[pablodiazjorge]

Summary

Two SQLite boundary tests fail on native Windows because of assumptions in the tests themselves; the provider behaves correctly. No issue is linked: I found them while verifying inspect_sqlite_database (#2271) on Windows, which CI does not cover (sqlite-runtime.yml runs on Ubuntu only). The fix touches only these two test files, like #2312. The same verification found a real misclassification of EBUSY on Windows, reported separately in #2822.

Problem and expected behavior

Smallest trigger, on Windows with Node 24.18.0: npx vitest run tests/boundary/sqlite/sqliteDatabaseProvider.test.ts tests/boundary/sqlite/sqliteDatabaseDescriptorCleanup.test.ts gives 2 failures on 117bd23dd. Expected: both tests reach their assertions and pass.

Test	Failure on Windows	Cause
sqliteDatabaseProvider › cancels an acquired SQLite worker and removes its private snapshot	ProviderAdapterError ("Provider process fixture exited before readiness") instead of AnalysisCancelledError	The test builds the provider with an empty environment ({}). Without SystemRoot, Node aborts at startup on Windows (Assertion failed: ncrypto::CSPRNG(nullptr, 0), exit 134), so the fixture never prints ready
sqliteDatabaseDescriptorCleanup › retains reply-read cancellation and retries the failed reply descriptor close	top-level execution_failure: "execution_failure" instead of "cancelled"	The test keeps the reply descriptor open on purpose. Windows cannot remove the private root while a file inside it is open (Delete owned object by handle failed […] (Win32 145), directory not empty), so the provider reports the root cleanup failure and keeps the cancellation as diagnostics.previous_error. POSIX unlinks open files, so the root is removed there

Change and scope

• sqliteDatabaseProvider.test.ts: the cancellation test passes process.env, like createProvider in the same file.
• sqliteDatabaseDescriptorCleanup.test.ts: on Windows, the reply-read test expects the root cleanup error with the original cancellation as its previous error (same execution_failure, resource and captured output as before). Elsewhere the expectation is unchanged. The shutdown retry assertions that follow are unchanged and pass on both platforms.

No product code changes.

Contribution scope: not applicable (test-only fix within existing boundaries).

Contract and boundary impact

• Semantic owner and earliest changed stage: test files only.
• Contribution scope fit and linked maintainer decision: not applicable.
• CLI and MCP/tool-catalog contract: none.
• Provider, bridge, target-format, or platform compatibility: none; the tests now run on native Windows.
• Evidence, artifact, provenance, or reconstruction contract: none.
• Target selection, authority, upstream behavior, and cleanup details (if applicable): not applicable.
• Generated metadata (docs/public/product-catalog.json), package, or installation impact: none.

Evidence and regression coverage

• Tests added or updated: the two tests above.
• Base reproduction or other evidence: on 117bd23dd the SQLite test files give 2 failed, 89 passed, 9 skipped on Windows; the errors are in the table.
• Remaining proof gaps: macOS not run.

Validation performed

Windows 11 x64 (10.0.26300), non-elevated, Node 24.18.0, with the Windows native addon built from the same commit.

• The 8 SQLite test files (git ls-files | grep sqlite | grep test.ts) on 117bd23dd: 2 failed, 89 passed, 9 skipped.
• The same files with this change: 91 passed, 9 skipped.
• Linux (WSL Ubuntu, Node 24.18.0), same files with this change: 98 passed, 2 skipped.
• npx oxfmt --check on the two files: passed.
• npm run check:fast: typecheck passed; lint reports two max-lines-per-function errors that are already on main (ghidraLauncher.test.ts and GhidraFunctionValues.test.ts, from fix(ghidra): warn when callee/caller reads stop at a truncated function body #2360), none in the changed files.

Compatibility, safety, and release

• Breaking changes or migration steps: none.
• Real Hopper/Ghidra, browser, or OS coverage: native Windows for these suites; no provider involved.
• Package or release metadata impact: none.
• Security, privacy, process, or containment review: none; tests only.

Before merge

• This PR addresses a concrete problem or an agreed enhancement.
• Material scope and authority questions are resolved, and any required maintainer scope decision covers the final diff.
• The PR has one focused outcome and the title follows type(scope): outcome.
• Related issue is linked, or the reason for not linking one is stated above.
• Tests cover changed observable behavior and meaningful failure paths.
• Owning docs, contracts, and generated metadata are updated where needed.
• User-visible CLI/MCP changes include representative output.
• I checked the final diff for secrets, unrelated cleanup, and unsupported claims.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-13T18:56:59.704200Z	e9a946b	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[greptile-apps]

Confidence Score: 5/5

[Low impact] The test-only changes appear safe to merge.

Summary

Updates two SQLite boundary tests for native Windows without changing product code.

• SQLite boundary tests now account for Windows worker startup and file cleanup.

_{Reviews (5) · Last reviewed commit: "test(sqlite): fix two SQLite boundary te..." · Reviewed by Greptile}

[pablodiazjorge]

Closing, since these changes landed on main in 5b7fe94: the descriptor cleanup test as proposed, and the provider test with only SystemRoot on Windows. Thanks!