CREATE_FTS_INDEX segfaults on Linux x86_64 with non-trivial data volumes

[WingedGuardian]

Environment

• OS: Ubuntu Linux x86_64 (6.8.0-101-generic)
• Node.js: v22.x
• @ladybugdb/core: v0.15.4 and v0.15.4-dev.20260428
• Consumer: GitNexus v1.6.4-rc.21

Description

CREATE_FTS_INDEX segfaults (SIGSEGV, exit code 139) when called on node tables with non-trivial data volumes on Linux x86_64. The FTS extension loads successfully — the crash occurs specifically during index creation.

Works: In-memory DB with 2 rows — index creates, queries return correct results.
Crashes: On-disk DB with 1,000+ rows — segfaults immediately on CREATE_FTS_INDEX.

Minimal Reproduction

const Database = require("@ladybugdb/core/database.js");
const Connection = require("@ladybugdb/core/connection.js");

async function test() {
  // This works fine:
  const memDb = new Database(":memory:");
  const memConn = new Connection(memDb);
  await memConn.init();
  await memConn.query("CREATE NODE TABLE T(id INT64, name STRING, content STRING, PRIMARY KEY(id))");
  await memConn.query("CREATE (n:T {id: 1, name: 'foo', content: 'hello world'})");
  await memConn.query("LOAD EXTENSION fts");
  await memConn.query("CALL CREATE_FTS_INDEX('T', 'test_fts', ['name', 'content'], stemmer := 'porter')");
  console.log("In-memory: OK"); // ✅ Works

  // This segfaults:
  const diskDb = new Database("/path/to/existing/db/with/1000+/rows");
  const diskConn = new Connection(diskDb);
  await diskConn.init();
  await diskConn.query("LOAD EXTENSION fts");
  // Next line causes SIGSEGV:
  await diskConn.query("CALL CREATE_FTS_INDEX('SomeTable', 'some_fts', ['name', 'content'], stemmer := 'porter')");
}
test();

Likely Root Cause

This appears to be inherited from the upstream KuzuDB codebase:

• KuzuDB #6017: All FTS end-to-end tests were failing FSM (Free Space Manager) leak checks
• KuzuDB #5671: Assertion failure in hash_index.cpp:174 during FTS checkpoint path when processing non-trivial data volumes

KuzuDB was archived with these bugs unresolved, and LadybugDB inherited the affected code.

Downstream Impact

• GitNexus issue #1160 tracks this — GitNexus's query tool returns empty results on Linux because FTS indexes can never be created
• Affects all GitNexus Linux users (FTS couldn't even load before v0.15.4 due to GLIBC requirements, and now that it loads, it segfaults on index creation)

Additional Notes

• Tested with both @ladybugdb/core@0.15.4 (stable) and @ladybugdb/core@0.15.4-dev.20260428 (today's nightly) — same crash on both
• Indexing name-only columns (excluding content) also segfaults — it's not a data size issue per row, but total row count
• The segfault is in the native binary (lbugjs.node), not in JavaScript code

[adsharma]

The old extension binaries are incompatible due to ABI changes. We need to respin them. Root cause is #332

The simplest fix is to bump up the major rev to 0.16.0 and update the extensions.

[adsharma]

v0.16.0 is out. Please retest.

[adsharma]

Merged LadybugDB/extensions#1. Concurrent access to the unordered_map could be the root cause. Please retest when 0.17.x is released.