[Tomasz Naumowicz (tnaum-ms)]

Summary

Make the SRV DNS resolver tests reliable on Windows and restricted CI runners without changing production resolver behavior.

Why

The UDP refusal test expected a closed localhost port to promptly produce ECONNREFUSED. Windows or restricted runners can suppress that notification and produce a timeout instead. Other tests also depended on short real-time deadlines, delayed TCP writes, or randomly generated query IDs being different. Increasing timeouts would retain these OS-dependent and probabilistic assumptions, so the tests now control the relevant events directly.

Changes

• Inject ECONNREFUSED and Windows-style ECONNRESET without sending a datagram, and verify immediate rejection and cleanup.
• Freeze clocks and resolver timers; advance time explicitly only in timeout scenarios and assert timer cleanup.
• Split client TCP data events deterministically instead of relying on delayed writes to create separate frames.
• Remove the random-ID inequality assertion and explicitly synchronize probe cancellation.
• Guard system DNS discovery and driver lookups against accidental real-network use.
• Clean up partially started DNS stubs, retry address-in-use conflicts, and add five lifecycle regression tests.
• Document the test-isolation guarantees and remaining loopback requirement.
• Keep the forwarded TCP event result explicitly boolean and its saved signature non-overloaded for both root and strict package-test configurations.
• Align the Case 2 checklists with CI: require strict workspace test typechecks and unconditional localization validation.

Scope

Test code, test helpers, feature documentation, and Case 2 verification instructions only. No production resolver changes, no external DNS access, and no changes to CI permissions. Localhost UDP/TCP sockets must still be permitted.

Validation — Case 2 (2026-10-21)

• npm run l10n — not applicable: no vscode.l10n.t() strings added, changed, or removed.
• npm run prettier-fix — passed.
• npm run l10n:check — passed; localization is up to date.
• npm run lint — passed after fixing the TCP event mock return type.
• npx vitest run — passed: 315 files, 5,190 tests passed, 1 skipped.
• npm run build — passed.
• npm run typecheck:tests — passed across all workspaces after reproducing and fixing the CI errors.
• npm run package — passed: VSIX generated successfully. Non-fatal dependency annotation and CSS warnings were emitted.
• After the final lint correction, formatting, full lint, build, and all 84 resolver tests passed again.
• Ten shuffled-order package runs also passed during implementation.
• Validation ran on Linux. Windows and restricted-runner behavior remain to be verified by CI.

Release and readiness

• Assigned to the existing 1.0.2 milestone.
• The operator explicitly requested skipping the AI review-agent/pre-review step for this PR. No AI review agent was launched and no review record was fabricated.

Context

• Related: mongodb+srv:// connections fail with "querySrv EBADRESP" on some networks (c-ares ≥ 1.34.7 rejects compressed SRV targets) #993 (original SRV fallback; not closed by this test-only change).
• Feature documentation: SRV DNS fallback.

Static-check failure follow-up

The initial local Case 2 list omitted npm run typecheck:tests, which CI runs separately. Root build and Vitest runtime success did not cover strict package-test typing: Socket.emit.call forwarding produced TS2554 and TS2556. The saved emitter now has an explicit rest-argument signature. The exact CI workspace typecheck, build, resolver tests, lint, and formatting passed locally. Both contributor and Copilot Case 2 checklists now require that script and npm run l10n:check, and explain why runtime tests are not a substitute.

[github-actions]

✅ CI passed: standard tier (ready PR)

Commit 7086750, workflow run.

Layer	Result	Ran or skipped because
Static checks	✅ passed	every push
Unit tests	✅ passed	every push
npm packages: verify tarballs	✅ passed	packages/**, package-lock.json or the CI workflow changed
VSIX: inspect contents	✅ passed	ready PR
VSIX: self-test of the checks	⏭️ skipped	no change to build/verification/**, build/vite/**, the Vite configs, package-lock.json or the CI workflow
VSIX: activate in VS Code	✅ passed	ready PR
VSIX: self-test of the activation check	⏭️ skipped	no change to build/verification/**, build/vite/**, the Vite configs, package-lock.json or the CI workflow
Webview host in unbundled extensions	⏭️ skipped	no change to packages/vscode-ext-webview/**, package-lock.json or the CI workflow
E2E	⏭️ not available yet	arrives with the E2E iteration (full tier)

Full run: add the ci:full-requested label to run every layer once on the current commit.
No full run yet.

Code quality

Check	Status	How to fix
Localization (l10n)	✅ Passed	Nothing to fix
ESLint	✅ Passed	Nothing to fix
Prettier formatting	✅ Passed	Nothing to fix

Build size

No page, activation, VSIX or other bundle size changed by 5% or 50 KiB or more against main. VSIX: 8.5 MiB.

Artifacts: VSIX | Bundle-reports-39580842714 | L1-manifest-report-39580842714

All graphs (exact bytes)

Graph	PR bytes	Base bytes	Delta bytes	Delta %
Collection View page (collectionView)	6,036,286	6,036,286	✅ 0	0.0%
Document View page (documentView)	4,808,762	4,808,762	✅ 0	0.0%
Local Quick Start page (localQuickStart)	1,551,408	1,551,408	✅ 0	0.0%
Atlas Credentials page (atlasCredentials)	1,504,612	1,504,612	✅ 0	0.0%
Cluster Dashboard page (clusterDashboard)	1,550,863	1,550,863	✅ 0	0.0%
Activation (mainStartup)	3,447,535	3,447,535	✅ 0	0.0%
Extension host (main)	9,974,227	9,974,227	✅ 0	0.0%
Playground worker (playgroundWorker)	5,384,529	5,384,529	✅ 0	0.0%
VSIX (vsix)	8,940,785	8,940,785	✅ 0	0.0%
playgroundTsPlugin	6,175	6,175	✅ 0	0.0%
viewsEntry	1,456,593	1,456,593	✅ 0	0.0%
viewsAssets	80,340	80,340	✅ 0	0.0%
editorWorker	259,528	259,528	✅ 0	0.0%
jsonWorker	389,389	389,389	✅ 0	0.0%

Packaged bytes. Shared chunks count in every graph that loads them, so rows do not add up.

Updated in place on every run.

[Tomasz Naumowicz (tnaum-ms)]

The static-check failure is fixed and the local verification instructions now match CI.

• Reproduced TS2554 and TS2556 with the exact npm run typecheck:tests command, then fixed the saved TCP emitter signature for strict package-test typing.
• Updated both Case 2 checklists to require npm run typecheck:tests and unconditional npm run l10n:check, build before strict test validation, and explain why build/Vitest runtime success is not a substitute.
• Verified locally: all workspace strict test typechecks, build, resolver tests (84 passed), localization check, formatting, and lint.
• Latest CI is green, including Static checks, the strict package-test step, and Merge gate: https://github.com/microsoft/vscode-documentdb/actions/runs/39580842714

The earlier local Case 2 result covered the documented commands at the time, but missed the separate strict test-typecheck that CI required. The updated checklist closes that gap.