Skip to content

fix(ci): repair checks broken by #2374 - #2377

Merged
georgeh0 merged 1 commit into
mainfrom
claude/fix-pr-2374-ci-errors-5dc3a1
Sep 5, 2026
Merged

fix(ci): repair checks broken by #2374#2377
georgeh0 merged 1 commit into
mainfrom
claude/fix-pr-2374-ci-errors-5dc3a1

Conversation

@georgeh0

@georgeh0 georgeh0 commented Sep 5, 2026

Copy link
Copy Markdown
Member

Summary

#2374 was merged as a critical fix while its CI was red, leaving main failing on Fast Check, E2E Type Check, and Build & Test. Two real errors, both from that commit:

  • end-of-file-fixer (Fast Check + Build & Test): the new python/tests/connectorkits/__init__.py holds a lone newline. Every sibling test package (python/tests/connectors/, core/, resources/) is a 0-byte file. Truncated to match.
  • mypy (E2E Type Check): python/tests/connectorkits/test_async_adapters.py:32: error: "AsyncIterator[int]" has no attribute "aclose". sync_to_async_iter is an async generator but declares -> AsyncIterator[_T], which hides aclose(). Since the entire point of that fix is that closing the iterator no longer blocks the event loop, the honest, more specific annotation is -> AsyncGenerator[_T, None]. AsyncGenerator is a subtype of AsyncIterator, so the four existing async for call sites are unaffected.

The Windows build-test failure on the PR run was unrelated infrastructure — setup-uv got fetch failed pulling the uv version manifest; the remaining matrix legs were cancelled by fail-fast.

Testing

  • prek run --all-files mypy-check — passes, both against the default dev deps and against the E2E job's uv sync --no-dev --group ci set (the two dependency sets differ; verified separately).
  • PREK_SKIP=<fast-check skip list> prek run --all-files — all hooks pass, fix end of files included.
  • uv run pytest python/tests/connectorkits/ -q — 1 passed.

Note, out of scope

uv.lock is stale relative to pyproject.toml: #2359 added pymysql and aiohttp to the ci-enabled-optional-deps group without relocking, and the uv-lock hook is file-filtered so it never fired. Any local uv run rewrites the lock. Harmless in CI today (the sync steps are not --locked), and deliberately left out of this PR to keep the diff scoped to the fallout.

🤖 Generated with Claude Code

#2374 landed as a critical fix while its CI was red. Two checks fail on
main because of it:

- `end-of-file-fixer` (Fast Check, Build & Test): the new
  `python/tests/connectorkits/__init__.py` holds a lone newline; the
  sibling test packages are all empty files. Truncate it.
- `mypy` (E2E Type Check): the regression test calls
  `async_iter.aclose()`, but `sync_to_async_iter` declares
  `-> AsyncIterator[_T]`, which has no `aclose`. It is an async
  generator, and its whole point post-#2374 is that closing it does not
  block the event loop, so declare `-> AsyncGenerator[_T, None]` — the
  honest, more specific type. `AsyncGenerator` is a subtype of
  `AsyncIterator`, so no caller changes.

The Windows build-test failure on the PR run was a `setup-uv` network
flake, not a code problem.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@georgeh0
georgeh0 merged commit 0164016 into main Sep 5, 2026
18 checks passed
@georgeh0
georgeh0 deleted the claude/fix-pr-2374-ci-errors-5dc3a1 branch September 5, 2026 17:28
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant