base_nid, "extends", lineno, confidence="INFERRED") + if cm.group(3): + for iface in cm.group(3).split(","): + iface = iface.strip() + if iface: + iface_nid = _make_id(stem, iface) + if iface_nid not in seen_ids: + iface_nid = _make_id(iface) + if iface_nid not in seen_ids: + add_node(iface_nid, iface, lineno) + add_edge(class_nid, iface_nid, "implements", lineno, confidence="INFERRED") + current_class_nid = class_nid + pending_annotations = [] + continue + + im = iface_re.match(stripped) + if im: + iface_name = im.group(1) + if iface_name.lower() in _CONTROL_FLOW: + pending_annotations = [] + continue + iface_nid = _make_id(stem, iface_name) + add_node(iface_nid, iface_name, lineno) + add_edge(file_nid if current_class_nid is None else current_class_nid, + iface_nid, "contains", lineno) + if im.group(2): + for parent in im.group(2).split(","): + parent = parent.strip() + if parent: + parent_nid = _make_id(stem, parent) + if parent_nid not in seen_ids: + parent_nid = _make_id(parent) + if parent_nid not in seen_ids: + add_node(parent_nid, parent, lineno) + add_edge(iface_nid, parent_nid, "extends", lineno, confidence="INFERRED") + pending_annotations = [] + continue + + em = enum_re.match(stripped) + if em: + enum_name = em.group(1) + if enum_name.lower() in _CONTROL_FLOW: + pending_annotations = [] + continue + enum_nid = _make_id(stem, enum_name) + add_node(enum_nid, enum_name, lineno) + add_edge(file_nid if current_class_nid is None else current_class_nid, + enum_nid, "contains", lineno) + pending_annotations = [] + continue + + if current_class_nid is not None: + mm = method_re.match(stripped) + if mm: + method_name = mm.group(1) + if method_name.lower() not in _CONTROL_FLOW: + method_nid = _make_id(current_class_nid, method_name) + method_label = f".{method_name}()" + add_node(method_nid, method_label, lineno) + add_edge(current_class_nid, method_nid, "method", lineno) + if "auraenabled" in pending_annotations or "invocablemethod" in pending_annotations: + add_edge(file_nid, method_nid, "contains", lineno, confidence="INFERRED") + pending_annotations = [] + continue + + pending_annotations = [] + + for sm in soql_re.finditer(line_text): + sobject = sm.group(1) + sob_nid = _make_id(sobject) + if sob_nid not in seen_ids: + add_node(sob_nid, sobject, lineno) + src = current_class_nid or file_nid + add_edge(src, sob_nid, "uses", lineno, confidence="INFERRED") + + for dm in dml_re.finditer(line_text): + dml_op = dm.group(1).lower() + dml_nid = _make_id(f"dml_{dml_op}") + if dml_nid not in seen_ids: + add_node(dml_nid, dml_op, lineno) + src = current_class_nid or file_nid + add_edge(src, dml_nid, "uses", lineno, confidence="INFERRED") + + return {"nodes": nodes, "edges": edges} + + def extract_kotlin(path: Path) -> dict: """Extract classes, objects, functions, and imports from a .kt/.kts file.""" return _extract_generic(path, _KOTLIN_CONFIG) @@ -1652,406 +6704,5352 @@ def extract_php(path: Path) -> dict: return _extract_generic(path, _PHP_CONFIG) -# One level of balanced parens (e.g. `Foo #(Bar #(int))`) — bounded so malformed -# input cannot trigger pathological backtracking. -def extract_lua(path: Path) -> dict: - """Extract functions, methods, require() imports, and calls from a .lua file.""" - return _extract_generic(path, _LUA_CONFIG) +def extract_dart(path: Path) -> dict: + """Extract classes, mixins, functions, imports, generic calls, and annotations from a .dart file using regex.""" + try: + src = path.read_text(encoding="utf-8", errors="replace") + except OSError: + return {"error": f"cannot read {path}"} + + # Remove inline and multi-line comments while leaving string literals untouched to prevent stripping URLs/paths inside strings + comment_string_pattern = re.compile( + r'"""(?:\\.|[\s\S])*?"""' + r"|'''(?:\\.|[\s\S])*?'''" + r'|"(?:\\.|[^"\\])*"' + r"|'(?:\\.|[^'\\])*'" + r"|/\*[\s\S]*?\*/" + r"|//[^\n]*" + ) + def _comment_replace(match: re.Match) -> str: + token = match.group(0) + if token.startswith("/"): + return "" + return token + src_clean = comment_string_pattern.sub(_comment_replace, src) + stem = _file_stem(path) + file_nid = _make_id(str(path)) -def extract_swift(path: Path) -> dict: - """Extract classes, structs, protocols, functions, imports, and calls from a .swift file.""" - return _extract_generic(path, _SWIFT_CONFIG) + # Check if this is a part-of file and redirect to parent + part_of_match = re.search(r"^\s*part\s+of\s+['\"]([^'\"]+)['\"]", src_clean, re.MULTILINE) + is_part = False + if part_of_match: + parent_ref = part_of_match.group(1) + if parent_ref.endswith(".dart"): + try: + parent_path = (path.parent / parent_ref).resolve() + if parent_path.exists(): + stem = _file_stem(parent_path) + file_nid = _make_id(str(parent_path)) + is_part = True + except Exception: + pass + nodes = [] + if not is_part: + nodes.append({"id": file_nid, "label": path.name, "file_type": "code", + "source_file": str(path), "source_location": None}) + edges = [] + defined: set[str] = set() + + def add_node(nid: str, label: str, ftype: str = "code", source_file: str | None = str(path)) -> None: + if nid not in defined: + nodes.append({"id": nid, "label": label, "file_type": ftype, + "source_file": source_file, "source_location": None}) + defined.add(nid) + + def add_edge(src_id: str, tgt_id: str, relation: str, weight: float = 1.0, context: str | None = None) -> None: + edge = {"source": src_id, "target": tgt_id, "relation": relation, + "confidence": "EXTRACTED", "confidence_score": 1.0, + "source_file": str(path), "source_location": None, "weight": weight} + if context: + edge["context"] = context + edges.append(edge) -# ── Julia extractor (custom walk) ──────────────────────────────────────────── + def _split_types(text: str) -> list[str]: + parts = [] + current = [] + depth = 0 + for char in text: + if char == "<": + depth += 1 + current.append(char) + elif char == ">": + depth -= 1 + current.append(char) + elif char == "," and depth == 0: + parts.append("".join(current).strip()) + current = [] + else: + current.append(char) + if current: + parts.append("".join(current).strip()) + return [p for p in parts if p] + + def _find_matching_brace(text: str, start_pos: int) -> int: + brace_count = 0 + in_double_quote = False + in_single_quote = False + escape = False + + first_brace = text.find("{", start_pos) + if first_brace == -1: + return len(text) + + brace_count = 1 + i = first_brace + 1 + n = len(text) + while i < n: + char = text[i] + if escape: + escape = False + i += 1 + continue + if char == "\\": + escape = True + i += 1 + continue + if text[i:i+3] == '"""' and not in_single_quote: + i += 3 + end = text.find('"""', i) + i = end + 3 if end != -1 else n + continue + if text[i:i+3] == "'''" and not in_double_quote: + i += 3 + end = text.find("'''", i) + i = end + 3 if end != -1 else n + continue + if char == '"' and not in_single_quote: + in_double_quote = not in_double_quote + elif char == "'" and not in_double_quote: + in_single_quote = not in_single_quote + elif not in_double_quote and not in_single_quote: + if char == "{": + brace_count += 1 + elif char == "}": + brace_count -= 1 + if brace_count == 0: + return i + 1 + i += 1 + return len(text) + + # 1. Classes, mixins, and enums declarations (with inheritance, mixins, interfaces, and generics) + # Supports multiple combined modifiers (e.g., abstract base class, mixin class) without capturing "class" as a name + class_pattern = r"^\s*(?:(?:abstract|sealed|base|interface|final|mixin)\s+)*(?:class|mixin|enum|extension\s+type)\s+(\w+)" + for m in re.finditer(class_pattern, src_clean, re.MULTILINE): + class_name = m.group(1) + class_nid = _make_id(stem, class_name) + add_node(class_nid, class_name) + add_edge(file_nid, class_nid, "defines") + + # Manually parse extends/on, with, and implements in header to handle nested generics brackets balanced + start_idx = m.end() + rest = src_clean[start_idx : start_idx + 500] + + # Skip class generic parameters + if rest.lstrip().startswith("<"): + offset = rest.find("<") + depth = 1 + i = offset + 1 + while i < len(rest) and depth > 0: + if rest[i] == "<": depth += 1 + elif rest[i] == ">": depth -= 1 + i += 1 + rest = rest[i:] + + # Skip primary constructor (e.g. extension type MyExt(int id)) + if rest.lstrip().startswith("("): + offset = rest.find("(") + depth = 1 + i = offset + 1 + while i < len(rest) and depth > 0: + if rest[i] == "(": depth += 1 + elif rest[i] == ")": depth -= 1 + i += 1 + rest = rest[i:] + + header_end = rest.find("{") + if header_end == -1: + header_end = rest.find(";") + if header_end == -1: + header_end = len(rest) + header = rest[:header_end] + + base_class = None + generics = None + mixins_list = [] + interfaces_list = [] + + # Parse extends or on + extends_m = re.search(r"^\s*(?:extends|on)\s+([a-zA-Z0-9_.]+)", header) + if extends_m: + base_class = extends_m.group(1) + rest_header = header[extends_m.end():] + if rest_header.strip().startswith("<"): + start_idx = rest_header.find("<") + depth = 1 + i = start_idx + 1 + while i < len(rest_header) and depth > 0: + if rest_header[i] == "<": + depth += 1 + elif rest_header[i] == ">": + depth -= 1 + if depth == 0: + generics = rest_header[start_idx + 1 : i] + break + i += 1 + if generics is not None: + header = rest_header[i + 1:] + else: + header = rest_header + else: + header = rest_header + + # Parse with + with_m = re.search(r"^\s*with\s+", header) + if with_m: + rest_header = header[with_m.end():] + impl_idx = rest_header.find("implements") + if impl_idx != -1: + mixins_str = rest_header[:impl_idx] + header = rest_header[impl_idx:] + else: + mixins_str = rest_header + header = "" + mixins_list = _split_types(mixins_str) + + # Parse implements + impl_m = re.search(r"^\s*implements\s+", header) + if impl_m: + interfaces_list = _split_types(header[impl_m.end():]) + + # Map extends inheritance relation + if base_class: + base_nid = _make_id(base_class) + add_node(base_nid, base_class, source_file=None) + add_edge(class_nid, base_nid, "inherits") + + # Map generic type arguments (e.g. MyBloc extends Bloc) + if generics: + for gen in _split_types(generics): + gen_clean = gen.split("<")[0].strip() + if gen_clean not in {"String", "int", "double", "bool", "num", "dynamic", "Object", "void"}: + gen_nid = _make_id(gen_clean) + add_node(gen_nid, gen_clean, source_file=None) + add_edge(class_nid, gen_nid, "references") + + # Map mixins + for mixin in mixins_list: + mixin_clean = mixin.split("<")[0].strip() + mixin_nid = _make_id(mixin_clean) + add_node(mixin_nid, mixin_clean, source_file=None) + add_edge(class_nid, mixin_nid, "mixes_in") + + # Map interfaces + for interface in interfaces_list: + interface_clean = interface.split("<")[0].strip() + interface_nid = _make_id(interface_clean) + add_node(interface_nid, interface_clean, source_file=None) + add_edge(class_nid, interface_nid, "implements") + + # Extract class body for precise framework dependencies and event handling + start_idx = m.start() + brace_pos = src_clean.find("{", start_idx) + semi_pos = src_clean.find(";", start_idx) + + has_body = brace_pos != -1 + if has_body and semi_pos != -1 and semi_pos < brace_pos: + has_body = False + + if has_body: + end_pos = _find_matching_brace(src_clean, start_idx) + class_body = src_clean[brace_pos:end_pos] + + # Bloc event registration: on() + for em in re.finditer(r"\bon<(\w+)>\s*\(", class_body): + event_name = em.group(1) + event_nid = _make_id(event_name) + add_node(event_nid, event_name, source_file=None) + add_edge(class_nid, event_nid, "calls", context="bloc_event") + + # Bloc state emissions: emit(MyState) or yield MyState + for sm in re.finditer(r"\b(?:emit|yield)\s*\(?\s*(?:const\s+)?([A-Z]\w*)\b", class_body): + state_name = sm.group(1) + if state_name not in {"String", "List", "Map", "Set", "Future", "Stream", "Object"}: + state_nid = _make_id(state_name) + add_node(state_nid, state_name, source_file=None) + add_edge(class_nid, state_nid, "calls", context="emit_state") + + # Bloc event additions: widget.add(MyEvent()) or bloc.add(MyEvent()) + for am in re.finditer(r"\b(?:\w*[Bb]loc\w*|context\.read<\w+>\(\))\.add\(\s*(?:const\s+)?([A-Z]\w*)\b", class_body): + event_name = am.group(1) + if event_name not in {"String", "List", "Map", "Set", "Future", "Stream", "Object"}: + event_nid = _make_id(event_name) + add_node(event_nid, event_name, source_file=None) + add_edge(class_nid, event_nid, "calls", context="bloc_add_event") + + # Riverpod provider references: ref.watch(provider) + for rm in re.finditer(r"\bref\.(?:watch|read|listen)\s*\(\s*(\w+)\b", class_body): + provider_name = rm.group(1) + provider_nid = _make_id(provider_name) + add_node(provider_nid, provider_name, source_file=None) + add_edge(class_nid, provider_nid, "references", context="riverpod_reference") + + # Widget to Bloc references: BlocBuilder + for bm in re.finditer(r"\bBloc(?:Builder|Listener|Consumer|Provider|Selector)\s*<\s*([a-zA-Z0-9_]+)\b", class_body): + bloc_name = bm.group(1) + if bloc_name not in {"String", "int", "double", "bool", "num", "dynamic", "Object", "void"}: + bloc_nid = _make_id(bloc_name) + add_node(bloc_nid, bloc_name, source_file=None) + add_edge(class_nid, bloc_nid, "references", context="bloc_widget_binding") + + # context.read() or BlocProvider.of(context) + for lm in re.finditer(r"\b(?:read|watch|select|of)\s*<([a-zA-Z0-9_]+)>", class_body): + bloc_name = lm.group(1) + if bloc_name not in {"String", "int", "double", "bool", "num", "dynamic", "Object", "void"}: + bloc_nid = _make_id(bloc_name) + add_node(bloc_nid, bloc_name, source_file=None) + add_edge(class_nid, bloc_nid, "references", context="bloc_lookup") + + # 2. Annotations mapping (class, mixin, enum, or function level annotations) + # Support: @riverpod, @Riverpod(...), @injectable, @singleton, @RoutePage(), @HiveType(typeId: 0), @RestApi() + # Matches `@annotation` and links it to the next class/mixin/enum/function declaration in the file + annotation_pattern = r"@(\w+)(?:\([^)]*\))?" + for am in re.finditer(annotation_pattern, src_clean): + annotation_name = am.group(1) + if annotation_name in {"override", "deprecated", "required", "protected", "mustCallSuper"}: + continue + annotation_pos = am.end() + intervening_text = src_clean[annotation_pos : annotation_pos + 300] + class_m = re.search(r"^\s*(?:(?:abstract|sealed|base|interface|final|mixin)\s+)*(?:class|mixin|enum|extension\s+type)\s+(\w+)", intervening_text, re.MULTILINE) + func_m = re.search(r"^\s*(?:factory\s+|static\s+|async\s+|external\s+|abstract\s+)?(?:\([^)]+\)|[a-zA-Z0-9_<>,.?]+)(?:\s+[a-zA-Z0-9_<>,.?]+){0,3}\s+(\w+)\s*\(", intervening_text, re.MULTILINE) -# ── Go extractor (custom walk) ──────────────────────────────────────────────── + target_nid = None + target_name = None + target_type = None + if class_m and func_m: + if class_m.start() < func_m.start(): + target_name = class_m.group(1) + target_type = "class" + target_nid = _make_id(stem, target_name) + else: + target_name = func_m.group(1) + target_type = "function" + target_nid = _make_id(stem, target_name) + elif class_m: + target_name = class_m.group(1) + target_type = "class" + target_nid = _make_id(stem, target_name) + elif func_m: + target_name = func_m.group(1) + target_type = "function" + target_nid = _make_id(stem, target_name) + + if target_nid and target_name: + actual_intervening = intervening_text[:min(class_m.start() if class_m else 300, func_m.start() if func_m else 300)] + if ";" not in actual_intervening and "}" not in actual_intervening and "{" not in actual_intervening: + annotation_nid = _make_id("annotation", annotation_name.lower()) + add_node(annotation_nid, f"@{annotation_name}", ftype="concept", source_file=None) + add_edge(target_nid, annotation_nid, "configures") + + # Riverpod specific provider generation mapping (supports camelCase class and functional providers) + if annotation_name.lower() == "riverpod": + if target_type == "class": + provider_name = target_name[0].lower() + target_name[1:] + "Provider" if len(target_name) > 1 else target_name.lower() + "Provider" + else: + provider_name = target_name + "Provider" + provider_nid = _make_id(provider_name) + add_node(provider_nid, provider_name, ftype="concept", source_file=str(path)) + add_edge(target_nid, provider_nid, "defines", context="riverpod_provider") + + # 2.5 Typedefs (Type Aliases) + typedef_pattern = r"^\s*typedef\s+(\w+)\s*(?:<[^>]+>)?\s*=\s*([a-zA-Z0-9_<>,.?\s]+);" + for m in re.finditer(typedef_pattern, src_clean, re.MULTILINE): + typedef_name = m.group(1) + target_type = m.group(2).split("<")[0].split(".")[-1].strip() + if target_type not in {"String", "int", "double", "bool", "num", "dynamic", "Object", "List", "Map", "Set", "void", "Function"}: + typedef_nid = _make_id(stem, typedef_name) + add_node(typedef_nid, typedef_name) + add_edge(file_nid, typedef_nid, "defines") + target_nid = _make_id(target_type) + add_node(target_nid, target_type, source_file=None) + add_edge(typedef_nid, target_nid, "references", context="typedef") + + # 3. Extensions (extension MyExt on MyClass) + ext_pattern = r"^\s{0,4}extension\s+(\w+)?(?:<[^>]+>)?\s+on\s+(\w+)" + for m in re.finditer(ext_pattern, src_clean, re.MULTILINE): + ext_name = m.group(1) or f"{stem}_anonymous_extension" + target_class = m.group(2) + + ext_nid = _make_id(stem, ext_name) + label = m.group(1) or f"Extension on {target_class}" + add_node(ext_nid, label) + add_edge(file_nid, ext_nid, "defines") + + target_nid = _make_id(target_class) + add_node(target_nid, target_class, source_file=None) + add_edge(ext_nid, target_nid, "extends") + + # 4. Top-level and class-level variable declarations (generic variables, records, late, and destructuring) + # Restrict indentation to 0-2 spaces to avoid matching local variables inside functions or switch expressions + var_pattern = r"^\s{0,2}(?:late\s+)?(?:(?:final|const|var)\s+)?(?:\([^)]+\)\s+|([a-zA-Z0-9_<>,.?]+(?:\s+[a-zA-Z0-9_<>,.?]+){0,3})\s+)?(?:(\w+)|(?:\w+\s*)?\(([^)]+)\))\s*(?:=|$|;)" + for m in re.finditer(var_pattern, src_clean, re.MULTILINE): + var_type = m.group(1) + single_name = m.group(2) + destructured_names = m.group(3) + + if not re.match(r"^\s*(?:late|final|const|var)\b", m.group(0)) and not var_type: + continue -# ── Rust extractor (custom walk) ────────────────────────────────────────────── + if single_name: + if single_name not in {"if", "for", "while", "switch", "catch", "return"}: + var_nid = _make_id(stem, single_name) + add_node(var_nid, single_name) + add_edge(file_nid, var_nid, "defines") + + if var_type and var_type not in {"String", "int", "double", "bool", "num", "dynamic", "Object", "List", "Map", "Set", "void"}: + clean_type = var_type.split("<")[0].split(".")[-1].strip() + type_nid = _make_id(clean_type) + add_node(type_nid, clean_type, source_file=None) + add_edge(file_nid, type_nid, "references", context="variable_type") + elif destructured_names: + for name in [n.strip() for n in destructured_names.split(",") if n.strip()]: + if ":" in name: + name = name.split(":")[-1].strip() + if re.match(r"^[a-zA-Z_]\w*$", name) and not re.match(r"^[A-Z]", name): + if name not in {"if", "for", "while", "switch", "catch", "return"}: + var_nid = _make_id(stem, name) + add_node(var_nid, name) + add_edge(file_nid, var_nid, "defines") + + # 5. Top-level and member functions/methods (supports typed/generic/record return types and Riverpod/Bloc references) + # Restrict indentation to 0-2 spaces to avoid matching nested local functions or methods inside multiline switch statements + method_pattern = r"^\s{0,2}(?:factory\s+|static\s+|async\s+|external\s+|abstract\s+)?(?:\([^)]+\)|[a-zA-Z0-9_<>,.?]+)(?:\s+[a-zA-Z0-9_<>,.?]+){0,3}\s+(\w+(?:\.\w+)?)\s*\(" + for m in re.finditer(method_pattern, src_clean, re.MULTILINE): + raw_name = m.group(1) + name = raw_name.split(".")[-1] + if name in {"if", "for", "while", "switch", "catch", "return", "void", "dynamic", "final", "const", "get", "set"}: + continue + if re.match(r"^[A-Z]", name): + continue + nid = _make_id(stem, name) + add_node(nid, name) + add_edge(file_nid, nid, "defines") + + # Get function body using matching brace to extract Riverpod reference patterns + start_idx = m.start() + brace_pos = src_clean.find("{", start_idx) + semi_pos = src_clean.find(";", start_idx) + arrow_pos = src_clean.find("=>", start_idx) + + has_body = brace_pos != -1 + if has_body and semi_pos != -1 and semi_pos < brace_pos: + has_body = False + if has_body and arrow_pos != -1 and arrow_pos < brace_pos: + has_body = False + + if has_body: + end_pos = _find_matching_brace(src_clean, start_idx) + func_body = src_clean[brace_pos:end_pos] + + # Extract Riverpod provider references: ref.watch(provider) + for rm in re.finditer(r"\bref\.(?:watch|read|listen)\s*\(\s*(\w+)\b", func_body): + provider_name = rm.group(1) + provider_nid = _make_id(provider_name) + add_node(provider_nid, provider_name, source_file=None) + add_edge(nid, provider_nid, "references", context="riverpod_reference") + + # Extract Bloc event additions: widget.add(MyEvent()) or bloc.add(MyEvent()) + for am in re.finditer(r"\b(?:\w*[Bb]loc\w*|context\.read<\w+>\(\))\.add\(\s*(?:const\s+)?([A-Z]\w*)\b", func_body): + event_name = am.group(1) + if event_name not in {"String", "List", "Map", "Set", "Future", "Stream", "Object"}: + event_nid = _make_id(event_name) + add_node(event_nid, event_name, source_file=None) + add_edge(nid, event_nid, "calls", context="bloc_add_event") + + # context.read() or BlocProvider.of(context) + for lm in re.finditer(r"\b(?:read|watch|select|of)\s*<([a-zA-Z0-9_]+)>", func_body): + bloc_name = lm.group(1) + if bloc_name not in {"String", "int", "double", "bool", "num", "dynamic", "Object", "void"}: + bloc_nid = _make_id(bloc_name) + add_node(bloc_nid, bloc_name, source_file=None) + add_edge(nid, bloc_nid, "references", context="bloc_lookup") + + # Universal Navigation Patters (GoRouter, AutoRoute, Navigator) + for nm in re.finditer(r"\b(?:go|push|goNamed|pushNamed|replace|replaceNamed)\s*\(\s*(?:context\s*,\s*)?['\"]([a-zA-Z0-9_/?=&%-]+)['\"]", func_body): + route_path = nm.group(1) + route_nid = _make_id("route", route_path.replace("/", "_").replace("?", "_").replace("=", "_").replace("&", "_")) + add_node(route_nid, f"Route {route_path}", ftype="concept", source_file=None) + add_edge(nid, route_nid, "navigates", context="route_path") + + for cm in re.finditer(r"\b(?:go|push|goNamed|pushNamed|replace|replaceNamed)\s*\(\s*(?:context\s*,\s*)?([A-Z][a-zA-Z0-9_]*\.[a-zA-Z0-9_]+)", func_body): + route_const = cm.group(1) + route_nid = _make_id("route", route_const.replace(".", "_")) + add_node(route_nid, route_const, ftype="concept", source_file=None) + add_edge(nid, route_nid, "navigates", context="route_const") + + for om in re.finditer(r"\b(?:push|replace)\s*\(\s*(?:context\s*,\s*)?.*?\b([A-Z]\w*(?:Route|Screen|Page))\b", func_body): + route_class = om.group(1) + route_nid = _make_id(route_class) + add_node(route_nid, route_class, source_file=None) + add_edge(nid, route_nid, "navigates", context="route_object") + + # 6. Imports and Exports + for m in re.finditer(r"""^\s*import\s+['"]([^'"]+)['"]""", src_clean, re.MULTILINE): + pkg = m.group(1) + tgt_nid = _make_id(pkg) + add_node(tgt_nid, pkg, source_file=None) + add_edge(file_nid, tgt_nid, "imports") + + for m in re.finditer(r"""^\s*export\s+['"]([^'"]+)['"]""", src_clean, re.MULTILINE): + pkg = m.group(1) + tgt_nid = _make_id(pkg) + add_node(tgt_nid, pkg, source_file=None) + add_edge(file_nid, tgt_nid, "exports") + + # 7. Generic Invocations / Type Lookups (Universal Dependency Lookup) + # Matches any method call with type parameters: methodName() or object.methodName() + # Automatically extracts GetIt, Injectable, Riverpod, Provider, BlocProvider, and InheritedWidget type lookups! + generic_call_pattern = r"\b\w+<([a-zA-Z0-9_.]+(?:<[a-zA-Z0-9_.,\s<>]+>)?)\s*>\s*\(" + type_blacklist = {"String", "int", "double", "bool", "num", "dynamic", "Object", "List", "Map", "Set", "Future", "Stream", "void"} + for m in re.finditer(generic_call_pattern, src_clean): + type_name = m.group(1).split(".")[-1].strip() + clean_name = type_name.split("<")[0].strip() + if clean_name not in type_blacklist: + target_nid = _make_id(clean_name) + add_node(target_nid, clean_name, source_file=None) + add_edge(file_nid, target_nid, "references", context="type_lookup") -# Common Rust trait/stdlib method names that appear in virtually every codebase. -# Resolving these cross-file produces spurious INFERRED edges across crate -# boundaries (issue #908) — skip them from the unresolved-call queue entirely. + return {"nodes": nodes, "edges": edges} -# ── Zig ─────────────────────────────────────────────────────────────────────── +def _sv_first_identifier(node, source: bytes) -> str | None: + """First `simple_identifier` under node in pre-order, or None. + tree-sitter-verilog 1.0.3 nests declaration names a few levels deep instead + of exposing a `name` field. Scope the search to the right child node (e.g. + `function_identifier`) or this returns the return-type instead of the name. + """ + if node is None: + return None + for child in node.children: + if child.type == "simple_identifier": + return _read_text(child, source) + found = _sv_first_identifier(child, source) + if found: + return found + return None -# ── PowerShell ──────────────────────────────────────────────────────────────── +def _sv_child(node, type_name: str) -> object | None: + if node is None: + return None + for child in node.children: + if child.type == type_name: + return child + return None -# ── PowerShell manifest (.psd1) ────────────────────────────────────────────── -# Keys in a .psd1 whose values are module names/paths we treat as imports. +_SV_BUILTIN_TYPES = frozenset({ + "bit", "logic", "reg", "wire", "int", "integer", "shortint", "longint", + "byte", "time", "real", "shortreal", "void", "string", "type", "event", + "mailbox", "semaphore", "process", "chandle", +}) +_SV_NON_TYPE_WORDS = frozenset({ + "return", "if", "else", "for", "foreach", "while", "case", "begin", "end", + "function", "task", "class", "endclass", "endfunction", "endtask", +}) -# ── Cross-file import resolution ────────────────────────────────────────────── +# One level of balanced parens (e.g. `Foo #(Bar #(int))`) — bounded so malformed +# input cannot trigger pathological backtracking. +_SV_PARENS_INNER = r"(?:[^()]|\([^()]*\))*" +_SV_PARENS = r"\(" + _SV_PARENS_INNER + r"\)" +_SV_FUNC_RE = re.compile( + r"\bfunction\s+([A-Za-z_]\w*(?:\s*#\s*" + _SV_PARENS + r")?)\s+(\w+)\s*" + r"\((" + _SV_PARENS_INNER + r")\)\s*;", + re.MULTILINE, +) -def _canonicalize_csharp_namespace_nodes(all_nodes: list[dict], all_edges: list[dict]) -> None: - """Collapse duplicate C# namespace node entries to one canonical node per label.""" - by_label: dict[str, list[dict]] = {} - for node in all_nodes: - if node.get("type") != "namespace": - continue - label = node.get("label") - if isinstance(label, str): - by_label.setdefault(label, []).append(node) +_SV_PARAM_RE = re.compile( + r"\s*(?:input|output|inout|ref|const\s+ref)?\s*" + r"([A-Za-z_]\w*(?:\s*#\s*" + _SV_PARENS + r")?)\s+\w+" +) - remap: dict[str, str] = {} - drop_node_ids: set[int] = set() - for group in by_label.values(): - if len(group) < 2: - continue - canonical = sorted( - group, - key=lambda node: ( - str(node.get("source_file") or ""), - str(node.get("source_location") or ""), - str(node.get("id") or ""), - ), - )[0] - canonical_id = canonical.get("id") - for node in group: - if node is canonical: - continue - drop_node_ids.add(id(node)) - dup_id = node.get("id") - if isinstance(dup_id, str) and isinstance(canonical_id, str): - remap[dup_id] = canonical_id - if remap: - for edge in all_edges: - if edge.get("source") in remap: - edge["source"] = remap[str(edge["source"])] - if edge.get("target") in remap: - edge["target"] = remap[str(edge["target"])] +def _sv_strip_comments(text: str) -> str: + text = re.sub(r"/\*.*?\*/", "", text, flags=re.DOTALL) + return re.sub(r"//.*", "", text) - if drop_node_ids: - all_nodes[:] = [node for node in all_nodes if id(node) not in drop_node_ids] +def _sv_split_type_list(text: str) -> list[str]: + parts: list[str] = [] + depth = 0 + start = 0 + for idx, ch in enumerate(text): + if ch == "(": + depth += 1 + elif ch == ")": + depth = max(0, depth - 1) + elif ch == "," and depth == 0: + item = text[start:idx].strip() + if item: + parts.append(item) + start = idx + 1 + item = text[start:].strip() + if item: + parts.append(item) + return parts -# Languages whose identifiers are case-insensitive, so cross-file name resolution -# may fold case. Everywhere else, case is semantic (`Path` the class vs `PATH` the -# env var are distinct) and folding manufactures false edges / super-hubs (#1581). -_CASE_INSENSITIVE_EXTS = frozenset({ - ".php", ".phtml", ".php3", ".php4", ".php5", ".php7", ".phps", # PHP fns/classes - ".sql", # SQL identifiers - ".nim", ".nims", ".nimble", # Nim (style-insensitive) -}) +def _sv_collect_type_refs(type_text: str, generic: bool = False, + skip: frozenset[str] = frozenset()) -> list[tuple[str, str]]: + refs: list[tuple[str, str]] = [] + text = type_text.strip() + if not text: + return refs + head = re.match(r"([A-Za-z_]\w*)", text) + if head: + name = head.group(1) + # `skip` carries the enclosing class's `#(type T = ...)` parameters so + # they are not mistaken for referenced types. + if name not in _SV_BUILTIN_TYPES and name not in _SV_NON_TYPE_WORDS and name not in skip: + refs.append((name, "generic_arg" if generic else "type")) + params = re.search(r"#\s*\((" + _SV_PARENS_INNER + r")\)", text) + if params: + for arg in _sv_split_type_list(params.group(1)): + refs.extend(_sv_collect_type_refs(arg, generic=True, skip=skip)) + return refs + + +def _augment_systemverilog_semantics( + raw: str, + stem: str, + str_path: str, + file_nid: str, + nodes: list[dict], + edges: list[dict], + seen_ids: set[str], +) -> None: + label_to_nid = {node["label"]: node["id"] for node in nodes} -def _lang_is_case_insensitive(source_file: object) -> bool: - """True when the file's language resolves identifiers case-insensitively (#1581).""" - if not source_file: - return False - return Path(str(source_file)).suffix.lower() in _CASE_INSENSITIVE_EXTS + def line_for(offset: int) -> int: + return raw.count("\n", 0, offset) + 1 + def add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({"id": nid, "label": label, "file_type": "code", + "source_file": str_path, "source_location": f"L{line}", + "confidence_score": 1.0}) + label_to_nid[label] = nid + + def ensure_type(label: str, line: int) -> str: + if label in label_to_nid: + return label_to_nid[label] + nid = _make_id(stem, label) + add_node(nid, label, line) + return nid + + def add_edge(src: str, target_label: str, relation: str, line: int, context: str | None = None) -> None: + tgt = ensure_type(target_label, line) + edge = {"source": src, "target": tgt, "relation": relation, + "confidence": "EXTRACTED", "confidence_score": 1.0, + "source_file": str_path, "source_location": f"L{line}", "weight": 1.0} + if context: + edge["context"] = context + edges.append(edge) -# Language interop families for cross-file call resolution. A call in one language -# can never bind by name to a definition in another family — a TSX component does -# not invoke a Kotlin method, and a Python function does not invoke a Java one. -# Families are grouped by REAL interop so legitimate cross-language resolution -# keeps working: Kotlin/Java/Scala/Groovy share the JVM, C/C++/Objective-C/CUDA -# share headers and symbols (Swift bridges to Objective-C), and JS/TS variants -# (plus Vue/Svelte/Astro SFC script blocks) compile into one module graph. -# Extensions absent from this map (docs, configs, unknown languages) resolve to -# no family and are never filtered — same permissive default as before. -_LANG_FAMILY_BY_EXT: dict[str, str] = { - # JS/TS module graph (SFCs embed JS/TS) - ".js": "jsts", ".jsx": "jsts", ".mjs": "jsts", ".cjs": "jsts", - ".ts": "jsts", ".tsx": "jsts", ".mts": "jsts", ".cts": "jsts", - ".vue": "jsts", ".svelte": "jsts", ".astro": "jsts", - # JVM interop - ".java": "jvm", ".kt": "jvm", ".kts": "jvm", - ".scala": "jvm", ".groovy": "jvm", ".gradle": "jvm", - # C-family: shared headers, Objective-C/C++ mix, Swift↔ObjC bridging - ".c": "native", ".h": "native", ".cpp": "native", ".cc": "native", - ".cxx": "native", ".hpp": "native", ".cu": "native", ".cuh": "native", - ".metal": "native", ".m": "native", ".mm": "native", ".swift": "native", - # Single-language families - ".py": "python", - ".go": "go", - ".rs": "rust", - ".rb": "ruby", ".rake": "ruby", - ".php": "php", ".phtml": "php", ".php3": "php", ".php4": "php", - ".php5": "php", ".php7": "php", ".phps": "php", - ".cs": "dotnet", ".razor": "dotnet", ".cshtml": "dotnet", ".xaml": "dotnet", - ".lua": "lua", ".luau": "lua", - ".zig": "zig", - ".ex": "elixir", ".exs": "elixir", - ".jl": "julia", - ".dart": "dart", - ".sh": "shell", ".bash": "shell", - ".ps1": "powershell", ".psm1": "powershell", ".psd1": "powershell", -} + text = _sv_strip_comments(raw) + # Consuming `endclass` (rather than a lookahead) makes each match own its + # terminator, so back-to-back or malformed classes cannot bleed bodies. + class_re = re.compile( + r"\b(?:(interface)\s+)?class\s+(\w+)([^;{]*)\s*;(.*?)\bendclass\b", + re.DOTALL, + ) + for match in class_re.finditer(text): + class_name = match.group(2) + header = match.group(3) or "" + body = match.group(4) or "" + line = line_for(match.start()) + # `#(type T = Payload)` declares `T` as a class type parameter, not a + # referenced type — collect these to skip below. + type_params = frozenset(re.findall(r"\btype\s+(\w+)", header)) + class_nid = _make_id(stem, class_name) + add_node(class_nid, class_name, line) + edges.append({"source": file_nid, "target": class_nid, "relation": "defines", + "confidence": "EXTRACTED", "confidence_score": 1.0, + "source_file": str_path, "source_location": f"L{line}", "weight": 1.0}) + + ext = re.search(r"\bextends\s+(\w+)", header) + if ext: + add_edge(class_nid, ext.group(1), "inherits", line) + impl = re.search(r"\bimplements\s+([^;{]+)", header) + if impl: + for iface_name in _sv_split_type_list(impl.group(1)): + add_edge(class_nid, iface_name.split("#", 1)[0].strip(), "implements", line) + + body_without_functions = re.sub( + r"\bfunction\b.*?\bendfunction\b", + lambda m: "\n" * m.group(0).count("\n"), + body, + flags=re.DOTALL, + ) + # Optional leading class-property qualifiers (rand/local/protected/etc.) + # must be consumed: otherwise a qualified field like `rand Config x;` + # (three tokens) fails the ` ;` shape and its type reference + # is silently dropped. + for field in re.finditer(r"^\s*(?:(?:rand|randc|local|protected|static|const|automatic|var)\s+)*([A-Za-z_]\w*(?:\s*#\s*\([^;]+?\))?)\s+\w+\s*;", body_without_functions, re.MULTILINE): + # Count to the start of the type token (group 1), not the match + # start: `^\s*` consumes the leading newline(s), so field.start() + # would resolve to the class's line instead of the field's. + field_line = line + body_without_functions.count("\n", 0, field.start(1)) + for ref_name, role in _sv_collect_type_refs(field.group(1), skip=type_params): + add_edge(class_nid, ref_name, "references", field_line, "generic_arg" if role == "generic_arg" else "field") + + for fm in _SV_FUNC_RE.finditer(body): + return_type, func_name, params = fm.group(1), fm.group(2), fm.group(3) + func_line = line + body.count("\n", 0, fm.start()) + func_nid = _make_id(class_nid, func_name) + add_node(func_nid, func_name, func_line) + edges.append({"source": class_nid, "target": func_nid, "relation": "method", + "confidence": "EXTRACTED", "confidence_score": 1.0, + "source_file": str_path, "source_location": f"L{func_line}", "weight": 1.0}) + for ref_name, role in _sv_collect_type_refs(return_type, skip=type_params): + add_edge(func_nid, ref_name, "references", func_line, "generic_arg" if role == "generic_arg" else "return_type") + for param in _sv_split_type_list(params): + pm = _SV_PARAM_RE.match(param) + if not pm: + continue + for ref_name, role in _sv_collect_type_refs(pm.group(1), skip=type_params): + add_edge(func_nid, ref_name, "references", func_line, "generic_arg" if role == "generic_arg" else "parameter_type") -def _lang_family(source_file: object) -> str | None: - """Interop family of the file's language, or None when unknown/not code.""" - if not source_file: - return None - return _LANG_FAMILY_BY_EXT.get(Path(str(source_file)).suffix.lower()) +def extract_verilog(path: Path) -> dict: + """Extract modules, functions, tasks, package imports, instantiations, and + SystemVerilog class semantics (inherits/implements edges, field/parameter/ + return-type references) from .v/.sv files.""" + try: + import tree_sitter_verilog as tsverilog + from tree_sitter import Language, Parser + except ImportError: + return {"nodes": [], "edges": [], "error": "tree_sitter_verilog not installed"} + try: + language = Language(tsverilog.language()) + parser = Parser(language) + source = path.read_bytes() + tree = parser.parse(source) + root = tree.root_node + except Exception as e: + return {"nodes": [], "edges": [], "error": str(e)} -def _node_label_key(node: dict, fold: bool = False) -> str: - label = str(node.get("label", "")).strip() - key = re.sub(r"[^a-zA-Z0-9]+", "", label) - return key.lower() if fold else key + stem = _file_stem(path) + str_path = str(path) + nodes: list[dict] = [] + edges: list[dict] = [] + seen_ids: set[str] = set() + + def add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({"id": nid, "label": label, "file_type": "code", + "source_file": str_path, "source_location": f"L{line}", + "confidence_score": 1.0}) + def add_edge(src: str, tgt: str, relation: str, line: int, + confidence: str = "EXTRACTED", score: float = 1.0) -> None: + edges.append({"source": src, "target": tgt, "relation": relation, + "confidence": confidence, "confidence_score": score, + "source_file": str_path, "source_location": f"L{line}", "weight": 1.0}) + + file_nid = _make_id(str(path)) + add_node(file_nid, path.name, 1) -def _is_top_level_function_definition(node: dict) -> bool: - """A free/top-level function def (label ``name()``), not a method or type. + def walk(node, module_nid: str | None = None) -> None: + t = node.type - Methods carry a leading dot (``.foo()``) or a qualifier (``Class.foo()``); - excluding those keeps a bare-name reference from binding to a receiver-scoped - method, which the receiver-typed resolvers own (#1781). - """ - label = str(node.get("label", "")).strip() - return ( - node.get("file_type") == "code" - and label.endswith(")") - and not label.startswith(".") - and "." not in label + # SystemVerilog class bodies are handled by _augment_systemverilog_semantics + # (regex over source text). Skip their subtrees so in-class methods are not + # double-emitted here — and with the wrong, return-type-derived name. + if t in ("class_declaration", "interface_class_declaration"): + return + + if t == "module_declaration": + mod_name = _sv_first_identifier(_sv_child(node, "module_header"), source) + if mod_name: + line = node.start_point[0] + 1 + nid = _make_id(stem, mod_name) + add_node(nid, mod_name, line) + add_edge(file_nid, nid, "defines", line) + for child in node.children: + walk(child, nid) + return + + # `function_prototype` only appears inside class/interface-class bodies + # (skipped above) and nests its name differently; it is intentionally not + # handled here. + elif t == "function_declaration": + fn_body = _sv_child(node, "function_body_declaration") + func_name = _sv_first_identifier(_sv_child(fn_body, "function_identifier"), source) + if func_name: + line = node.start_point[0] + 1 + parent = module_nid or file_nid + nid = _make_id(parent, func_name) + add_node(nid, f"{func_name}()", line) + add_edge(parent, nid, "contains", line) + + elif t == "task_declaration": + tk_body = _sv_child(node, "task_body_declaration") + task_name = _sv_first_identifier(_sv_child(tk_body, "task_identifier"), source) + if task_name: + line = node.start_point[0] + 1 + parent = module_nid or file_nid + nid = _make_id(parent, task_name) + add_node(nid, task_name, line) + add_edge(parent, nid, "contains", line) + + elif t == "package_import_declaration": + for child in node.children: + if child.type == "package_import_item": + pkg_text = _read_text(child, source) + pkg_name = pkg_text.split("::")[0].strip() + if pkg_name: + line = node.start_point[0] + 1 + tgt_nid = _make_id(pkg_name) + add_node(tgt_nid, pkg_name, line) + src_nid = module_nid or file_nid + add_edge(src_nid, tgt_nid, "imports_from", line) + + elif t in ("module_instantiation", "checker_instantiation"): + # `leaf u_leaf();` parses as checker_instantiation in 1.0.3; + # module_instantiation (when it occurs) exposes a `module_type` field. + # Both reduce to the first identifier under the node — the instantiated + # type, not the instance name (which appears later). + if module_nid: + type_node = node.child_by_field_name("module_type") + inst_type = (_read_text(type_node, source).strip() if type_node + else _sv_first_identifier(node, source)) + if inst_type: + line = node.start_point[0] + 1 + tgt_nid = _make_id(inst_type) + add_node(tgt_nid, inst_type, line) + add_edge(module_nid, tgt_nid, "instantiates", line) + + for child in node.children: + walk(child, module_nid) + + walk(root) + _augment_systemverilog_semantics( + source.decode("utf-8", errors="replace"), + stem, + str_path, + file_nid, + nodes, + edges, + seen_ids, ) + return {"nodes": nodes, "edges": edges} -def _rewire_unique_stub_nodes(nodes: list[dict], edges: list[dict]) -> None: - """Map unresolved no-source stubs to a unique real definition with the same label.""" - real_by_label: dict[str, list[dict]] = {} # exact-case type-like (all languages) - real_by_label_ci: dict[str, list[dict]] = {} # case-INSENSITIVE-language reals only - func_by_label: dict[str, list[dict]] = {} # top-level function defs (#1781) - stubs: list[dict] = [] +def extract_sql(path: Path, content: str | bytes | None = None) -> dict: + """Extract tables, views, functions, and relationships from .sql files via tree-sitter.""" + try: + import tree_sitter_sql as tssql + from tree_sitter import Language, Parser + except ImportError: + return {"nodes": [], "edges": [], "error": "tree_sitter_sql not installed. Run: pip install tree-sitter-sql"} - for node in nodes: + try: + language = Language(tssql.language()) + parser = Parser(language) + source = ( + content.encode("utf-8") if isinstance(content, str) + else content if content is not None + else path.read_bytes() + ) + tree = parser.parse(source) + root = tree.root_node + except Exception as e: + return {"nodes": [], "edges": [], "error": str(e)} + + + stem = _file_stem(path) + str_path = str(path) + file_nid = _make_id(str_path) + nodes: list[dict] = [{"id": file_nid, "label": path.name, "file_type": "code", + "source_file": str_path, "source_location": None}] + edges: list[dict] = [] + seen_ids: set[str] = {file_nid} + table_nids: dict[str, str] = {} # name → nid for reference resolution + + def _read(n) -> str: + return source[n.start_byte:n.end_byte].decode("utf-8", errors="replace") + + def _obj_name(n) -> str | None: + for c in n.children: + if c.type == "object_reference": + return _read(c) + return None + + def _add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({"id": nid, "label": label, "file_type": "code", + "source_file": str_path, "source_location": f"L{line}"}) + edges.append({"source": file_nid, "target": nid, "relation": "contains", + "confidence": "EXTRACTED", "source_file": str_path, + "source_location": f"L{line}", "weight": 1.0}) + + def _add_edge(src: str, tgt: str, relation: str, line: int) -> None: + edges.append({"source": src, "target": tgt, "relation": relation, + "confidence": "EXTRACTED", "source_file": str_path, + "source_location": f"L{line}", "weight": 1.0}) + + def walk(node) -> None: + t = node.type + line = node.start_point[0] + 1 + + if t == "create_table": + name = _obj_name(node) + if name: + nid = _make_id(stem, name) + _add_node(nid, name, line) + table_nids[name.lower()] = nid + # Foreign key REFERENCES + for col in node.children: + if col.type == "column_definitions": + has_error = any(cd.type == "ERROR" for cd in col.children) + seen_refs: set[str] = set() + for cd in col.children: + if cd.type == "column_definition": + # Inline column-level REFERENCES + ref_name: str | None = None + found_ref = False + for cc in cd.children: + if cc.type == "keyword_references": + found_ref = True + elif found_ref and cc.type == "object_reference": + ref_name = _read(cc) + break + if ref_name: + ref_nid = table_nids.get(ref_name.lower()) or _make_id(stem, ref_name) + _add_edge(nid, ref_nid, "references", line) + seen_refs.add(ref_name.lower()) + elif cd.type == "constraints": + # Table-level FOREIGN KEY ... REFERENCES ... constraints + for constraint in cd.children: + if constraint.type != "constraint": + continue + ref_name = None + found_ref = False + for cc in constraint.children: + if cc.type == "keyword_references": + found_ref = True + elif found_ref and cc.type == "object_reference": + ref_name = _read(cc) + break + if ref_name: + ref_nid = table_nids.get(ref_name.lower()) or _make_id(stem, ref_name) + _add_edge(nid, ref_nid, "references", line) + seen_refs.add(ref_name.lower()) + if has_error: + # Dialect-specific syntax (e.g. Firebird COMPUTED BY) causes ERROR + # nodes that make the parser drop the trailing constraints block. + # Regex-scan the raw column_definitions text as fallback. + col_text = _read(col) + for rm in re.finditer(r"\bREFERENCES\s+([\w$]+)", col_text, re.IGNORECASE): + ref_name = rm.group(1) + if ref_name.lower() not in seen_refs: + ref_nid = table_nids.get(ref_name.lower()) or _make_id(stem, ref_name) + _add_edge(nid, ref_nid, "references", line) + seen_refs.add(ref_name.lower()) + + elif t == "create_view": + name = _obj_name(node) + if name: + nid = _make_id(stem, name) + _add_node(nid, name, line) + table_nids[name.lower()] = nid + # FROM/JOIN table references inside view body + _walk_from_refs(node, nid, line) + + elif t == "create_function": + name = _obj_name(node) + if name: + nid = _make_id(stem, name) + _add_node(nid, f"{name}()", line) + _walk_from_refs(node, nid, line) + + elif t == "create_procedure": + name = _obj_name(node) + if name: + nid = _make_id(stem, name) + _add_node(nid, f"{name}()", line) + _walk_from_refs(node, nid, line) + + elif t == "alter_table": + name = _obj_name(node) + if name: + src_nid = table_nids.get(name.lower()) + if not src_nid: + src_nid = _make_id(stem, name) + _add_node(src_nid, name, line) + table_nids[name.lower()] = src_nid + for child in node.children: + if child.type == "add_constraint": + for cc in child.children: + if cc.type != "constraint": + continue + found_ref = False + ref_name: str | None = None + for ccc in cc.children: + if ccc.type == "keyword_references": + found_ref = True + elif found_ref and ccc.type == "object_reference": + ref_name = _read(ccc) + break + if ref_name: + ref_nid = table_nids.get(ref_name.lower()) + if not ref_nid: + ref_nid = _make_id(stem, ref_name) + _add_edge(src_nid, ref_nid, "references", line) + + elif t == "create_trigger": + trig_name: str | None = None + tbl_name: str | None = None + after_trigger = False + after_for = False + for c in node.children: + if c.type == "keyword_trigger": + after_trigger = True + elif after_trigger and not trig_name and c.type == "object_reference": + trig_name = _read(c) + elif c.type == "keyword_for": + after_for = True + elif after_for and not tbl_name and c.type == "object_reference": + tbl_name = _read(c) + if trig_name: + trig_nid = _make_id(stem, trig_name) + _add_node(trig_nid, trig_name, line) + if tbl_name: + tbl_nid = table_nids.get(tbl_name.lower()) or _make_id(stem, tbl_name) + _add_edge(trig_nid, tbl_nid, "triggers", line) + + elif t == "fb_proc_or_trigger": + text = _read(node) + m = re.match( + r"CREATE\s+(?:OR\s+(?:REPLACE|ALTER)\s+)?" + r"(PROCEDURE|TRIGGER|FUNCTION)\s+([\w$]+)", + text, re.IGNORECASE, + ) + if m: + obj_type = m.group(1).upper() + obj_name = m.group(2) + obj_nid = _make_id(stem, obj_name) + label = obj_name if obj_type == "TRIGGER" else f"{obj_name}()" + _add_node(obj_nid, label, line) + if obj_type == "TRIGGER": + fm = re.search(r"\bFOR\s+([\w$]+)", text, re.IGNORECASE) + if fm: + tbl = fm.group(1) + tbl_nid = table_nids.get(tbl.lower()) or _make_id(stem, tbl) + _add_edge(obj_nid, tbl_nid, "triggers", line) + _NON_TABLES = { + "select", "where", "set", "dual", "null", "true", "false", + "first", "skip", "rows", "next", "only", "lateral", + } + seen_tbls: set[str] = set() + for rm in re.finditer(r"\b(?:FROM|JOIN|INTO)\s+([\w$]+)", text, re.IGNORECASE): + tbl = rm.group(1) + if tbl.lower() not in _NON_TABLES and tbl.lower() not in seen_tbls: + seen_tbls.add(tbl.lower()) + tbl_nid = table_nids.get(tbl.lower()) or _make_id(stem, tbl) + _add_edge(obj_nid, tbl_nid, "reads_from", line) + for rm in re.finditer(r"\bUPDATE\s+([\w$]+)", text, re.IGNORECASE): + tbl = rm.group(1) + if tbl.lower() not in _NON_TABLES and tbl.lower() not in seen_tbls: + seen_tbls.add(tbl.lower()) + tbl_nid = table_nids.get(tbl.lower()) or _make_id(stem, tbl) + _add_edge(obj_nid, tbl_nid, "reads_from", line) + + for child in node.children: + walk(child) + + def _walk_from_refs(node, caller_nid: str, line: int) -> None: + """Recursively find FROM/JOIN table references inside a node.""" + if node.type in ("from", "join"): + for c in node.children: + if c.type == "relation": + for cc in c.children: + if cc.type == "object_reference": + tbl = _read(cc) + tbl_nid = _make_id(stem, tbl) + _add_edge(caller_nid, tbl_nid, "reads_from", + c.start_point[0] + 1) + for child in node.children: + _walk_from_refs(child, caller_nid, line) + + for stmt in root.children: + if stmt.type == "statement": + for child in stmt.children: + walk(child) + elif stmt.type in ("fb_proc_or_trigger", "set_term", "declare_external_function"): + walk(stmt) + + # Global regex fallback: catch any REFERENCES missed due to ERROR nodes in the parse tree + # (e.g. Firebird COMPUTED BY columns push constraints out of the tree entirely). + # Snapshot after tree walk so we don't re-emit edges already captured above. + emitted = {(e["source"], e["target"]) for e in edges if e["relation"] == "references"} + src_text = source.decode("utf-8", errors="replace") + for m in re.finditer(r"CREATE\s+TABLE\s+([\w$]+)\s*\(", src_text, re.IGNORECASE): + tbl_name = m.group(1) + tbl_nid = table_nids.get(tbl_name.lower()) + if tbl_nid is None: + continue + tbl_line = src_text[: m.start()].count("\n") + 1 + tail = src_text[m.start():] + end = re.search(r"(?:^|\n)(?:CREATE|SET\s+TERM|ALTER)\s", tail[1:], re.IGNORECASE) + block = tail[: end.start() + 1] if end else tail + for rm in re.finditer(r"\bREFERENCES\s+([\w$]+)", block, re.IGNORECASE): + ref_name = rm.group(1) + ref_nid = table_nids.get(ref_name.lower()) or _make_id(stem, ref_name) + if (tbl_nid, ref_nid) not in emitted: + _add_edge(tbl_nid, ref_nid, "references", tbl_line) + emitted.add((tbl_nid, ref_nid)) + + return {"nodes": nodes, "edges": edges} + + +def extract_lua(path: Path) -> dict: + """Extract functions, methods, require() imports, and calls from a .lua file.""" + return _extract_generic(path, _LUA_CONFIG) + + +def extract_swift(path: Path) -> dict: + """Extract classes, structs, protocols, functions, imports, and calls from a .swift file.""" + return _extract_generic(path, _SWIFT_CONFIG) + + +# ── Julia extractor (custom walk) ──────────────────────────────────────────── + +def extract_julia(path: Path) -> dict: + """Extract modules, structs, functions, imports, and calls from a .jl file.""" + try: + import tree_sitter_julia as tsjulia + from tree_sitter import Language, Parser + except ImportError: + return {"nodes": [], "edges": [], "error": "tree-sitter-julia not installed"} + + try: + language = Language(tsjulia.language()) + parser = Parser(language) + source = path.read_bytes() + tree = parser.parse(source) + root = tree.root_node + except Exception as e: + return {"nodes": [], "edges": [], "error": str(e)} + + stem = _file_stem(path) + str_path = str(path) + nodes: list[dict] = [] + edges: list[dict] = [] + seen_ids: set[str] = set() + function_bodies: list[tuple[str, object]] = [] + + def add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": label, + "file_type": "code", + "source_file": str_path, + "source_location": f"L{line}", + }) + + def add_edge(src: str, tgt: str, relation: str, line: int, + confidence: str = "EXTRACTED", weight: float = 1.0, + context: str | None = None) -> None: + edge = { + "source": src, + "target": tgt, + "relation": relation, + "confidence": confidence, + "source_file": str_path, + "source_location": f"L{line}", + "weight": weight, + } + if context: + edge["context"] = context + edges.append(edge) + + file_nid = _make_id(str(path)) + add_node(file_nid, path.name, 1) + + def ensure_named_node(name: str, line: int) -> str: + nid = _make_id(stem, name) + if nid in seen_ids: + return nid + nid = _make_id(name) + if nid not in seen_ids: + # The name isn't defined in this file, so this is a cross-file reference + # (e.g. a `Thing` type annotation imported from another module). Emit a + # SOURCELESS stub — like the inheritance-base path below — so the + # corpus-level rewire can collapse it onto the real definition. A sourced + # stub here makes _disambiguate_colliding_node_ids bake the referencing + # file's path (with extension) into the id and blocks the rewire, which is + # the phantom-duplicate-node bug (#1402). + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": name, + "file_type": "code", + "source_file": "", + "source_location": "", + "origin_file": str_path, + }) + return nid + + def _func_name_from_signature(sig_node) -> str | None: + """Extract function name from a Julia signature node (call_expression > identifier).""" + for child in sig_node.children: + if child.type == "call_expression": + callee = child.children[0] if child.children else None + if callee and callee.type == "identifier": + return _read_text(callee, source) + return None + + def walk_calls(body_node, func_nid: str) -> None: + if body_node is None: + return + t = body_node.type + if t in ("function_definition", "short_function_definition"): + return + if t == "call_expression" and body_node.children: + callee = body_node.children[0] + # Direct call: foo(...) + if callee.type == "identifier": + callee_name = _read_text(callee, source) + target_nid = _make_id(stem, callee_name) + add_edge(func_nid, target_nid, "calls", body_node.start_point[0] + 1, + confidence="EXTRACTED", context="call") + # Method call: obj.method(...) + elif callee.type == "field_expression" and len(callee.children) >= 3: + method_node = callee.children[-1] + method_name = _read_text(method_node, source) + target_nid = _make_id(stem, method_name) + add_edge(func_nid, target_nid, "calls", body_node.start_point[0] + 1, + confidence="EXTRACTED", context="call") + for child in body_node.children: + walk_calls(child, func_nid) + + def walk(node, scope_nid: str) -> None: + t = node.type + + # Module + if t == "module_definition": + name_node = next((c for c in node.children if c.type == "identifier"), None) + if name_node: + mod_name = _read_text(name_node, source) + mod_nid = _make_id(stem, mod_name) + line = node.start_point[0] + 1 + add_node(mod_nid, mod_name, line) + add_edge(file_nid, mod_nid, "defines", line) + for child in node.children: + walk(child, mod_nid) + return + + # Struct (struct / mutable struct — both map to struct_definition in tree-sitter-julia) + if t == "struct_definition": + # type_head may contain: identifier (simple) or binary_expression (Foo <: Bar) + type_head = next((c for c in node.children if c.type == "type_head"), None) + if not type_head: + return + struct_name: str | None = None + super_name: str | None = None + bin_expr = next((c for c in type_head.children if c.type == "binary_expression"), None) + if bin_expr: + identifiers = [c for c in bin_expr.children if c.type == "identifier"] + if identifiers: + struct_name = _read_text(identifiers[0], source) + if len(identifiers) >= 2: + super_name = _read_text(identifiers[-1], source) + else: + name_node = next((c for c in type_head.children if c.type == "identifier"), None) + if name_node: + struct_name = _read_text(name_node, source) + if not struct_name: + return + struct_nid = _make_id(stem, struct_name) + line = node.start_point[0] + 1 + add_node(struct_nid, struct_name, line) + add_edge(scope_nid, struct_nid, "defines", line) + if super_name: + add_edge(struct_nid, ensure_named_node(super_name, line), + "inherits", line, confidence="EXTRACTED") + # Field types: each `name::Type` lowers to a typed_expression child of struct_definition + for child in node.children: + if child.type == "typed_expression": + type_ids = [c for c in child.children if c.type == "identifier"] + if len(type_ids) >= 2: + field_line = child.start_point[0] + 1 + type_name = _read_text(type_ids[-1], source) + type_nid = ensure_named_node(type_name, field_line) + edges.append(_semantic_reference_edge( + struct_nid, type_nid, "field", str_path, field_line)) + return + + # Abstract type + if t == "abstract_definition": + # type_head > identifier + type_head = next((c for c in node.children if c.type == "type_head"), None) + if type_head: + name_node = next((c for c in type_head.children if c.type == "identifier"), None) + if name_node: + abs_name = _read_text(name_node, source) + abs_nid = _make_id(stem, abs_name) + line = node.start_point[0] + 1 + add_node(abs_nid, abs_name, line) + add_edge(scope_nid, abs_nid, "defines", line) + return + + # Function: function foo(...) ... end + if t == "function_definition": + sig_node = next((c for c in node.children if c.type == "signature"), None) + if sig_node: + func_name = _func_name_from_signature(sig_node) + if func_name: + func_nid = _make_id(stem, func_name) + line = node.start_point[0] + 1 + add_node(func_nid, f"{func_name}()", line) + add_edge(scope_nid, func_nid, "defines", line) + function_bodies.append((func_nid, node)) + return + + # Short function: foo(x) = expr + if t == "assignment": + lhs = node.children[0] if node.children else None + if lhs and lhs.type == "call_expression" and lhs.children: + callee = lhs.children[0] + if callee.type == "identifier": + func_name = _read_text(callee, source) + func_nid = _make_id(stem, func_name) + line = node.start_point[0] + 1 + add_node(func_nid, f"{func_name}()", line) + add_edge(scope_nid, func_nid, "defines", line) + # Only walk the RHS (index 2 after lhs and operator) to avoid self-loops + rhs = node.children[-1] if len(node.children) >= 3 else None + if rhs: + function_bodies.append((func_nid, rhs)) + return + + # Using / Import + if t in ("using_statement", "import_statement"): + line = node.start_point[0] + 1 + + def _julia_mod_name(n): + # identifier (`Foo`), scoped_identifier (`Base.Threads`), or + # import_path (relative `..Sibling`) -> the module name. Only bare + # identifiers were handled, so qualified/relative imports — and the + # scoped package of a `selected_import` — were silently dropped. + if n.type == "import_path": + ids = [c for c in n.children if c.type == "identifier"] + return _read_text(ids[-1], source) if ids else None + if n.type in ("identifier", "scoped_identifier"): + return _read_text(n, source) + return None + + def _emit_import(name): + if not name: + return + imp_nid = _make_id(name) + add_node(imp_nid, name, line) + add_edge(scope_nid, imp_nid, "imports", line, context="import") + + for child in node.children: + if child.type in ("identifier", "scoped_identifier", "import_path"): + _emit_import(_julia_mod_name(child)) + elif child.type == "selected_import": + # `import Base.Threads: nthreads` — the package (first named + # child) may itself be a scoped_identifier/import_path. + pkg = next( + (c for c in child.children + if c.type in ("identifier", "scoped_identifier", "import_path")), + None, + ) + if pkg is not None: + _emit_import(_julia_mod_name(pkg)) + return + + for child in node.children: + walk(child, scope_nid) + + walk(root, file_nid) + + for func_nid, body_node in function_bodies: + # For function_definition nodes, walk children directly to avoid + # the boundary check returning early on the top-level node itself. + # Skip the "signature" child — it contains the function's own call_expression + # which would create a self-loop. + if body_node.type == "function_definition": + for child in body_node.children: + if child.type != "signature": + walk_calls(child, func_nid) + else: + walk_calls(body_node, func_nid) + + return {"nodes": nodes, "edges": edges} + + +_FORTRAN_CPP_EXTS = {".F", ".F90", ".F95", ".F03", ".F08"} + + +def _cpp_preprocess(path: Path) -> bytes: + """Run cpp -w -P on a capital-F Fortran file and return preprocessed bytes. + + Falls back to raw file bytes if cpp is not available. Capital-F extensions + conventionally require C preprocessor expansion (#ifdef MPI, #define REAL8, etc.) + before parsing. + + Security (F-007): we pass `-nostdinc` and `-I /dev/null` so a malicious + source file containing `#include "/home/victim/.ssh/id_rsa"` (or any other + include directive) cannot inline arbitrary host files into the output that + we then ship to an LLM. Without these flags `cpp` happily resolves any + relative or absolute include path it can read, which is a corpus-side + file-exfiltration vector. + """ + import shutil + import subprocess + if not shutil.which("cpp"): + return path.read_bytes() + try: + # Pass an absolute path so a corpus file named like "-I/etc/x.F90" cannot + # be parsed by cpp as an option (cpp does not accept a "--" end-of-options + # terminator). An absolute path always begins with "/". + result = subprocess.run( + ["cpp", "-w", "-P", "-nostdinc", "-I", "/dev/null", str(path.resolve())], + capture_output=True, + timeout=30, + ) + if result.returncode == 0 and result.stdout: + return result.stdout + except Exception: + pass + return path.read_bytes() + + +def extract_fortran(path: Path) -> dict: + """Extract programs, modules, subroutines, functions, use statements, and calls from Fortran files. + + Capital-F extensions (.F, .F90, etc.) are run through the C preprocessor before + parsing so #ifdef/#define macros are resolved. + """ + try: + import tree_sitter_fortran as tsfortran + from tree_sitter import Language, Parser + except ImportError: + return {"nodes": [], "edges": [], "error": "tree-sitter-fortran not installed"} + + try: + language = Language(tsfortran.language()) + parser = Parser(language) + source = _cpp_preprocess(path) if path.suffix in _FORTRAN_CPP_EXTS else path.read_bytes() + tree = parser.parse(source) + root = tree.root_node + except Exception as e: + return {"nodes": [], "edges": [], "error": str(e)} + + stem = _file_stem(path) + str_path = str(path) + nodes: list[dict] = [] + edges: list[dict] = [] + seen_ids: set[str] = set() + scope_bodies: list[tuple[str, object]] = [] + + def add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": label, + "file_type": "code", + "source_file": str_path, + "source_location": f"L{line}", + }) + + def add_edge(src: str, tgt: str, relation: str, line: int, + confidence: str = "EXTRACTED", weight: float = 1.0, + context: str | None = None) -> None: + edge = { + "source": src, + "target": tgt, + "relation": relation, + "confidence": confidence, + "source_file": str_path, + "source_location": f"L{line}", + "weight": weight, + } + if context: + edge["context"] = context + edges.append(edge) + + file_nid = _make_id(str(path)) + add_node(file_nid, path.name, 1) + + def _fortran_name(stmt_node) -> str | None: + """Extract name from a *_statement node. Fortran is case-insensitive; lowercase.""" + for child in stmt_node.children: + if child.type in ("name", "identifier"): + return _read_text(child, source).lower() + return None + + def ensure_named_node(name: str, line: int) -> str: + nid = _make_id(stem, name) + if nid in seen_ids: + return nid + nid = _make_id(name) + if nid not in seen_ids: + # The name isn't defined in this file, so this is a cross-file reference + # (e.g. a `Thing` type annotation imported from another module). Emit a + # SOURCELESS stub — like the inheritance-base path below — so the + # corpus-level rewire can collapse it onto the real definition. A sourced + # stub here makes _disambiguate_colliding_node_ids bake the referencing + # file's path (with extension) into the id and blocks the rewire, which is + # the phantom-duplicate-node bug (#1402). + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": name, + "file_type": "code", + "source_file": "", + "source_location": "", + "origin_file": str_path, + }) + return nid + + def emit_signature_refs(scope_node, fn_nid: str, is_function: bool) -> None: + """Emit references[parameter_type] / references[return_type] edges for + a subroutine/function based on its variable_declaration siblings.""" + stmt_type = "function_statement" if is_function else "subroutine_statement" + stmt = next((c for c in scope_node.children if c.type == stmt_type), None) + if stmt is None: + return + param_names: set[str] = set() + params_node = next((c for c in stmt.children if c.type == "parameters"), None) + if params_node is not None: + for c in params_node.children: + if c.type == "identifier": + param_names.add(_read_text(c, source).lower()) + result_name: str | None = None + if is_function: + result_node = next((c for c in stmt.children if c.type == "function_result"), None) + if result_node is not None: + res_id = next((c for c in result_node.children if c.type == "identifier"), None) + if res_id is not None: + result_name = _read_text(res_id, source).lower() + else: + # implicit result variable: same name as the function + result_name = _fortran_name(stmt) + for child in scope_node.children: + if child.type != "variable_declaration": + continue + derived = next((c for c in child.children if c.type == "derived_type"), None) + if derived is None: + continue + type_name_node = next((c for c in derived.children if c.type == "type_name"), None) + if type_name_node is None: + continue + type_name = _read_text(type_name_node, source).lower() + for var in child.children: + if var.type != "identifier": + continue + var_name = _read_text(var, source).lower() + var_line = var.start_point[0] + 1 + if var_name in param_names: + tgt = ensure_named_node(type_name, var_line) + if tgt != fn_nid: + add_edge(fn_nid, tgt, "references", var_line, context="parameter_type") + elif is_function and var_name == result_name: + tgt = ensure_named_node(type_name, var_line) + if tgt != fn_nid: + add_edge(fn_nid, tgt, "references", var_line, context="return_type") + + def walk_calls(node, scope_nid: str) -> None: + if node is None: + return + t = node.type + if t in ("subroutine", "function", "module", "program", "internal_procedures"): + return + # call FOO(args) — tree-sitter-fortran uses subroutine_call + if t == "subroutine_call": + name_node = next((c for c in node.children if c.type == "identifier"), None) + if name_node: + callee = _read_text(name_node, source).lower() + target_nid = _make_id(stem, callee) + add_edge(scope_nid, target_nid, "calls", node.start_point[0] + 1, + confidence="EXTRACTED", context="call") + # x = compute(args) — function invocations are `call_expression`, which + # shares Fortran's `name(...)` syntax with array indexing. Only emit a + # call edge when the callee resolves to a procedure defined in this file + # (an array variable produces no matching node), so array accesses can't + # fabricate spurious `calls` edges. + elif t == "call_expression": + name_node = next((c for c in node.children if c.type == "identifier"), None) + if name_node: + callee = _read_text(name_node, source).lower() + target_nid = _make_id(stem, callee) + if target_nid in seen_ids and target_nid != scope_nid: + add_edge(scope_nid, target_nid, "calls", node.start_point[0] + 1, + confidence="EXTRACTED", context="call") + for child in node.children: + walk_calls(child, scope_nid) + + def walk(node, scope_nid: str) -> None: + t = node.type + + if t == "program": + stmt = next((c for c in node.children if c.type == "program_statement"), None) + name = _fortran_name(stmt) if stmt else None + if name: + nid = _make_id(stem, name) + line = node.start_point[0] + 1 + add_node(nid, name, line) + add_edge(file_nid, nid, "defines", line) + scope_bodies.append((nid, node)) + for child in node.children: + walk(child, nid) + return + + if t == "module": + stmt = next((c for c in node.children if c.type == "module_statement"), None) + name = _fortran_name(stmt) if stmt else None + if name: + nid = _make_id(stem, name) + line = node.start_point[0] + 1 + add_node(nid, name, line) + add_edge(file_nid, nid, "defines", line) + for child in node.children: + walk(child, nid) + return + + # subroutines/functions inside a module live under internal_procedures + if t == "internal_procedures": + for child in node.children: + walk(child, scope_nid) + return + + if t == "derived_type_definition": + stmt = next((c for c in node.children if c.type == "derived_type_statement"), None) + if stmt is not None: + name_node = next((c for c in stmt.children if c.type == "type_name"), None) + if name_node is not None: + type_name = _read_text(name_node, source).lower() + type_nid = _make_id(stem, type_name) + line = node.start_point[0] + 1 + add_node(type_nid, type_name, line) + add_edge(scope_nid, type_nid, "defines", line) + return + + if t == "subroutine": + stmt = next((c for c in node.children if c.type == "subroutine_statement"), None) + name = _fortran_name(stmt) if stmt else None + if name: + nid = _make_id(stem, name) + line = node.start_point[0] + 1 + add_node(nid, f"{name}()", line) + add_edge(scope_nid, nid, "defines", line) + scope_bodies.append((nid, node)) + emit_signature_refs(node, nid, is_function=False) + for child in node.children: + walk(child, nid) + return + + if t == "function": + stmt = next((c for c in node.children if c.type == "function_statement"), None) + name = _fortran_name(stmt) if stmt else None + if name: + nid = _make_id(stem, name) + line = node.start_point[0] + 1 + add_node(nid, f"{name}()", line) + add_edge(scope_nid, nid, "defines", line) + scope_bodies.append((nid, node)) + emit_signature_refs(node, nid, is_function=True) + for child in node.children: + walk(child, nid) + return + + if t == "use_statement": + line = node.start_point[0] + 1 + # tree-sitter-fortran uses module_name node for the used module + name_node = next((c for c in node.children if c.type in ("module_name", "name", "identifier")), None) + if name_node: + mod_name = _read_text(name_node, source).lower() + imp_nid = _make_id(mod_name) + add_node(imp_nid, mod_name, line) + add_edge(scope_nid, imp_nid, "imports", line, context="use") + return + + for child in node.children: + walk(child, scope_nid) + + walk(root, file_nid) + + _stmt_headers = { + "subroutine_statement", "function_statement", + "program_statement", "module_statement", + } + for scope_nid, body_node in scope_bodies: + for child in body_node.children: + if child.type not in _stmt_headers: + walk_calls(child, scope_nid) + + return {"nodes": nodes, "edges": edges} + + +# ── Go extractor (custom walk) ──────────────────────────────────────────────── + +def extract_go(path: Path) -> dict: + """Extract functions, methods, type declarations, and imports from a .go file.""" + try: + import tree_sitter_go as tsgo + from tree_sitter import Language, Parser + except ImportError: + return {"nodes": [], "edges": [], "error": "tree-sitter-go not installed"} + + try: + language = Language(tsgo.language()) + parser = Parser(language) + source = path.read_bytes() + tree = parser.parse(source) + root = tree.root_node + except Exception as e: + return {"nodes": [], "edges": [], "error": str(e)} + + stem = _file_stem(path) + # Use directory name as package scope so methods on the same type across + # multiple files in a package share one canonical type node. + pkg_scope = path.parent.name or stem + str_path = str(path) + nodes: list[dict] = [] + edges: list[dict] = [] + seen_ids: set[str] = set() + function_bodies: list[tuple[str, object]] = [] + go_imported_pkgs: set[str] = set() # local names of imported packages + + def add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": label, + "file_type": "code", + "source_file": str_path, + "source_location": f"L{line}", + }) + + def add_edge(src: str, tgt: str, relation: str, line: int, + confidence: str = "EXTRACTED", weight: float = 1.0, + context: str | None = None) -> None: + edge = { + "source": src, + "target": tgt, + "relation": relation, + "confidence": confidence, + "source_file": str_path, + "source_location": f"L{line}", + "weight": weight, + } + if context: + edge["context"] = context + edges.append(edge) + + file_nid = _make_id(str(path)) + add_node(file_nid, path.name, 1) + + def ensure_named_node(name: str, line: int) -> str: + nid = _make_id(pkg_scope, name) + if nid in seen_ids: + return nid + nid = _make_id(name) + if nid not in seen_ids: + # The name isn't declared in this file, so this is a cross-file reference + # (e.g. a type defined in another file of the package). Emit a SOURCELESS + # stub — like the inheritance-base path in the other extractors — so the + # corpus-level rewire can collapse it onto the real definition. A sourced + # stub here makes _disambiguate_colliding_node_ids bake the referencing + # file's path (with extension) into the id and blocks the rewire, which is + # the phantom-duplicate-node bug (#1402). + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": name, + "file_type": "code", + "source_file": "", + "source_location": "", + "origin_file": str_path, + }) + return nid + + def emit_go_method_refs(func_node, func_nid: str, line: int) -> None: + params = func_node.child_by_field_name("parameters") + if params is not None: + for p in params.children: + if p.type != "parameter_declaration": + continue + type_node = p.child_by_field_name("type") + refs: list[tuple[str, str]] = [] + _go_collect_type_refs(type_node, source, False, refs) + for ref_name, role in refs: + ctx = "generic_arg" if role == "generic_arg" else "parameter_type" + tgt = ensure_named_node(ref_name, line) + if tgt != func_nid: + add_edge(func_nid, tgt, "references", line, context=ctx) + result = func_node.child_by_field_name("result") + if result is not None: + if result.type == "parameter_list": + for p in result.children: + if p.type != "parameter_declaration": + continue + type_node = p.child_by_field_name("type") + if type_node is None: + for c in p.children: + if c.is_named: + type_node = c + break + refs = [] + _go_collect_type_refs(type_node, source, False, refs) + for ref_name, role in refs: + ctx = "generic_arg" if role == "generic_arg" else "return_type" + tgt = ensure_named_node(ref_name, line) + if tgt != func_nid: + add_edge(func_nid, tgt, "references", line, context=ctx) + else: + refs = [] + _go_collect_type_refs(result, source, False, refs) + for ref_name, role in refs: + ctx = "generic_arg" if role == "generic_arg" else "return_type" + tgt = ensure_named_node(ref_name, line) + if tgt != func_nid: + add_edge(func_nid, tgt, "references", line, context=ctx) + + def walk(node) -> None: + t = node.type + + if t == "function_declaration": + name_node = node.child_by_field_name("name") + if name_node: + func_name = _read_text(name_node, source) + line = node.start_point[0] + 1 + func_nid = _make_id(stem, func_name) + add_node(func_nid, f"{func_name}()", line) + add_edge(file_nid, func_nid, "contains", line) + emit_go_method_refs(node, func_nid, line) + body = node.child_by_field_name("body") + if body: + function_bodies.append((func_nid, body)) + return + + if t == "method_declaration": + receiver = node.child_by_field_name("receiver") + receiver_type: str | None = None + if receiver: + for param in receiver.children: + if param.type == "parameter_declaration": + type_node = param.child_by_field_name("type") + if type_node: + receiver_type = _read_text(type_node, source).lstrip("*").strip() + break + name_node = node.child_by_field_name("name") + if not name_node: + return + method_name = _read_text(name_node, source) + line = node.start_point[0] + 1 + + if receiver_type: + parent_nid = _make_id(pkg_scope, receiver_type) + add_node(parent_nid, receiver_type, line) + method_nid = _make_id(parent_nid, method_name) + add_node(method_nid, f".{method_name}()", line) + add_edge(parent_nid, method_nid, "method", line) + else: + method_nid = _make_id(stem, method_name) + add_node(method_nid, f"{method_name}()", line) + add_edge(file_nid, method_nid, "contains", line) + + emit_go_method_refs(node, method_nid, line) + body = node.child_by_field_name("body") + if body: + function_bodies.append((method_nid, body)) + return + + if t == "type_declaration": + for child in node.children: + if child.type != "type_spec": + continue + name_node = child.child_by_field_name("name") + if not name_node: + continue + type_name = _read_text(name_node, source) + line = child.start_point[0] + 1 + type_nid = _make_id(pkg_scope, type_name) + add_node(type_nid, type_name, line) + add_edge(file_nid, type_nid, "contains", line) + # Type body: struct fields (with embeds) or interface embedding. + type_body = None + for tc in child.children: + if tc.type in ("struct_type", "interface_type"): + type_body = tc + break + if type_body is None: + continue + if type_body.type == "struct_type": + for fdl in type_body.children: + if fdl.type != "field_declaration_list": + continue + for field in fdl.children: + if field.type != "field_declaration": + continue + has_name = any( + fc.type == "field_identifier" for fc in field.children + ) + type_node = field.child_by_field_name("type") + if type_node is None: + for fc in field.children: + if fc.is_named and fc.type != "field_identifier": + type_node = fc + break + refs: list[tuple[str, str]] = [] + _go_collect_type_refs(type_node, source, False, refs) + for ref_name, role in refs: + tgt = ensure_named_node(ref_name, field.start_point[0] + 1) + if tgt == type_nid: + continue + if not has_name and role == "type": + add_edge(type_nid, tgt, "embeds", + field.start_point[0] + 1) + else: + ctx = "generic_arg" if role == "generic_arg" else "field" + add_edge(type_nid, tgt, "references", + field.start_point[0] + 1, context=ctx) + elif type_body.type == "interface_type": + for elem in type_body.children: + if elem.type != "type_elem": + continue + refs = [] + for sub in elem.children: + if sub.is_named: + _go_collect_type_refs(sub, source, False, refs) + for ref_name, role in refs: + tgt = ensure_named_node(ref_name, elem.start_point[0] + 1) + if tgt == type_nid: + continue + if role == "type": + add_edge(type_nid, tgt, "embeds", + elem.start_point[0] + 1) + else: + add_edge(type_nid, tgt, "references", + elem.start_point[0] + 1, context="generic_arg") + return + + if t == "import_declaration": + for child in node.children: + if child.type == "import_spec_list": + for spec in child.children: + if spec.type == "import_spec": + path_node = spec.child_by_field_name("path") + if path_node: + raw = _read_text(path_node, source).strip('"') + # Prefix with go_pkg_ so stdlib names (e.g. "context") + # don't collide with local files of the same basename. + tgt_nid = _make_id("go", "pkg", raw) + add_edge(file_nid, tgt_nid, "imports_from", spec.start_point[0] + 1, context="import") + # Track local name (alias or last path segment) + alias = spec.child_by_field_name("name") + local_name = _read_text(alias, source) if alias else raw.split("/")[-1] + if local_name and local_name != "_" and local_name != ".": + go_imported_pkgs.add(local_name) + elif child.type == "import_spec": + path_node = child.child_by_field_name("path") + if path_node: + raw = _read_text(path_node, source).strip('"') + tgt_nid = _make_id("go", "pkg", raw) + add_edge(file_nid, tgt_nid, "imports_from", child.start_point[0] + 1, context="import") + alias = child.child_by_field_name("name") + local_name = _read_text(alias, source) if alias else raw.split("/")[-1] + if local_name and local_name != "_" and local_name != ".": + go_imported_pkgs.add(local_name) + return + + for child in node.children: + walk(child) + + walk(root) + + label_to_nid: dict[str, str] = {} + for n in nodes: + raw = n["label"] + normalised = raw.strip("()").lstrip(".") + label_to_nid[normalised] = n["id"] + + seen_call_pairs: set[tuple[str, str]] = set() + raw_calls: list[dict] = [] + + def walk_calls(node, caller_nid: str) -> None: + if node.type in ("function_declaration", "method_declaration"): + return + if node.type == "call_expression": + func_node = node.child_by_field_name("function") + callee_name: str | None = None + is_member_call: bool = False + if func_node: + if func_node.type == "identifier": + callee_name = _read_text(func_node, source) + elif func_node.type == "selector_expression": + field = func_node.child_by_field_name("field") + operand = func_node.child_by_field_name("operand") + receiver_name = _read_text(operand, source) if operand else "" + # Package-qualified call (e.g. fmt.Println) → allow cross-file resolution. + # Receiver method call (e.g. s.logger.Log) → skip, no import evidence. + is_member_call = receiver_name not in go_imported_pkgs + if field: + callee_name = _read_text(field, source) + if callee_name and callee_name not in _LANGUAGE_BUILTIN_GLOBALS: + tgt_nid = label_to_nid.get(callee_name) + if tgt_nid and tgt_nid != caller_nid: + pair = (caller_nid, tgt_nid) + if pair not in seen_call_pairs: + seen_call_pairs.add(pair) + line = node.start_point[0] + 1 + edges.append({ + "source": caller_nid, + "target": tgt_nid, + "relation": "calls", + "context": "call", + "confidence": "EXTRACTED", + "source_file": str_path, + "source_location": f"L{line}", + "weight": 1.0, + }) + elif callee_name: + raw_calls.append({ + "caller_nid": caller_nid, + "callee": callee_name, + "is_member_call": is_member_call, + "source_file": str_path, + "source_location": f"L{node.start_point[0] + 1}", + }) + for child in node.children: + walk_calls(child, caller_nid) + + for caller_nid, body_node in function_bodies: + walk_calls(body_node, caller_nid) + + valid_ids = seen_ids + clean_edges = [] + for edge in edges: + src, tgt = edge["source"], edge["target"] + if src in valid_ids and (tgt in valid_ids or edge["relation"] in ("imports", "imports_from")): + clean_edges.append(edge) + + return {"nodes": nodes, "edges": clean_edges, "raw_calls": raw_calls} + + +# ── Rust extractor (custom walk) ────────────────────────────────────────────── + +# Common Rust trait/stdlib method names that appear in virtually every codebase. +# Resolving these cross-file produces spurious INFERRED edges across crate +# boundaries (issue #908) — skip them from the unresolved-call queue entirely. +_RUST_TRAIT_METHOD_BLOCKLIST: frozenset[str] = frozenset({ + "new", "default", "parse", "from_str", "now", "clone", "into", "from", + "to_string", "to_owned", "len", "is_empty", "iter", "next", "build", + "start", "run", "init", "app", "get", "set", "push", "pop", "insert", + "remove", "contains", "collect", "map", "filter", "unwrap", "expect", + "ok", "err", "some", "none", "send", "recv", "lock", "read", "write", +}) + +def extract_rust(path: Path) -> dict: + """Extract functions, structs, enums, traits, impl methods, and use declarations from a .rs file.""" + try: + import tree_sitter_rust as tsrust + from tree_sitter import Language, Parser + except ImportError: + return {"nodes": [], "edges": [], "error": "tree-sitter-rust not installed"} + + try: + language = Language(tsrust.language()) + parser = Parser(language) + source = path.read_bytes() + tree = parser.parse(source) + root = tree.root_node + except Exception as e: + return {"nodes": [], "edges": [], "error": str(e)} + + stem = _file_stem(path) + str_path = str(path) + nodes: list[dict] = [] + edges: list[dict] = [] + seen_ids: set[str] = set() + function_bodies: list[tuple[str, object]] = [] + + def add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": label, + "file_type": "code", + "source_file": str_path, + "source_location": f"L{line}", + }) + + def add_edge(src: str, tgt: str, relation: str, line: int, + confidence: str = "EXTRACTED", weight: float = 1.0, + context: str | None = None) -> None: + edge = { + "source": src, + "target": tgt, + "relation": relation, + "confidence": confidence, + "source_file": str_path, + "source_location": f"L{line}", + "weight": weight, + } + if context: + edge["context"] = context + edges.append(edge) + + file_nid = _make_id(str(path)) + add_node(file_nid, path.name, 1) + + def ensure_named_node(name: str, line: int) -> str: + nid = _make_id(stem, name) + if nid in seen_ids: + return nid + nid = _make_id(name) + if nid not in seen_ids: + # The name isn't defined in this file, so this is a cross-file reference + # (e.g. a `Thing` type annotation imported from another module). Emit a + # SOURCELESS stub — like the inheritance-base path below — so the + # corpus-level rewire can collapse it onto the real definition. A sourced + # stub here makes _disambiguate_colliding_node_ids bake the referencing + # file's path (with extension) into the id and blocks the rewire, which is + # the phantom-duplicate-node bug (#1402). + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": name, + "file_type": "code", + "source_file": "", + "source_location": "", + "origin_file": str_path, + }) + return nid + + def emit_param_return_refs(func_node, func_nid: str, line: int) -> None: + params = func_node.child_by_field_name("parameters") + if params is not None: + for p in params.children: + if p.type != "parameter": + continue + type_node = p.child_by_field_name("type") + refs: list[tuple[str, str]] = [] + _rust_collect_type_refs(type_node, source, False, refs) + for ref_name, role in refs: + ctx = "generic_arg" if role == "generic_arg" else "parameter_type" + tgt = ensure_named_node(ref_name, line) + if tgt != func_nid: + add_edge(func_nid, tgt, "references", line, context=ctx) + return_type = func_node.child_by_field_name("return_type") + if return_type is not None: + refs = [] + _rust_collect_type_refs(return_type, source, False, refs) + for ref_name, role in refs: + ctx = "generic_arg" if role == "generic_arg" else "return_type" + tgt = ensure_named_node(ref_name, line) + if tgt != func_nid: + add_edge(func_nid, tgt, "references", line, context=ctx) + + def walk(node, parent_impl_nid: str | None = None) -> None: + t = node.type + + if t == "function_item": + name_node = node.child_by_field_name("name") + if name_node: + func_name = _read_text(name_node, source) + line = node.start_point[0] + 1 + if parent_impl_nid: + func_nid = _make_id(parent_impl_nid, func_name) + add_node(func_nid, f".{func_name}()", line) + add_edge(parent_impl_nid, func_nid, "method", line) + else: + func_nid = _make_id(stem, func_name) + add_node(func_nid, f"{func_name}()", line) + add_edge(file_nid, func_nid, "contains", line) + emit_param_return_refs(node, func_nid, line) + body = node.child_by_field_name("body") + if body: + function_bodies.append((func_nid, body)) + return + + if t in ("struct_item", "enum_item", "trait_item"): + name_node = node.child_by_field_name("name") + if name_node: + item_name = _read_text(name_node, source) + line = node.start_point[0] + 1 + item_nid = _make_id(stem, item_name) + add_node(item_nid, item_name, line) + add_edge(file_nid, item_nid, "contains", line) + if t == "trait_item": + for c in node.children: + if c.type != "trait_bounds": + continue + for sub in c.children: + if not sub.is_named: + continue + refs: list[tuple[str, str]] = [] + _rust_collect_type_refs(sub, source, False, refs) + for idx, (ref_name, _role) in enumerate(refs): + tgt = ensure_named_node(ref_name, line) + if tgt == item_nid: + continue + rel = "inherits" if idx == 0 else "references" + if rel == "inherits": + add_edge(item_nid, tgt, "inherits", line) + else: + add_edge(item_nid, tgt, "references", line, + context="generic_arg") + if t == "struct_item": + for c in node.children: + if c.type != "field_declaration_list": + continue + for field in c.children: + if field.type != "field_declaration": + continue + type_node = field.child_by_field_name("type") + if type_node is None: + for fc in field.children: + if fc.type in ("type_identifier", "generic_type", + "scoped_type_identifier", + "reference_type", "primitive_type"): + type_node = fc + break + refs = [] + _rust_collect_type_refs(type_node, source, False, refs) + for ref_name, role in refs: + ctx = "generic_arg" if role == "generic_arg" else "field" + tgt = ensure_named_node(ref_name, field.start_point[0] + 1) + if tgt != item_nid: + add_edge(item_nid, tgt, "references", + field.start_point[0] + 1, context=ctx) + # Tuple structs (`struct Wrapper(pub Logger, Config);`) nest their + # positional field types directly under ordered_field_declaration_list + # with no field_declaration wrapper -- the same shape handled for tuple + # enum variants below. Without this branch these field type references + # are silently dropped. + for c in node.children: + if c.type != "ordered_field_declaration_list": + continue + fline = c.start_point[0] + 1 + for tc in c.children: + if tc.type not in ("type_identifier", "generic_type", + "scoped_type_identifier", "reference_type", + "primitive_type", "tuple_type", "array_type"): + continue + refs = [] + _rust_collect_type_refs(tc, source, False, refs) + for ref_name, role in refs: + ctx = "generic_arg" if role == "generic_arg" else "field" + tgt = ensure_named_node(ref_name, fline) + if tgt != item_nid: + add_edge(item_nid, tgt, "references", fline, context=ctx) + if t == "enum_item": + # Variant payload types nest under enum_variant_list -> + # enum_variant -> ordered_field_declaration_list (tuple variant, + # `Click(Logger)`) | field_declaration_list (struct variant, + # `Resize { size: Dim }`). Neither was traversed, so every + # enum-variant type reference was silently dropped. + _TYPE_NODES = ("type_identifier", "generic_type", + "scoped_type_identifier", "reference_type", + "primitive_type", "tuple_type", "array_type") + + def _emit_enum_type(type_node, at_line): + if type_node is None: + return + refs2: list[tuple[str, str]] = [] + _rust_collect_type_refs(type_node, source, False, refs2) + for ref_name, role in refs2: + ctx = "generic_arg" if role == "generic_arg" else "field" + tgt = ensure_named_node(ref_name, at_line) + if tgt != item_nid: + add_edge(item_nid, tgt, "references", at_line, context=ctx) + + for c in node.children: + if c.type != "enum_variant_list": + continue + for variant in c.children: + if variant.type != "enum_variant": + continue + vline = variant.start_point[0] + 1 + for vc in variant.children: + if vc.type == "ordered_field_declaration_list": + for tc in vc.children: + if tc.type in _TYPE_NODES: + _emit_enum_type(tc, vline) + elif vc.type == "field_declaration_list": + for field in vc.children: + if field.type != "field_declaration": + continue + type_node = field.child_by_field_name("type") + _emit_enum_type(type_node, field.start_point[0] + 1) + return + + if t == "impl_item": + type_node = node.child_by_field_name("type") + trait_node = node.child_by_field_name("trait") + impl_nid: str | None = None + if type_node: + type_name = _read_text(type_node, source).strip() + impl_nid = _make_id(stem, type_name) + add_node(impl_nid, type_name, node.start_point[0] + 1) + if trait_node is not None and impl_nid is not None: + refs: list[tuple[str, str]] = [] + _rust_collect_type_refs(trait_node, source, False, refs) + for idx, (ref_name, _role) in enumerate(refs): + tgt = ensure_named_node(ref_name, node.start_point[0] + 1) + if tgt == impl_nid: + continue + if idx == 0: + add_edge(impl_nid, tgt, "implements", node.start_point[0] + 1) + else: + add_edge(impl_nid, tgt, "references", node.start_point[0] + 1, + context="generic_arg") + body = node.child_by_field_name("body") + if body: + for child in body.children: + walk(child, parent_impl_nid=impl_nid) + return + + if t == "use_declaration": + arg = node.child_by_field_name("argument") + if arg: + raw = _read_text(arg, source) + clean = raw.split("{")[0].rstrip(":").rstrip("*").rstrip(":") + module_name = clean.split("::")[-1].strip() + if module_name: + tgt_nid = _make_id(module_name) + add_edge(file_nid, tgt_nid, "imports_from", node.start_point[0] + 1, context="import") + return + + for child in node.children: + walk(child, parent_impl_nid=None) + + walk(root) + + label_to_nid: dict[str, str] = {} + for n in nodes: + raw = n["label"] + normalised = raw.strip("()").lstrip(".") + label_to_nid[normalised] = n["id"] + + seen_call_pairs: set[tuple[str, str]] = set() + raw_calls: list[dict] = [] + + def walk_calls(node, caller_nid: str) -> None: + if node.type == "function_item": + return + if node.type == "call_expression": + func_node = node.child_by_field_name("function") + callee_name: str | None = None + is_member_call: bool = False + is_scoped_call: bool = False + if func_node: + if func_node.type == "identifier": + callee_name = _read_text(func_node, source) + elif func_node.type == "field_expression": + is_member_call = True + field = func_node.child_by_field_name("field") + if field: + callee_name = _read_text(field, source) + elif func_node.type == "scoped_identifier": + # Type::method() — still allow in-file EXTRACTED match, but + # skip cross-file resolution: bare last-segment lookup ignores + # crate boundaries and produces spurious INFERRED edges (#908). + is_scoped_call = True + name = func_node.child_by_field_name("name") + if name: + callee_name = _read_text(name, source) + if callee_name and callee_name not in _LANGUAGE_BUILTIN_GLOBALS: + tgt_nid = label_to_nid.get(callee_name) + if tgt_nid and tgt_nid != caller_nid: + pair = (caller_nid, tgt_nid) + if pair not in seen_call_pairs: + seen_call_pairs.add(pair) + line = node.start_point[0] + 1 + edges.append({ + "source": caller_nid, + "target": tgt_nid, + "relation": "calls", + "context": "call", + "confidence": "EXTRACTED", + "source_file": str_path, + "source_location": f"L{line}", + "weight": 1.0, + }) + elif not is_scoped_call and callee_name.lower() not in _RUST_TRAIT_METHOD_BLOCKLIST: + raw_calls.append({ + "caller_nid": caller_nid, + "callee": callee_name, + "is_member_call": is_member_call, + "source_file": str_path, + "source_location": f"L{node.start_point[0] + 1}", + }) + for child in node.children: + walk_calls(child, caller_nid) + + for caller_nid, body_node in function_bodies: + walk_calls(body_node, caller_nid) + + valid_ids = seen_ids + clean_edges = [] + for edge in edges: + src, tgt = edge["source"], edge["target"] + if src in valid_ids and (tgt in valid_ids or edge["relation"] in ("imports", "imports_from")): + clean_edges.append(edge) + + return {"nodes": nodes, "edges": clean_edges, "raw_calls": raw_calls} + + +# ── Zig ─────────────────────────────────────────────────────────────────────── + + + +# ── PowerShell ──────────────────────────────────────────────────────────────── + +def extract_powershell(path: Path) -> dict: + """Extract functions, classes, methods, and using statements from a .ps1 file.""" + try: + import tree_sitter_powershell as tsps + from tree_sitter import Language, Parser + except ImportError: + return {"nodes": [], "edges": [], "error": "tree_sitter_powershell not installed"} + + try: + language = Language(tsps.language()) + parser = Parser(language) + source = path.read_bytes() + tree = parser.parse(source) + root = tree.root_node + except Exception as e: + return {"nodes": [], "edges": [], "error": str(e)} + + stem = _file_stem(path) + str_path = str(path) + nodes: list[dict] = [] + edges: list[dict] = [] + seen_ids: set[str] = set() + function_bodies: list[tuple[str, Any]] = [] + + def add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({"id": nid, "label": label, "file_type": "code", + "source_file": str_path, "source_location": f"L{line}"}) + + def add_edge(src: str, tgt: str, relation: str, line: int, + confidence: str = "EXTRACTED", weight: float = 1.0, + context: str | None = None) -> None: + edge = {"source": src, "target": tgt, "relation": relation, + "confidence": confidence, "source_file": str_path, + "source_location": f"L{line}", "weight": weight} + if context: + edge["context"] = context + edges.append(edge) + + file_nid = _make_id(str(path)) + add_node(file_nid, path.name, 1) + + _PS_SKIP = frozenset({ + "using", "return", "if", "else", "elseif", "foreach", "for", + "while", "do", "switch", "try", "catch", "finally", "throw", + "break", "continue", "exit", "param", "begin", "process", "end", + # Import commands — handled as import edges, not function calls + "import-module", + }) + + def _find_script_block_body(node): + for child in node.children: + if child.type == "script_block": + for sc in child.children: + if sc.type == "script_block_body": + return sc + return child + return None + + def ensure_named_node(name: str, line: int) -> str: + nid = _make_id(stem, name) + if nid in seen_ids: + return nid + nid = _make_id(name) + if nid not in seen_ids: + # The name isn't defined in this file, so this is a cross-file reference + # (e.g. a `Thing` type annotation imported from another module). Emit a + # SOURCELESS stub — like the inheritance-base path below — so the + # corpus-level rewire can collapse it onto the real definition. A sourced + # stub here makes _disambiguate_colliding_node_ids bake the referencing + # file's path (with extension) into the id and blocks the rewire, which is + # the phantom-duplicate-node bug (#1402). + seen_ids.add(nid) + nodes.append({ + "id": nid, + "label": name, + "file_type": "code", + "source_file": "", + "source_location": "", + "origin_file": str_path, + }) + return nid + + def _ps_type_name(type_literal_node) -> str | None: + """Drill into a type_literal node and return the inner type_identifier text.""" + if type_literal_node is None: + return None + for spec in type_literal_node.children: + if spec.type != "type_spec": + continue + for tname in spec.children: + if tname.type != "type_name": + continue + for tid in tname.children: + if tid.type == "type_identifier": + return _read_text(tid, source) + return None + + def walk(node, parent_class_nid: str | None = None) -> None: + t = node.type + + if t == "function_statement": + name_node = next((c for c in node.children if c.type == "function_name"), None) + if name_node: + func_name = _read_text(name_node, source) + line = node.start_point[0] + 1 + func_nid = _make_id(stem, func_name) + add_node(func_nid, f"{func_name}()", line) + add_edge(file_nid, func_nid, "contains", line) + body = _find_script_block_body(node) + if body: + function_bodies.append((func_nid, body)) + # Also walk the body during the main pass so that + # Import-Module / dot-source inside functions emit + # file-level imports_from edges (#1331). + walk(body, parent_class_nid) + return + + if t == "class_statement": + name_node = next((c for c in node.children if c.type == "simple_name"), None) + if name_node: + class_name = _read_text(name_node, source) + line = node.start_point[0] + 1 + class_nid = _make_id(stem, class_name) + add_node(class_nid, class_name, line) + add_edge(file_nid, class_nid, "contains", line) + # Base type(s) after ':'. PowerShell has no syntactic base vs + # interface split, so (matching the C# convention) treat the + # first base as the superclass (inherits) and the rest as + # interfaces (implements). Bases are the simple_name children + # after the ':' token. + colon_seen = False + base_index = 0 + for child in node.children: + if child.type == ":": + colon_seen = True + elif colon_seen and child.type == "simple_name": + base_nid = ensure_named_node(_read_text(child, source), line) + if base_nid != class_nid: + rel = "inherits" if base_index == 0 else "implements" + add_edge(class_nid, base_nid, rel, line) + base_index += 1 + for child in node.children: + walk(child, parent_class_nid=class_nid) + return + + if t == "class_property_definition" and parent_class_nid: + type_literal = next((c for c in node.children if c.type == "type_literal"), None) + type_name = _ps_type_name(type_literal) + if type_name: + line = node.start_point[0] + 1 + target_nid = ensure_named_node(type_name, line) + if target_nid != parent_class_nid: + add_edge(parent_class_nid, target_nid, "references", + line, context="field") + return + + if t == "class_method_definition": + name_node = next((c for c in node.children if c.type == "simple_name"), None) + if name_node: + method_name = _read_text(name_node, source) + line = node.start_point[0] + 1 + if parent_class_nid: + method_nid = _make_id(parent_class_nid, method_name) + add_node(method_nid, f".{method_name}()", line) + add_edge(parent_class_nid, method_nid, "method", line) + else: + method_nid = _make_id(stem, method_name) + add_node(method_nid, f"{method_name}()", line) + add_edge(file_nid, method_nid, "contains", line) + # Return type: type_literal sibling of simple_name + return_type_literal = next( + (c for c in node.children if c.type == "type_literal"), None) + return_type_name = _ps_type_name(return_type_literal) + if return_type_name: + target_nid = ensure_named_node(return_type_name, line) + if target_nid != method_nid: + add_edge(method_nid, target_nid, "references", + line, context="return_type") + # Parameter types: class_method_parameter_list + param_list = next( + (c for c in node.children if c.type == "class_method_parameter_list"), None) + if param_list is not None: + for p in param_list.children: + if p.type != "class_method_parameter": + continue + ptype_literal = next( + (c for c in p.children if c.type == "type_literal"), None) + ptype_name = _ps_type_name(ptype_literal) + if not ptype_name: + continue + p_line = p.start_point[0] + 1 + target_nid = ensure_named_node(ptype_name, p_line) + if target_nid != method_nid: + add_edge(method_nid, target_nid, "references", + p_line, context="parameter_type") + body = _find_script_block_body(node) + if body: + function_bodies.append((method_nid, body)) + return + + if t == "command": + # Dot-sourcing: `. ./Shared.psm1` + # Uses command_invokation_operator '.' + command_name_expr (not command_name) + invoke_op = next( + (c for c in node.children if c.type == "command_invokation_operator"), None + ) + if invoke_op is not None and _read_text(invoke_op, source).strip() == ".": + name_expr = next( + (c for c in node.children if c.type == "command_name_expr"), None + ) + if name_expr is not None: + name_node = next( + (c for c in name_expr.children if c.type == "command_name"), None + ) + if name_node: + raw_path = _read_text(name_node, source) + # Strip relative path prefix (./ or .\ or just the dot) + module_stem = re.sub(r'^[./\\]+', '', raw_path) + # Drop extension to get bare module name + module_stem = re.sub(r'\.[^.]+$', '', module_stem).replace('\\', '/') + module_name = module_stem.split('/')[-1] + if module_name: + add_edge(file_nid, _make_id(module_name), "imports_from", + node.start_point[0] + 1) + return + + cmd_name_node = next((c for c in node.children if c.type == "command_name"), None) + if cmd_name_node: + cmd_text = _read_text(cmd_name_node, source).lower() + if cmd_text == "using": + tokens = [] + for child in node.children: + if child.type == "command_elements": + for el in child.children: + if el.type == "generic_token": + tokens.append(_read_text(el, source)) + module_tokens = [t for t in tokens + if t.lower() not in ("namespace", "module", "assembly")] + if module_tokens: + module_name = module_tokens[-1].split(".")[-1] + add_edge(file_nid, _make_id(module_name), "imports_from", + node.start_point[0] + 1) + elif cmd_text == "import-module": + # Collect generic_token args; skip command_parameter flags like -Name + # The module name is the first generic_token (or the one after -Name) + module_name: str | None = None + expect_name = False + for child in node.children: + if child.type != "command_elements": + continue + for el in child.children: + if el.type == "command_parameter": + param_text = _read_text(el, source).lstrip("-").lower() + expect_name = param_text in ("name", "n") + elif el.type == "generic_token": + token = _read_text(el, source) + if module_name is None or expect_name: + module_name = token + expect_name = False + if module_name: + # Strip extension; keep only the stem for the node ID + bare = re.sub(r'\.[^.]+$', '', module_name).split('/')[-1].split('\\')[-1] + if bare: + add_edge(file_nid, _make_id(bare), "imports_from", + node.start_point[0] + 1) + return + + for child in node.children: + walk(child, parent_class_nid) + + walk(root) + + label_to_nid = {n["label"].strip("()").lstrip(".").lower(): n["id"] for n in nodes} + seen_call_pairs: set[tuple[str, str]] = set() + raw_calls: list[dict] = [] + + def walk_calls(node, caller_nid: str) -> None: + if node.type in ("function_statement", "class_statement"): + return + if node.type == "command": + cmd_name_node = next((c for c in node.children if c.type == "command_name"), None) + if cmd_name_node: + cmd_text = _read_text(cmd_name_node, source) + if cmd_text.lower() not in _PS_SKIP: + tgt_nid = label_to_nid.get(cmd_text.lower()) + if tgt_nid and tgt_nid != caller_nid: + pair = (caller_nid, tgt_nid) + if pair not in seen_call_pairs: + seen_call_pairs.add(pair) + add_edge(caller_nid, tgt_nid, "calls", + node.start_point[0] + 1, + confidence="EXTRACTED", weight=1.0) + elif cmd_text: + raw_calls.append({ + "caller_nid": caller_nid, + "callee": cmd_text, + "is_member_call": False, + "source_file": str_path, + "source_location": f"L{node.start_point[0] + 1}", + }) + for child in node.children: + walk_calls(child, caller_nid) + + for caller_nid, body_node in function_bodies: + walk_calls(body_node, caller_nid) + + clean_edges = [e for e in edges if e["source"] in seen_ids and + (e["target"] in seen_ids or e["relation"] in ("imports_from", "imports"))] + return {"nodes": nodes, "edges": clean_edges, "raw_calls": raw_calls} + + +# ── PowerShell manifest (.psd1) ────────────────────────────────────────────── + +# Keys in a .psd1 whose values are module names/paths we treat as imports. +_PSD1_IMPORT_KEYS = frozenset({"RootModule", "NestedModules", "RequiredModules"}) + + +def _psd1_collect_string_literals(node, source: bytes) -> list[str]: + """Recursively collect all string_literal text values under *node*.""" + results: list[str] = [] + + def _walk(n) -> None: + if n.type == "string_literal": + raw = source[n.start_byte:n.end_byte].decode(errors="replace") + # Strip surrounding quote chars (' or ") + results.append(raw.strip("'\"")) + return + for child in n.children: + _walk(child) + + _walk(node) + return results + + +def _psd1_module_name(raw: str) -> str: + """Derive a bare module name from a raw string value. + + e.g. 'MyModule.psm1' → 'MyModule', './sub/Util.psm1' → 'Util', 'PSReadLine' → 'PSReadLine' + """ + # Strip path prefix and extension + name = raw.replace("\\", "/").split("/")[-1] + name = re.sub(r"\.[^.]+$", "", name) # remove last extension + return name.strip() + + +def extract_powershell_manifest(path: Path) -> dict: + """Extract module dependency edges from a PowerShell .psd1 manifest file. + + .psd1 files are PowerShell data hashtables, not scripts. tree-sitter-powershell + parses them correctly (they are syntactically valid PS). We walk the AST looking + for RootModule, NestedModules, and RequiredModules keys and emit imports_from + edges for every referenced module. + + RequiredModules supports two forms: + - Simple string: 'PSReadLine' + - Module specification: @{ ModuleName = 'Pester'; ModuleVersion = '5.0' } + For the hashtable form we only follow the ModuleName key. + """ + try: + import tree_sitter_powershell as tsps + from tree_sitter import Language, Parser + except ImportError: + return {"nodes": [], "edges": [], "error": "tree_sitter_powershell not installed"} + + try: + language = Language(tsps.language()) + parser = Parser(language) + source = path.read_bytes() + tree = parser.parse(source) + root = tree.root_node + except Exception as e: + return {"nodes": [], "edges": [], "error": str(e)} + + str_path = str(path) + nodes: list[dict] = [] + edges: list[dict] = [] + seen_ids: set[str] = set() + + def add_node(nid: str, label: str, line: int) -> None: + if nid not in seen_ids: + seen_ids.add(nid) + nodes.append({"id": nid, "label": label, "file_type": "code", + "source_file": str_path, "source_location": f"L{line}"}) + + def add_import_edge(src: str, module_raw: str, line: int) -> None: + name = _psd1_module_name(module_raw) + if not name: + return + tgt_nid = _make_id(name) + edges.append({ + "source": src, + "target": tgt_nid, + "relation": "imports_from", + "confidence": "EXTRACTED", + "source_file": str_path, + "source_location": f"L{line}", + "weight": 1.0, + "context": "import", + }) + + file_nid = _make_id(str(path)) + add_node(file_nid, path.name, 1) + + def walk_manifest(node) -> None: + """Walk the AST and emit edges for import-relevant hash_entry nodes.""" + if node.type != "hash_entry": + for child in node.children: + walk_manifest(child) + return + + # Identify the key + key_node = next((c for c in node.children if c.type == "key_expression"), None) + if key_node is None: + return + key_text = source[key_node.start_byte:key_node.end_byte].decode(errors="replace").strip() + + if key_text not in _PSD1_IMPORT_KEYS: + # Still recurse in case there are nested hashes (e.g. ModuleVersion entries + # contain sub-hashes, but we only care about top-level keys for imports) + return + + line = node.start_point[0] + 1 + value_node = next((c for c in node.children if c.type == "pipeline"), None) + if value_node is None: + return + + if key_text == "RootModule": + # Value is a single string + strings = _psd1_collect_string_literals(value_node, source) + for s in strings: + add_import_edge(file_nid, s, line) + + elif key_text == "NestedModules": + # Value is a string or @('a', 'b', ...) array — collect all string literals + strings = _psd1_collect_string_literals(value_node, source) + for s in strings: + add_import_edge(file_nid, s, line) + + elif key_text == "RequiredModules": + # Two forms: + # 1) 'SimpleModule' — direct string literals in the array + # 2) @{ ModuleName = 'Foo'; ModuleVersion = '2.0' } — use ModuleName only + # + # Strategy: walk the value for hash_entry nodes whose key is 'ModuleName'; + # collect their string values. For the remaining string_literal nodes that + # are NOT inside a hash_entry subtree, treat them as simple module names. + module_name_strings: list[str] = [] + inside_hash_entries: set[int] = set() # byte offsets of handled strings + + def find_modulename_entries(n) -> None: + if n.type == "hash_entry": + sub_key = next((c for c in n.children if c.type == "key_expression"), None) + if sub_key is not None: + sk_text = source[sub_key.start_byte:sub_key.end_byte].decode(errors="replace").strip() + # Collect strings inside *all* sub-keys so we can exclude them + for c in n.children: + if c.type == "pipeline": + for s_node in _collect_string_nodes(c): + inside_hash_entries.add(s_node.start_byte) + if sk_text == "ModuleName": + for c in n.children: + if c.type == "pipeline": + for s in _psd1_collect_string_literals(c, source): + module_name_strings.append(s) + return # don't recurse further into this hash_entry + for child in n.children: + find_modulename_entries(child) + + def _collect_string_nodes(n): + """Return all string_literal nodes in subtree.""" + if n.type == "string_literal": + yield n + return + for child in n.children: + yield from _collect_string_nodes(child) + + find_modulename_entries(value_node) + + # Now gather direct string literals not inside hash entries + direct_strings: list[str] = [] + for s_node in _collect_string_nodes(value_node): + if s_node.start_byte not in inside_hash_entries: + raw = source[s_node.start_byte:s_node.end_byte].decode(errors="replace") + direct_strings.append(raw.strip("'\"")) + + for s in direct_strings + module_name_strings: + add_import_edge(file_nid, s, line) + + walk_manifest(root) + + return {"nodes": nodes, "edges": edges, "raw_calls": []} + + +# ── Cross-file import resolution ────────────────────────────────────────────── + +def _source_key(source_file: str, root: Path) -> str: + if not source_file: + return "" + source_path = Path(source_file) + try: + return str(source_path.resolve().relative_to(root)) + except Exception: + return str(source_path) + + +def _node_disambiguation_source_key(node: dict, root: Path) -> str: + source_file = str(node.get("source_file", "")) + if source_file: + return _source_key(source_file, root) + return _source_key(str(node.get("origin_file", "")), root) + + +def _disambiguate_colliding_node_ids( + nodes: list[dict], + edges: list[dict], + raw_calls: list[dict], + root: Path, +) -> None: + """Rewrite only colliding node IDs, using source path as the disambiguator. + + Module anchor nodes (#1327) are exempt: ``import CoreKit`` from three files + yields three ``type=module`` nodes with the same id but different + source_files. Those are the *same* module, not distinct same-named symbols, + so they must collapse to one shared node — disambiguating them by path would + scatter a single module across N file-qualified duplicates. + """ + by_id: dict[str, list[dict]] = {} + for node in nodes: + if node.get("type") in ("module", "namespace"): + continue + nid = node.get("id") + if isinstance(nid, str) and nid: + by_id.setdefault(nid, []).append(node) + + remap: dict[tuple[str, str], str] = {} + ambiguous_ids: set[str] = set() + for old_id, group in by_id.items(): + source_keys = {_node_disambiguation_source_key(node, root) for node in group} + if len(group) < 2 or len(source_keys) < 2: + continue + ambiguous_ids.add(old_id) + # Salt the colliding id with the *path* it came from. The naive salt is + # ``_make_id(source_key, old_id)`` — source_key is the raw repo-relative + # path. But _make_id collapses every separator, so two DISTINCT paths + # whose only difference is a separator-vs-inner-punctuation swap + # (``a/b/c.md`` vs ``a.b/c.md``, ``foo/bar_baz.md`` vs ``foo_bar/baz.md``) + # normalize to the SAME salted id and still collide (#1522 — the residual + # of #1504 the 0.9.0 full-path stem didn't reach). When that happens, + # append a short stable hash of the *raw* source_key, which IS injective + # over distinct paths, so the colliders separate. Computed in code from + # source_file (never trusted from the LLM), so AST↔semantic parity holds. + naive: dict[str, str] = {} # source_key -> _make_id(source_key, old_id) + for source_key in source_keys: + if source_key: + naive[source_key] = _make_id(source_key, old_id) + # source_keys that, after normalization, are not unique among themselves. + seen: dict[str, int] = {} + for nid in naive.values(): + seen[nid] = seen.get(nid, 0) + 1 + needs_hash = {sk for sk, nid in naive.items() if seen.get(nid, 0) > 1} + for node in group: + source_key = _node_disambiguation_source_key(node, root) + if not source_key: + continue + if source_key in needs_hash: + salt = hashlib.sha1(source_key.encode("utf-8")).hexdigest()[:6] + new_id = _make_id(source_key, old_id, salt) + else: + new_id = naive.get(source_key) or _make_id(source_key, old_id) + remap[(old_id, source_key)] = new_id + if new_id != old_id: + node["id"] = new_id + + if not remap: + return + + unambiguous_remaps: dict[str, str] = {} + for old_id, group in by_id.items(): + if old_id in ambiguous_ids: + continue + candidates = { + node["id"] for node in group + if isinstance(node.get("id"), str) and node["id"] != old_id + } + if len(candidates) == 1: + unambiguous_remaps[old_id] = next(iter(candidates)) + + # A C/ObjC/C++ `#include "foo.h"` / `#import "foo.h"` resolves to the header's + # file node, but `foo.h` and its sibling `foo.c`/`foo.m`/`foo.cpp` collapse to + # the same `foo` file id, so disambiguation salts them apart by path. A + # cross-file import edge from a THIRD file carries neither salt's source_key, so + # the (target, edge_source_key) lookup misses and the edge dangles on the now + # dead `foo` id. Repoint those import edges to the HEADER variant (the include + # always targeted the header), keyed by the original colliding id (#1475). + _HEADER_SUFFIXES = (".h", ".hpp", ".hh", ".hxx") + header_remaps: dict[str, str] = {} + for old_id in ambiguous_ids: + for node in by_id.get(old_id, []): + sk = _node_disambiguation_source_key(node, root) + if sk and Path(sk).suffix.lower() in _HEADER_SUFFIXES: + new_id = remap.get((old_id, sk)) + if new_id: + header_remaps[old_id] = new_id + break + + for edge in edges: + edge_source_key = _source_key(str(edge.get("source_file", "")), root) + source_key = (edge.get("source", ""), edge_source_key) + target_key = (edge.get("target", ""), edge_source_key) + if source_key in remap: + edge["source"] = remap[source_key] + elif edge.get("source") in unambiguous_remaps: + edge["source"] = unambiguous_remaps[str(edge["source"])] + # imports/imports_from always target a header file, so they must resolve to + # the header variant BEFORE the same-source-file salt is considered. Keying + # the import target by the importer's own source file mis-points a `.m` + # importing its own `.h` back at itself (self-loop), and is wrong for any + # cross-file import whose importer shares the colliding id (#1475). + if (edge.get("relation") in ("imports", "imports_from") + and edge.get("target") in header_remaps): + edge["target"] = header_remaps[str(edge["target"])] + elif target_key in remap: + edge["target"] = remap[target_key] + elif edge.get("target") in unambiguous_remaps: + edge["target"] = unambiguous_remaps[str(edge["target"])] + + for raw_call in raw_calls: + call_source_key = _source_key(str(raw_call.get("source_file", "")), root) + caller_key = (raw_call.get("caller_nid", ""), call_source_key) + if caller_key in remap: + raw_call["caller_nid"] = remap[caller_key] + elif raw_call.get("caller_nid") in unambiguous_remaps: + raw_call["caller_nid"] = unambiguous_remaps[str(raw_call["caller_nid"])] + + +def _canonicalize_csharp_namespace_nodes(all_nodes: list[dict], all_edges: list[dict]) -> None: + """Collapse duplicate C# namespace node entries to one canonical node per label.""" + by_label: dict[str, list[dict]] = {} + for node in all_nodes: + if node.get("type") != "namespace": + continue + label = node.get("label") + if isinstance(label, str): + by_label.setdefault(label, []).append(node) + + remap: dict[str, str] = {} + drop_node_ids: set[int] = set() + for group in by_label.values(): + if len(group) < 2: + continue + canonical = sorted( + group, + key=lambda node: ( + str(node.get("source_file") or ""), + str(node.get("source_location") or ""), + str(node.get("id") or ""), + ), + )[0] + canonical_id = canonical.get("id") + for node in group: + if node is canonical: + continue + drop_node_ids.add(id(node)) + dup_id = node.get("id") + if isinstance(dup_id, str) and isinstance(canonical_id, str): + remap[dup_id] = canonical_id + + if remap: + for edge in all_edges: + if edge.get("source") in remap: + edge["source"] = remap[str(edge["source"])] + if edge.get("target") in remap: + edge["target"] = remap[str(edge["target"])] + + if drop_node_ids: + all_nodes[:] = [node for node in all_nodes if id(node) not in drop_node_ids] + + +# Languages whose identifiers are case-insensitive, so cross-file name resolution +# may fold case. Everywhere else, case is semantic (`Path` the class vs `PATH` the +# env var are distinct) and folding manufactures false edges / super-hubs (#1581). +_CASE_INSENSITIVE_EXTS = frozenset({ + ".php", ".phtml", ".php3", ".php4", ".php5", ".php7", ".phps", # PHP fns/classes + ".sql", # SQL identifiers + ".nim", ".nims", ".nimble", # Nim (style-insensitive) +}) + + +def _lang_is_case_insensitive(source_file: object) -> bool: + """True when the file's language resolves identifiers case-insensitively (#1581).""" + if not source_file: + return False + return Path(str(source_file)).suffix.lower() in _CASE_INSENSITIVE_EXTS + + +def _node_label_key(node: dict, fold: bool = False) -> str: + label = str(node.get("label", "")).strip() + key = re.sub(r"[^a-zA-Z0-9]+", "", label) + return key.lower() if fold else key + + +def _is_type_like_definition(node: dict) -> bool: + if node.get("type") == "namespace": + return False + label = str(node.get("label", "")).strip() + if not label: + return False + if label.endswith(")") or label.startswith("."): + return False + if "." in label: + return False + return node.get("file_type") == "code" + + +def _rewire_unique_stub_nodes(nodes: list[dict], edges: list[dict]) -> None: + """Map unresolved no-source stubs to a unique real definition with the same label.""" + real_by_label: dict[str, list[dict]] = {} # exact-case (all languages) + real_by_label_ci: dict[str, list[dict]] = {} # case-INSENSITIVE-language reals only + stubs: list[dict] = [] + + for node in nodes: key = _node_label_key(node) if not key: continue - if node.get("source_file"): - if _is_type_like_definition(node): - # Match stubs case-SENSITIVELY: a `Path` reference must not rewire to a - # `PATH` env var (#1581). Fold only for genuinely case-insensitive - # languages, where `foo` legitimately resolves to `Foo`. - real_by_label.setdefault(key, []).append(node) - if _lang_is_case_insensitive(node.get("source_file")): - real_by_label_ci.setdefault( - _node_label_key(node, fold=True), []).append(node) - elif _is_top_level_function_definition(node): - func_by_label.setdefault(key, []).append(node) + if node.get("source_file"): + if _is_type_like_definition(node): + # Match stubs case-SENSITIVELY: a `Path` reference must not rewire to a + # `PATH` env var (#1581). Fold only for genuinely case-insensitive + # languages, where `foo` legitimately resolves to `Foo`. + real_by_label.setdefault(key, []).append(node) + if _lang_is_case_insensitive(node.get("source_file")): + real_by_label_ci.setdefault( + _node_label_key(node, fold=True), []).append(node) + continue + stubs.append(node) + + remap: dict[str, str] = {} + for stub in stubs: + stub_id = str(stub.get("id", "")) + if not stub_id: + continue + candidates = real_by_label.get(_node_label_key(stub), []) + if len(candidates) != 1: + # No unique exact match — fall back to a case-insensitive match, but + # only against case-insensitive-language definitions (so a case-sensitive + # `PATH` can never absorb a `Path` reference). + candidates = real_by_label_ci.get(_node_label_key(stub, fold=True), []) + if len(candidates) != 1: + continue + target_id = candidates[0].get("id") + if isinstance(target_id, str) and target_id and target_id != stub_id: + remap[stub_id] = target_id + + if not remap: + return + + by_id = {node.get("id"): node for node in nodes if node.get("id")} + csharp_scoped_relations = {"inherits", "implements", "references", "imports"} + for edge in edges: + is_csharp_scoped_edge = ( + str(edge.get("source_file", "")).endswith(".cs") + and edge.get("relation") in csharp_scoped_relations + ) + source = edge.get("source") + if source in remap: + remapped_source = remap[str(source)] + if not ( + is_csharp_scoped_edge + and str(by_id.get(remapped_source, {}).get("source_file", "")).endswith(".cs") + ): + edge["source"] = remapped_source + target = edge.get("target") + if target in remap: + remapped_target = remap[str(target)] + if not ( + is_csharp_scoped_edge + and str(by_id.get(remapped_target, {}).get("source_file", "")).endswith(".cs") + ): + edge["target"] = remapped_target + + referenced = {x for e in edges for x in (e.get("source"), e.get("target"))} + drop_ids = {stub_id for stub_id in remap if stub_id not in referenced} + nodes[:] = [node for node in nodes if node.get("id") not in drop_ids] + + +def _js_source_path(source_file: str, root: Path) -> Path | None: + if not source_file: + return None + path = Path(source_file) + if not path.is_absolute(): + path = root / path + try: + return path.resolve() + except Exception: + return path + + +@dataclass(frozen=True) +class _SymbolDeclarationFact: + file_path: Path + name: str + line: int + + +@dataclass(frozen=True) +class _SymbolImportFact: + file_path: Path + local_name: str + target_path: Path + imported_name: str + line: int + + +@dataclass(frozen=True) +class _SymbolAliasFact: + file_path: Path + alias: str + target_name: str + line: int + + +@dataclass(frozen=True) +class _SymbolExportFact: + file_path: Path + exported_name: str + line: int + local_name: str | None = None + target_path: Path | None = None + target_name: str | None = None + + +@dataclass(frozen=True) +class _StarExportFact: + file_path: Path + target_path: Path + line: int + + +@dataclass(frozen=True) +class _NamespaceExportFact: + file_path: Path + exported_name: str + target_path: Path + line: int + + +@dataclass(frozen=True) +class _SymbolUseFact: + file_path: Path + source_id: str + local_name: str + relation: str + context: str + line: int + + +@dataclass +class _SymbolResolutionFacts: + declarations: list[_SymbolDeclarationFact] = field(default_factory=list) + imports: list[_SymbolImportFact] = field(default_factory=list) + aliases: list[_SymbolAliasFact] = field(default_factory=list) + exports: list[_SymbolExportFact] = field(default_factory=list) + star_exports: list[_StarExportFact] = field(default_factory=list) + namespace_exports: list[_NamespaceExportFact] = field(default_factory=list) + uses: list[_SymbolUseFact] = field(default_factory=list) + # File-to-file submodule imports from `from pkg import submod` (#1146). + # Each entry is (importing_file, submodule_file, line). + module_imports: list[tuple[Path, Path, int]] = field(default_factory=list) + + +def _apply_symbol_resolution_facts( + paths: list[Path], + nodes: list[dict], + edges: list[dict], + root: Path, + facts: _SymbolResolutionFacts, +) -> None: + """Apply language-provided import/export/use facts to graph edges.""" + if not ( + facts.declarations + or facts.imports + or facts.aliases + or facts.exports + or facts.star_exports + or facts.namespace_exports + or facts.uses + or facts.module_imports + ): + return + + path_by_resolved = {path.resolve(): path for path in paths} + source_file_id = {path.resolve(): _make_id(str(path)) for path in paths} + symbol_nodes: dict[tuple[Path, str], str] = {} + for node in nodes: + source_path = _js_source_path(str(node.get("source_file", "")), root) + if source_path is None: + continue + label = str(node.get("label", "")).strip().strip("()").lstrip(".") + if label and node.get("id"): + symbol_nodes[(source_path, label)] = str(node["id"]) + + def ensure_symbol_node(path: Path, name: str, line: int) -> str: + resolved_path = path.resolve() + existing = symbol_nodes.get((resolved_path, name)) + if existing is not None: + return existing + node_id = _make_id(_file_stem(path), name) + symbol_nodes[(resolved_path, name)] = node_id + nodes.append({ + "id": node_id, + "label": name, + "file_type": "code", + "source_file": str(path), + "source_location": f"L{line}", + }) + return node_id + + existing_edges = { + ( + str(edge.get("source")), + str(edge.get("target")), + str(edge.get("relation")), + str(edge.get("context") or ""), + ) + for edge in edges + } + + def add_edge(source: str, target: str, relation: str, context: str, line: int, source_path: Path) -> None: + key = (source, target, relation, context or "") + if key in existing_edges: + return + existing_edges.add(key) + edges.append({ + "source": source, + "target": target, + "relation": relation, + "context": context, + "confidence": "EXTRACTED", + "source_file": str(source_path), + "source_location": f"L{line}", + "weight": 1.0, + }) + + for declaration in facts.declarations: + ensure_symbol_node(declaration.file_path, declaration.name, declaration.line) + + local_aliases_by_file: dict[Path, dict[str, tuple[Path, str]]] = {} + for import_fact in facts.imports: + file_path = import_fact.file_path.resolve() + local_aliases_by_file.setdefault(file_path, {})[import_fact.local_name] = ( + import_fact.target_path.resolve(), + import_fact.imported_name, + ) + + pending_aliases_by_file: dict[Path, list[_SymbolAliasFact]] = {} + for alias_fact in facts.aliases: + pending_aliases_by_file.setdefault(alias_fact.file_path.resolve(), []).append(alias_fact) + + for file_path, aliases in pending_aliases_by_file.items(): + local_aliases = local_aliases_by_file.setdefault(file_path, {}) + changed = True + while changed: + changed = False + for alias_fact in aliases: + if alias_fact.alias in local_aliases: + continue + origin = local_aliases.get(alias_fact.target_name) + if origin is not None: + local_aliases[alias_fact.alias] = origin + changed = True + + named_exports_by_file: dict[Path, dict[str, tuple[Path, str]]] = {} + star_exports_by_file: dict[Path, list[Path]] = {} + + for star_fact in facts.star_exports: + source_path = star_fact.file_path.resolve() + target_path = star_fact.target_path.resolve() + star_exports_by_file.setdefault(source_path, []).append(target_path) + source_id = source_file_id.get(source_path) + if source_id is not None: + add_edge( + source_id, + _make_id(str(path_by_resolved.get(target_path, target_path))), + "re_exports", + "export", + star_fact.line, + star_fact.file_path, + ) + + for namespace_fact in facts.namespace_exports: + source_path = namespace_fact.file_path.resolve() + target_path = namespace_fact.target_path.resolve() + namespace_id = ensure_symbol_node( + namespace_fact.file_path, + namespace_fact.exported_name, + namespace_fact.line, + ) + named_exports_by_file.setdefault(source_path, {})[ + namespace_fact.exported_name + ] = (source_path, namespace_fact.exported_name) + source_id = source_file_id.get(source_path) + if source_id is not None: + add_edge( + source_id, + namespace_id, + "contains", + "namespace_export", + namespace_fact.line, + namespace_fact.file_path, + ) + add_edge( + source_id, + _make_id(str(path_by_resolved.get(target_path, target_path))), + "re_exports", + "export", + namespace_fact.line, + namespace_fact.file_path, + ) + + for export_fact in facts.exports: + file_path = export_fact.file_path.resolve() + origin: tuple[Path, str] | None = None + if export_fact.target_path is not None and export_fact.target_name is not None: + origin = (export_fact.target_path.resolve(), export_fact.target_name) + elif export_fact.local_name is not None: + origin = local_aliases_by_file.get(file_path, {}).get(export_fact.local_name) + if origin is None and (file_path, export_fact.local_name) in symbol_nodes: + origin = (file_path, export_fact.local_name) + if origin is None: + continue + named_exports_by_file.setdefault(file_path, {})[export_fact.exported_name] = origin + if origin[0] != file_path: + source_id = source_file_id.get(file_path) + if source_id is not None: + add_edge( + source_id, + _make_id(str(path_by_resolved.get(origin[0], origin[0]))), + "re_exports", + "export", + export_fact.line, + export_fact.file_path, + ) + + def resolve_exported_origin(target_path: Path, imported_name: str, seen: set[tuple[Path, str]] | None = None) -> tuple[Path, str]: + target_path = target_path.resolve() + key = (target_path, imported_name) + if seen is None: + seen = set() + if key in seen: + return key + seen.add(key) + origin = named_exports_by_file.get(target_path, {}).get(imported_name) + if origin is not None: + return resolve_exported_origin(origin[0], origin[1], seen) + for star_target in star_exports_by_file.get(target_path, []): + star_key = (star_target, imported_name) + if star_key in symbol_nodes: + return star_key + resolved = resolve_exported_origin(star_target, imported_name, seen) + if resolved in symbol_nodes: + return resolved + return key + + for import_fact in facts.imports: + source_id = source_file_id.get(import_fact.file_path.resolve()) + if source_id is None: + continue + origin_path, origin_symbol = resolve_exported_origin( + import_fact.target_path, + import_fact.imported_name, + ) + target_id = symbol_nodes.get((origin_path, origin_symbol)) + if target_id is None: + continue + add_edge( + source_id, + target_id, + "imports", + "import", + import_fact.line, + import_fact.file_path, + ) + + # #1146: emit file-to-file imports_from edges for package-form submodule imports. + for from_path, to_path, line in facts.module_imports: + try: + from_rel = from_path.relative_to(root) + to_rel = to_path.relative_to(root) + except ValueError: + continue + source_id = _make_id(_file_stem(from_rel)) + target_id = _make_id(_file_stem(to_rel)) + add_edge(source_id, target_id, "imports_from", "submodule_import", line, from_path) + + for use_fact in facts.uses: + file_path = use_fact.file_path.resolve() + target_id = None + unresolved_origin = local_aliases_by_file.get(file_path, {}).get(use_fact.local_name) + if unresolved_origin is not None: + origin_path, origin_symbol = resolve_exported_origin(*unresolved_origin) + target_id = symbol_nodes.get((origin_path, origin_symbol)) + if target_id is None and use_fact.relation in ("inherits", "implements"): + # Same-file fallback for HERITAGE only: a base declared in the same + # file (`class X extends Y`, `interface A extends B`) has no import + # alias, so resolve it directly against the file's own symbol nodes. + # Scoped to heritage because same-file calls/uses already resolve via + # the dedicated call-graph pass; widening this would duplicate those + # edges. Import resolution still takes precedence (#1095). + target_id = symbol_nodes.get((file_path, use_fact.local_name)) + if target_id is None: + continue + add_edge( + use_fact.source_id, + target_id, + use_fact.relation, + use_fact.context, + use_fact.line, + use_fact.file_path, + ) + + +def _parse_js_tree(path: Path): + try: + from tree_sitter import Language, Parser + # .vue embeds the script in non-JS markup; mask it out and parse the + #