ci: set git committer identity before merging the target branch

[yingjerkao]

Summary

The shared "Merge with latest target branch (pull_request only)" step runs

git merge --no-edit origin/${{ github.event.pull_request.base.ref }}

without a configured git committer identity. Whenever a PR head is behind its base branch, the merge is not a fast-forward no-op, so git must create a merge commit — and fails with:

fatal: empty ident name (for <runner@...>) not allowed

killing the job with exit code 128 before any real work runs (observed 2026-07-06 on PR #984, run 28790515780). PRs branched from the base tip pass only because the merge reports "Already up to date", which masks the bug.

Fix

Configure the github-actions[bot] identity before the merge, exactly as release_pypi.yml already does:

git config user.name "github-actions[bot]"
git config user.email "github-actions[bot]@users.noreply.github.com"

Applied to the four remaining workflows that carry the step:

• .github/workflows/ci-cmake_tests.yml
• .github/workflows/ci-downstream-find-package.yml
• .github/workflows/clang-format-check.yml
• .github/workflows/conda_build.yml

The step is otherwise unchanged. (release_pypi.yml already had the fix, so it needs no change.)

Test plan

• yaml.safe_load passes on all five workflow files
• Verified every git merge in .github/workflows/ is now preceded by the identity config
• CI on this PR exercises the merge step itself (this branch is at the master tip, so the merge will be a no-op here; the real proof is the next behind-base PR no longer failing with exit 128)

🤖 Generated with Claude Code

[gemini-code-assist]

Note

Gemini is unable to generate a review for this pull request due to the file types involved not being currently supported.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 30.82%. Comparing base (7b4d7dc) to head (22812b4).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1007   +/-   ##
=======================================
  Coverage   30.82%   30.82%           
=======================================
  Files         229      229           
  Lines       33348    33348           
  Branches    14069    14069           
=======================================
  Hits        10281    10281           
  Misses      15795    15795           
  Partials     7272     7272           

Flag	Coverage Δ
cpp	30.45% <ø> (ø)
python	61.44% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
C++ backend	32.25% <ø> (ø)
Python bindings	18.77% <ø> (ø)
Python package	61.44% <ø> (ø)

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7b4d7dc...22812b4. Read the comment docs.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[pcchen]

Code Review — LGTM

Correct, minimal CI fix. The two git config lines are placed first inside the "Merge with latest target branch" run: | block, before git fetch/git merge --no-edit — which is exactly right: git merge --no-edit of the target branch creates a non-fast-forward merge commit, and the ephemeral runner has no committer identity by default, so the merge would fail without it. The github-actions[bot] identity is the conventional one, indentation matches each block's existing git commands, and the fix is applied uniformly across all 4 workflows.

Full CI is green here — including the pull_request merge-target step itself, which validates the fix end-to-end.

Note: this supersedes #1002 (@IvanaGyro), which is a functionally identical fix to the same 4 files (only comment wrapping/indentation differs). Linking it there.

Posted by Claude Code on behalf of @pcchen

[pcchen]

Context for whoever merges: this bug is currently masking BuildAndTest on behind-base PRs

On the last (unchecked) test-plan item — #1007's own CI can't exercise the fix because this branch is at the master tip, so its merge step is a no-op. That's accurate. But the fix isn't speculative, and the "real proof" already exists:

• Master's ci-cmake_tests.yml (the BuildAndTest job) still lacks the git config, so any PR behind its base dies at the "Merge with latest target branch" step with fatal: empty ident name … not allowed / exit 128 — before BuildAndTest runs. (Direct evidence: the cited PR fix(linalg): replace lower-enum-index dtype selection with Type.type_promote #984 run 28790515780 fails exactly this way on Formatting Check.)
• This appears to be the cause of the missing/absent BuildAndTest checks on the currently-behind-base PRs in flight (e.g. refactor(pybind)!: guard UniTensor +/- on matching metadata; drop * / (#934) #997, feat(api)!: norm()->Scalar, underscore convention for mutators, unified permute/reshape syntax #1000, refactor(Bond)!: make Bond immutable; encapsulate BlockFermionicUniTensor signflip #1001, fix(UniTensor): thread non-mutating block views through contract() #1005): their wheel jobs run (the release_pypi/BuildWheel workflow already has the identity fix) but their BuildAndTest job does not — matching this failure exactly. On refactor(Bond)!: make Bond immutable; encapsulate BlockFermionicUniTensor signflip #1001, no BuildAndTest check is present at all.

So merging #1007 is higher-value than it looks: it should restore BuildAndTest CI coverage for every behind-base PR, which is what would let the recurring "test evidence is local-only / BuildAndTest didn't run" notes on those PRs finally be cleared by real CI.

To close the last test-plan item after merge: re-run BuildAndTest on any behind-base open PR (e.g. #1001) and confirm it now gets past the merge step.

Posted by Claude Code on behalf of @pcchen

[pcchen]

Approving. Correct, minimal, fully-green CI fix: the github-actions[bot] identity is set before the target-branch merge in all four affected workflows, mirroring the working release_pypi.yml precedent. The bug is real and evidenced (PR #984 run 28790515780: empty ident name / exit 128), and merging this restores BuildAndTest coverage for the behind-base PRs currently in flight.

Posted by Claude Code on behalf of @pcchen