Finally working!

Author: saulshanabrook

.gitignore

@@ -87,3 +87,4 @@ visualizer.tgz
 package
 .mypy_cache/
 *.json
+python/egglog/exp/param_eq/trace/

AGENTS.md

@@ -44,6 +44,7 @@
 ## Verification
 
 - Prefer the minimal code change and the minimal diff that solves the task; only broaden the change if the smaller fix is not sufficient.
+- For long-running profiling or trace probes, run them with explicit timeouts, check for lingering worker processes before and after, and inspect memory usage after any timeout or manual kill before starting the next experiment.
 - Run `make mypy` for typing changes.
 - Run targeted pytest for touched modules.
 - Run `make docs` for docs or public API changes.

python/egglog/bindings.pyi

@@ -78,6 +78,7 @@ __all__ = [
     "RustSpan",
     "Saturate",
     "Scan",
+    "SchedulerHandle",
     "Schema",
     "Sequence",
     "SerializedEGraph",
@@ -135,6 +136,22 @@ class EGraph:
     def run_program(
         self, *commands: _Command, traceparent: str | None = None, tracestate: str | None = None
     ) -> list[_CommandOutput]: ...
+    def add_backoff_scheduler(
+        self,
+        match_limit: int,
+        ban_length: int,
+        *,
+        egg_like: bool = False,
+        haskell_backoff: bool = False,
+    ) -> SchedulerHandle: ...
+    def run_ruleset_with_scheduler(
+        self,
+        ruleset: str,
+        scheduler: SchedulerHandle,
+        *,
+        traceparent: str | None = None,
+        tracestate: str | None = None,
+    ) -> RunReport: ...
     def serialize(
         self,
         root_eclasses: list[_Expr],
@@ -175,6 +192,9 @@ class Value:
     def __gt__(self, other: object) -> bool: ...
     def __ge__(self, other: object) -> bool: ...
 
+@final
+class SchedulerHandle: ...
+
 @final
 class EggSmolError(Exception):
     context: str

python/egglog/egraph.py

@@ -976,6 +976,31 @@ def _run_schedule(self, schedule: Schedule) -> bindings.RunReport:
         assert isinstance(command_output, bindings.RunScheduleOutput)
         return command_output.report
 
+    def _add_backoff_scheduler(
+        self,
+        *,
+        match_limit: int,
+        ban_length: int,
+        egg_like: bool,
+        haskell_backoff: bool = False,
+    ) -> bindings.SchedulerHandle:
+        return self._egraph.add_backoff_scheduler(
+            match_limit,
+            ban_length,
+            egg_like=egg_like,
+            haskell_backoff=haskell_backoff,
+        )
+
+    def _run_ruleset_with_scheduler(
+        self,
+        ruleset: Ruleset | UnstableCombinedRuleset,
+        scheduler: bindings.SchedulerHandle,
+    ) -> bindings.RunReport:
+        self._add_decls(ruleset)
+        ruleset_ident = ruleset.__egg_ident__
+        self._state.ruleset_to_egg(ruleset_ident)
+        return call_with_current_trace(self._egraph.run_ruleset_with_scheduler, str(ruleset_ident), scheduler)
+
     def stats(self) -> bindings.RunReport:
         """
         Returns the overall run report for the egraph.

python/egglog/exp/param_eq/Makefile

@@ -1,9 +1,10 @@
 # Canonical local workflow for the retained param-eq replication package.
 # The file dependencies document the expected order:
 # 1. normalize archived Haskell outputs
-# 2. run the Egglog corpus against those normalized rows
-# 3. refresh the golden cases and notebook
-# 4. run the replication-local tests
+# 2. run the current local Haskell pipeline across the same rows
+# 3. run the Egglog corpus baseline and scheduler ablations against those rows
+# 4. refresh the golden cases and notebook
+# 5. run the replication-local tests
 
 PARAM_EQ_DIR := $(abspath $(dir $(lastword $(MAKEFILE_LIST))))
 REPO_ROOT := $(abspath $(PARAM_EQ_DIR)/../../../..)
@@ -13,9 +14,12 @@ export EGGLOG_PARAM_EQ_DATA_DIR
 ARTIFACT_DIR := $(PARAM_EQ_DIR)/artifacts
 GOLDEN := $(PARAM_EQ_DIR)/haskell_golden.json
 NOTEBOOK := $(PARAM_EQ_DIR)/replication.ipynb
-HASKELL_ROWS := $(ARTIFACT_DIR)/haskell_paper_rows.csv
+ARCHIVED_HASKELL_ROWS := $(ARTIFACT_DIR)/haskell_paper_rows.csv
+LIVE_HASKELL_ROWS := $(ARTIFACT_DIR)/haskell_live_rows.csv
 RUNTIME_ROWS := $(ARTIFACT_DIR)/pagie_runtime_scatter.csv
 EGGLOG_ROWS := $(ARTIFACT_DIR)/egglog_paper_rows.csv
+EGGLOG_ABLATION_ROWS := $(ARTIFACT_DIR)/egglog_ablation_rows.csv
+ABLATION_SUMMARY := $(ARTIFACT_DIR)/ablation_summary.csv
 
 HASKELL_FIXTREE := $(EGGLOG_PARAM_EQ_DATA_DIR)/src/FixTree.hs
 HASKELL_MAIN := $(EGGLOG_PARAM_EQ_DATA_DIR)/src/Main.hs
@@ -28,23 +32,41 @@ HASKELL_EXPRS := \
 	$(wildcard $(HASKELL_RESULTS)/exprs/*_exprs_*) \
 	$(wildcard $(HASKELL_RESULTS)/exprs_simpl/*_exprs_*)
 
-.PHONY: golden artifacts notebook test all
+.PHONY: golden archived-artifacts live-haskell artifacts ablations notebook test all
 
-$(HASKELL_ROWS) $(RUNTIME_ROWS): $(PARAM_EQ_DIR)/normalize_archives.py $(PARAM_EQ_DIR)/paths.py $(HASKELL_TABLES) $(HASKELL_EXPRS) $(HASKELL_RUNTIME)
+$(ARCHIVED_HASKELL_ROWS) $(RUNTIME_ROWS): $(PARAM_EQ_DIR)/normalize_archives.py $(PARAM_EQ_DIR)/paths.py $(HASKELL_TABLES) $(HASKELL_EXPRS) $(HASKELL_RUNTIME)
 	cd $(REPO_ROOT) && uv run python -m egglog.exp.param_eq.normalize_archives
 
-$(EGGLOG_ROWS): $(PARAM_EQ_DIR)/run_egglog_corpus.py $(PARAM_EQ_DIR)/pipeline.py $(HASKELL_ROWS) $(RUNTIME_ROWS)
+$(LIVE_HASKELL_ROWS): $(PARAM_EQ_DIR)/run_haskell_corpus.py $(PARAM_EQ_DIR)/generate_haskell_golden.py $(PARAM_EQ_DIR)/pipeline.py $(PARAM_EQ_DIR)/paths.py $(ARCHIVED_HASKELL_ROWS) $(HASKELL_FIXTREE) $(HASKELL_MAIN)
+	cd $(REPO_ROOT) && uv run python -m egglog.exp.param_eq.run_haskell_corpus
+
+$(EGGLOG_ROWS): $(PARAM_EQ_DIR)/run_egglog_corpus.py $(PARAM_EQ_DIR)/pipeline.py $(ARCHIVED_HASKELL_ROWS) $(RUNTIME_ROWS)
 	cd $(REPO_ROOT) && uv run python -m egglog.exp.param_eq.run_egglog_corpus
 
-$(GOLDEN): $(PARAM_EQ_DIR)/generate_haskell_golden.py $(PARAM_EQ_DIR)/pipeline.py $(PARAM_EQ_DIR)/paths.py $(HASKELL_FIXTREE) $(HASKELL_MAIN) $(HASKELL_ROWS)
+$(EGGLOG_ABLATION_ROWS): $(PARAM_EQ_DIR)/run_egglog_corpus.py $(PARAM_EQ_DIR)/pipeline.py $(ARCHIVED_HASKELL_ROWS) $(LIVE_HASKELL_ROWS)
+	cd $(REPO_ROOT) && uv run python -m egglog.exp.param_eq.run_egglog_corpus \
+		--output $(EGGLOG_ABLATION_ROWS) \
+		--mode egglog-baseline \
+		--mode no-haskell-backoff \
+		--mode no-graph-size-stop \
+		--mode no-bound-scheduler \
+		--mode no-fresh-rematch
+
+$(GOLDEN): $(PARAM_EQ_DIR)/generate_haskell_golden.py $(PARAM_EQ_DIR)/pipeline.py $(PARAM_EQ_DIR)/paths.py $(HASKELL_FIXTREE) $(HASKELL_MAIN) $(ARCHIVED_HASKELL_ROWS)
 	cd $(REPO_ROOT) && uv run python -m egglog.exp.param_eq.generate_haskell_golden
 
-$(NOTEBOOK): $(PARAM_EQ_DIR)/replication.py $(HASKELL_ROWS) $(EGGLOG_ROWS) $(RUNTIME_ROWS)
+$(NOTEBOOK): $(PARAM_EQ_DIR)/replication.py $(ARCHIVED_HASKELL_ROWS) $(LIVE_HASKELL_ROWS) $(EGGLOG_ROWS) $(ABLATION_SUMMARY) $(RUNTIME_ROWS)
 	cd $(REPO_ROOT) && uv run python python/egglog/exp/param_eq/replication.py
 
+archived-artifacts: $(ARCHIVED_HASKELL_ROWS) $(RUNTIME_ROWS)
+
+live-haskell: $(LIVE_HASKELL_ROWS)
+
 golden: $(GOLDEN)
 
-artifacts: $(EGGLOG_ROWS)
+artifacts: $(ARCHIVED_HASKELL_ROWS) $(RUNTIME_ROWS) $(LIVE_HASKELL_ROWS) $(EGGLOG_ROWS)
+
+ablations: $(EGGLOG_ABLATION_ROWS)
 
 notebook: $(NOTEBOOK)