feat: upgrade @ladybugdb/core to 0.15.2 and remove segfault workarounds

[abhigyanpatwari]

Summary

• Upgrade @ladybugdb/core from ^0.15.1 to ^0.15.2 — includes the merged fix for child QueryResult lifetime segfault (ladybug-nodejs#1)
• Fix N-API destructor crashes that prevented lbug-db tests from running on macOS
• Add defense-in-depth write query validation to pool adapter executeQuery
• Unify CI test architecture — single vitest run with coverage, cross-platform smoke tests
• Update CI report with per-platform status and accurate test/coverage data

Root causes found & fixed

Issue	Root cause	Fix
0/8 lbug-db tests ran on macOS	global-setup.ts called db.close() without await — Promise<void> in 0.15.2 API left stale file lock	await conn.close() + await db.close()
Fork crash after 2-3 files with isolate: false	Repeated native DB open/close cycles corrupt vitest fork state	Removed isolate: false, each file gets own fork
poolAdapter tests segfault on macOS	writable→read-only switch calls db.close() which triggers N-API destructor segfault	initLbugWithDb() — pool adapter reuses core adapter's writable Database, no close needed
Unhandled Promise rejections from conn.close()	try { conn.close() } catch {} doesn't catch async rejections	Changed to conn.close().catch(() => {})
Write queries succeed on shared writable DB	Pool adapter had no application-level write protection	Added WRITE_QUERY_RE check in executeQuery (defense-in-depth)

Changes

File	Change
gitnexus/package.json	@ladybugdb/core ^0.15.1 → ^0.15.2
gitnexus-web/package.json	@ladybugdb/wasm-core ^0.15.1 → ^0.15.2
gitnexus/test/global-setup.ts	await async close() calls (were fire-and-forget)
gitnexus/vitest.config.ts	Removed isolate: false, updated comments
gitnexus/src/core/lbug/lbug-adapter.ts	Added getDatabase() export for pool adapter reuse
gitnexus/src/mcp/core/lbug-adapter.ts	initLbugWithDb() for external DB injection, WRITE_QUERY_RE in executeQuery, external flag in SharedDB, .catch() on fire-and-forget closes
gitnexus/test/helpers/test-indexed-db.ts	Pool adapter setup uses initLbugWithDb instead of close-and-reopen
.github/workflows/ci-tests.yml	Renamed from ci-unit-tests.yml
.github/workflows/ci.yml	Updated workflow reference
.github/workflows/ci-report.yml	Per-platform status columns, commit SHA, cleaner formatting

Test results

Platform	Files	Tests	Status
Ubuntu (coverage)	84	3077	✅
Windows	84	3077	✅
macOS	84	3077	✅ (was 0/8 lbug-db before)

All tests pass on all platforms. No tests skipped or modified — the initLbugWithDb + WRITE_QUERY_RE approach preserves all existing test assertions.

Test plan

• npx vitest run passes locally (84 files, 3077 tests, 0 errors)
• npx vitest run --project lbug-db passes (8 files, 111 tests)
• npx tsc --noEmit passes
• CI passes on Ubuntu, Windows, macOS
• Verify CI report comment shows accurate coverage data after merge

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
gitnexus	Ready	Preview, Comment	Mar 19, 2026 8:19am

[github-actions]

CI Report

✅ All checks passed

Pipeline Status

Stage	Status	Details
✅ Typecheck	success	tsc --noEmit
✅ Unit Tests	success	3 platforms
✅ Integration	success	3 OS x 4 groups = 12 jobs

Test Results

Suite	Tests	Passed	Failed	Skipped	Duration
Unit	1766	1765	0	1	8s
Integration	1348	1329	0	19	69s
Total	3114	3094	0	20	77s

✅ All 3094 tests passed

20 test(s) skipped — expand for details

Integration:

• Python match/case as-pattern type binding > resolves u.save() to User#save via match/case as-pattern binding
• Python match/case as-pattern type binding > does NOT resolve u.save() to Repo#save (negative disambiguation)
• Swift constructor-inferred type resolution > detects User and Repo classes, both with save methods
• Swift constructor-inferred type resolution > resolves user.save() to Models/User.swift via constructor-inferred type
• Swift constructor-inferred type resolution > resolves repo.save() to Models/Repo.swift via constructor-inferred type
• Swift constructor-inferred type resolution > emits exactly 2 save() CALLS edges (one per receiver type)
• Swift self resolution > detects User and Repo classes, each with a save function
• Swift self resolution > resolves self.save() inside User.process to User.save, not Repo.save
• Swift parent resolution > detects BaseModel and User classes plus Serializable protocol
• Swift parent resolution > emits EXTENDS edge: User → BaseModel
• Swift parent resolution > emits IMPLEMENTS edge: User → Serializable (protocol conformance)
• Swift cross-file User.init() inference > resolves user.save() via User.init(name:) inference
• Swift cross-file User.init() inference > resolves user.greet() via User.init(name:) inference
• Swift return type inference > detects User class and getUser function
• Swift return type inference > detects save function on User (Swift class methods are Function nodes)
• Swift return type inference > resolves user.save() to User#save via return type of getUser() -> User
• Swift return-type inference via function return type > resolves user.save() to User#save via return type of getUser()
• Swift return-type inference via function return type > user.save() does NOT resolve to Repo#save
• Swift return-type inference via function return type > resolves repo.save() to Repo#save via return type of getRepo()

Unit:

• buildTypeEnv > known limitations (documented skip tests) > Ruby block parameter: users.each { |user| } — closure param inference, different feature

Code Coverage

Combined (Unit + Integration)

Metric	Coverage	Covered	Base	Delta	Status
Statements	61.51%	7195/11697	46.88%	📈 +14.6	🟢 ████████████░░░░░░░░
Branches	55.44%	4996/9010	41.48%	📈 +14.0	🟢 ███████████░░░░░░░░░
Functions	60.91%	611/1003	46.56%	📈 +14.3	🟢 ████████████░░░░░░░░
Lines	63.46%	6428/10128	48.48%	📈 +15.0	🟢 ████████████░░░░░░░░

Coverage breakdown by test suite

Unit Tests

Metric	Coverage	Covered	Base	Delta	Status
Statements	46.83%	5478/11697	46.88%	📉 -0.1	🔴 █████████░░░░░░░░░░░
Branches	41.47%	3737/9010	41.48%	📉 -0.0	🔴 ████████░░░░░░░░░░░░
Functions	46.56%	467/1003	46.56%	＝ 0.0	🟢 █████████░░░░░░░░░░░
Lines	48.43%	4906/10128	48.48%	📉 -0.0	🔴 █████████░░░░░░░░░░░

Integration Tests

Metric	Coverage	Covered	Base	Delta	Status
Statements	47.08%	5508/11697	46.88%	📈 +0.2	🟢 █████████░░░░░░░░░░░
Branches	44.15%	3978/9010	41.48%	📈 +2.7	🟢 ████████░░░░░░░░░░░░
Functions	44.56%	447/1003	46.56%	📉 -2.0	🔴 ████████░░░░░░░░░░░░
Lines	48.25%	4887/10128	48.48%	📉 -0.2	🔴 █████████░░░░░░░░░░░

---

_{📋 View full run · Generated by CI}

[xkonjin]

Code Review

Overall: Solid upgrade that properly addresses the N-API segfault workarounds. The move from "let the OS reclaim on exit" to explicit .close() calls is a meaningful reliability improvement.

Key Changes Reviewed

Pool connection lifecycle (lbug-adapter.ts)

• The new closed flag on PoolEntry is the right pattern. Connections checked out during eviction get closed on checkin() instead of being silently orphaned. This fixes a real resource leak.
• closeOne() now properly closes available connections before decrementing the shared DB refcount. The ordering is correct: close connections first, then conditionally close the DB when refCount hits 0.

Test cleanup

• Removing the platform-gated .close() calls (previously Windows-only) and making cleanup universal is the correct call given the 0.15.2 fixes.
• Deleting test/setup.ts entirely is clean. The _getActiveHandles unref hack was always fragile.

Notes

1. dangerouslyIgnoreUnhandledErrors still present - The comment explains why (vitest forks + N-API destructor ordering at exit). This is reasonable as a transient workaround, but the TODO should be tracked somewhere to avoid it becoming permanent. Consider filing an issue upstream.

2. try { conn.close(); } catch {} pattern - Used in several places (closeOne, checkin). Silent catches are fine here since these are cleanup paths where failure is non-actionable, but a debug-level log would help diagnose issues if they recur.

3. Race window in closeOne - Between setting entry.closed = true and pool.delete(repoId), a concurrent checkout() could grab a connection from entry.available that is about to be closed in the loop. The available.length = 0 at the end mitigates this, but since the close loop iterates the array while checkout might .pop() from it, there is a narrow window. Since this is single-threaded Node.js and there are no awaits in the loop, this is safe in practice, but worth a comment.

4. tree-sitter-swift removed from lockfile - Looks like an unrelated cleanup that landed in this PR. Not a problem, just noting it.

[xkonjin]

Code Review

Clean upgrade — removes the segfault workarounds that were papered over the N-API destructor bug in @ladybugdb/core 0.15.1. A few notes:

Key Observations

1. Connection pool closed flag — The new closed field on PoolEntry with the checkin() orphan-close logic is a good pattern. One edge case: if a connection is checked out, the entry is closed (LRU eviction), and then checkin closes the connection — but if closeOne is called again for the same repoId before checkin happens (e.g. rapid re-init), pool.delete(repoId) already ran, so checkin would need to handle a missing pool entry. Currently it receives the entry object directly (not looked up by key), so this is actually safe. Just noting the reasoning for future readers.

2. Database refCount close — closeOne now decrements shared.refCount and closes the Database when it hits 0. The previous code only decremented without closing. This is correct now that the N-API destructors are fixed, but the dangerouslyIgnoreUnhandledErrors: true in vitest.config.ts is still there with a TODO — make sure that gets tracked.

3. Removed test/setup.ts — The old afterAll that unref'd active handles is gone. This is correct since 0.15.2 handles cleanup properly, but verify CI passes on all platforms (the comment in vitest.config.ts mentions macOS still has destructor ordering issues at exit).

4. tree-sitter-swift removed from package.json — This seems unrelated to the ladybugdb upgrade. Is this intentional? If Swift parsing support is being dropped, it should probably be its own commit or at least mentioned in the PR description.

Good

• Test helpers now properly close resources instead of relying on process exit, which is the right approach.
• The prebuilt platform binaries in optionalDependencies should eliminate the cmake-js build step on CI.
• Comments explain why dangerouslyIgnoreUnhandledErrors is still needed (exit-time destructor ordering, not the fixed QueryResult lifetime issue).

Looks good. Minor question on the tree-sitter-swift removal.

[xkonjin]

Code Review

Overall: Major quality-of-life upgrade. Bumps @ladybugdb/core to 0.15.2, removes segfault workarounds, merges the CI integration test matrix into a single test job, and properly closes native handles. This is a significant reduction in CI complexity and tech debt.

Key Changes

Native handle cleanup is now correct. The closeOne() function properly:

1. Marks the pool entry as closed to prevent orphaned connections
2. Closes available connections immediately
3. Handles in-flight connections via checkin() (detects entry.closed and closes on return)
4. Reference-counts the shared Database and only closes when refCount === 0

This is a clean fix for a real resource leak / crash pattern.

CI simplification is well-structured. Merging unit + integration into a single tests workflow reduces the matrix from 3 OS x 4 groups = 12 jobs to 3 OS x 1 = 3 jobs, which should cut CI cost and time significantly. The ci-report.yml changes consistently rename unit/integration to the unified tests naming.

Concerns

1. dangerouslyIgnoreUnhandledErrors is still enabled in vitest.config.ts. The TODO says "remove once LadybugDB fixes all N-API destructor ordering issues." Now that 0.15.2 fixes the segfault-on-close path, is this flag still needed? If the main segfault was the closeSync() destructor bug, this upgrade might have fixed it. Worth testing a CI run with the flag removed to see if it passes clean.

2. tree-sitter-swift was removed from package.json dependencies (line in the lockfile diff). If this is intentional, the PR description should mention it. If any Swift parsing tests exist, they would silently fail or skip.

3. Integration test coverage reporting is gone. The old CI ran integration tests separately with --coverage and merged coverage. The new unified npx vitest run with coverage should cover both, but the test:integration script still exists in package.json. This could confuse contributors who run npm run test:integration locally and expect it to still work as a standalone command.

4. Potential race in closeOne() + checkin() interaction: If closeOne() is called while a connection is checked out, entry.closed = true is set. When checkin() runs later, it checks entry.closed and closes the connection. But closeOne() also calls pool.delete(repoId), so the entry object is orphaned from the pool. This works because checkin() receives the entry reference directly (not via pool lookup), but it's worth a comment noting this contract.

No security issues found. The resource cleanup improvements are a net positive for stability.

[github-actions]

CI Report

✅ All checks passed   e637bf1

Pipeline

Stage	Status	Ubuntu	Windows	macOS
Typecheck	✅ success	—	—	—
Tests	✅ success	✅	✅	✅

Tests

Metric	Value
Total	3301
Passed	3281
Skipped	20
Files	932
Duration	1m 42s

✅ All 3281 tests passed across 932 files

20 test(s) skipped

• buildTypeEnv > known limitations (documented skip tests) > Ruby block parameter: users.each { |user| } — closure param inference, different feature
• Python match/case as-pattern type binding > resolves u.save() to User#save via match/case as-pattern binding
• Python match/case as-pattern type binding > does NOT resolve u.save() to Repo#save (negative disambiguation)
• Swift constructor-inferred type resolution > detects User and Repo classes, both with save methods
• Swift constructor-inferred type resolution > resolves user.save() to Models/User.swift via constructor-inferred type
• Swift constructor-inferred type resolution > resolves repo.save() to Models/Repo.swift via constructor-inferred type
• Swift constructor-inferred type resolution > emits exactly 2 save() CALLS edges (one per receiver type)
• Swift self resolution > detects User and Repo classes, each with a save function
• Swift self resolution > resolves self.save() inside User.process to User.save, not Repo.save
• Swift parent resolution > detects BaseModel and User classes plus Serializable protocol
• Swift parent resolution > emits EXTENDS edge: User → BaseModel
• Swift parent resolution > emits IMPLEMENTS edge: User → Serializable (protocol conformance)
• Swift cross-file User.init() inference > resolves user.save() via User.init(name:) inference
• Swift cross-file User.init() inference > resolves user.greet() via User.init(name:) inference
• Swift return type inference > detects User class and getUser function
• Swift return type inference > detects save function on User (Swift class methods are Function nodes)
• Swift return type inference > resolves user.save() to User#save via return type of getUser() -> User
• Swift return-type inference via function return type > resolves user.save() to User#save via return type of getUser()
• Swift return-type inference via function return type > user.save() does NOT resolve to Repo#save
• Swift return-type inference via function return type > resolves repo.save() to Repo#save via return type of getRepo()

Coverage

Metric	Coverage	Covered	Base (main)	Delta
Statements	67.08%	7864/11722	46.88%	📈 +20.2%
Branches	58.56%	5281/9018	41.48%	📈 +17.1%
Functions	68.97%	696/1009	46.56%	📈 +22.4%
Lines	69.42%	7049/10154	48.48%	📈 +20.9%

---

_{📋 Full run · Coverage from Ubuntu · Generated by CI}