Commit Graph
2 Commits
Author SHA1 Message Date
morluto 116f159a8a fix: isolate concurrent doctor cleanup scopes, trace JS fixture helpers, and share MCP/home lifecycles
Address all four PR review findings structurally:
- Isolate concurrent Doctor probes and host subsystems in child
  ArtifactResourceScope instances tracked via resources.track() so a failed
  cleanup on one probe cannot deadlock or abort sibling reads before close().
- Include tests/ JS/TS helper modules and fixture path literals in
  throughHelpers() and named() so changes to JS fixtures such as
  fakeHopperReplies.mjs reach transitive helper and socket tests.
- Extract generic createSharedMcpSession() and reuse it in lazyLocalToolsMcp()
  to preserve retryable cleanup draining and startup-failure attribution.
- Replace test-scoped createTestTempDirectory() with a file-scoped workspace
  in jsonInputStatus.test.ts so the shared isolated home persists across tests.
2026-11-17 19:58:40 +08:00
40428da1dd perf(test): reuse one stdio MCP server per JavaScript application file (#5658)
* perf(test): reuse one stdio MCP server per JavaScript application file

Repeated analyzeJavaScriptDirectory MCP cases started a full server for
every case. Share one file-scoped server and compile cache so later cases
skip that startup while CLI cases still launch a fresh process.

Co-authored-by: morluto <morluto@users.noreply.github.com>

* fix(test): close in-flight shared MCP startup with the file

File teardown waits for a stdio session that is still starting, skips a
server once close has begun, and still closes one registered during that
wait. The later-case reuse test can run without the earlier case.

Co-authored-by: morluto <morluto@users.noreply.github.com>

* fix(test): keep a failed MCP handshake when cleanup fails

The shared session reports the startup error and carries the cleanup failure
as its cause. A cleanup that failed stays owned and is retried on close
without running a cleanup that already succeeded.

Co-authored-by: morluto <morluto@users.noreply.github.com>

* fix(test): fail shared MCP cleanup when the rejection has no reason

A cleanup rejection of undefined still fails the drain and stays owned.
Close caches completion only after every retained cleanup succeeds, so a
later close can release it.

Co-authored-by: morluto <morluto@users.noreply.github.com>

* fix(test): keep one MCP server while failed cleanup is retained

A later call retries retained cleanup before opening another server. The
new server starts only after that cleanup succeeds.

Co-authored-by: morluto <morluto@users.noreply.github.com>

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: morluto <morluto@users.noreply.github.com>
2026-10-21 03:46:14 +08:00