build(ci): ensure C++ toolchain before pytest

[wanghan-iapcm]

Summary

The Test Python job (.github/workflows/test_python.yml) does checkout → setup-python → pip install → pytest and never verifies the C++ compiler, yet torch inductor shells out to the system g++ to JIT-compile C++ kernels at test time (e.g. the _CompiledModel / torch.compile tests).

When a GitHub-hosted runner VM occasionally boots with a broken gcc — the g++ driver present but its backend cc1plus missing — this surfaces as an opaque failure deep inside the suite:

torch._inductor.exc.InductorError: CppCompileError: C++ compile error
g++: fatal error: cannot execute 'cc1plus': execvp: No such file or directory

This was observed once on an unrelated docs-only PR. The re-run on a fresh VM with identical code passed, confirming it is a transient per-VM toolchain corruption, not a code defect — and a survey of other recent Test Python failures showed all of them were genuine code/test failures, none with this cc1plus signature.

Change

Add a step right after setup-python that:

• reinstalls build-essential (re-laying down gcc, which ships cc1plus), self-healing a corrupted VM, and
• verifies a trivial C++ compile, so if the toolchain is still broken the job fails immediately with a clear message instead of 40 minutes into pytest.

Notes / known limitations

• This makes the flake self-correct without a manual job re-run. It does not eliminate the underlying hosted-runner image variability; pinning to a container image would be the fully-deterministic alternative but is a larger change.
• pytest-rerunfailures would not help here: retrying the same test on the same VM hits the same missing cc1plus.

Summary by CodeRabbit

• Chores
  • Updated internal CI/CD workflow to enhance build validation during automated testing.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ed5b92ddc6

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Reinstall the compiler package, not the metapackage

For the runner corruption described here (the g++ driver exists but cc1plus is missing), this step does not actually restore the missing backend: build-essential is only a small informational metapackage whose dependency list includes g++, and an apt-get -s install --reinstall build-essential dry run reinstalls exactly one package, build-essential, not its already-installed dependencies. As a result, a VM with a damaged g++/g++-11 package will still fail the compile check instead of self-healing; reinstall g++/the versioned compiler package directly.

Useful? React with 👍 / 👎.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 68815877-1ac9-411b-aea9-72397e0e288c

📥 Commits

Reviewing files that changed from the base of the PR and between 03682bf and ed5b92d.

📒 Files selected for processing (1)

• .github/workflows/test_python.yml

---

📝 Walkthrough

Walkthrough

The Python CI workflow gains a new step in the testpython job that reinstalls build-essential via apt-get and compiles a minimal C++ program with g++ to verify that cc1plus is present before the torch inductor tests execute.

Changes

Python CI C++ Toolchain Verification

Layer / File(s)	Summary
g++ compile verification step .github/workflows/test_python.yml	Adds a step that reinstalls build-essential and compiles a trivial C++ program with g++, failing the job early if cc1plus is absent.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'build(ci): ensure C++ toolchain before pytest' directly and specifically describes the main change—adding a CI step to verify the C++ toolchain before running pytest tests.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.15%. Comparing base (03682bf) to head (ed5b92d).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #5573   +/-   ##
=======================================
  Coverage   82.14%   82.15%           
=======================================
  Files         900      900           
  Lines      104139   104138    -1     
  Branches     4471     4470    -1     
=======================================
  Hits        85550    85550           
  Misses      17178    17178           
+ Partials     1411     1410    -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[wanghan-iapcm]

Closing this — the diagnosis is wrong, as @njzjz pointed out: if the gcc toolchain were broken at boot, the editable pip install -e . (which compiles deepmd's PyTorch C++ ops) would have failed first.

The failed-run log confirms it. On the affected VM (run 27929887537, job Test Python (10, 3.13)):

• 04:36:50 -> 04:38:16: deepmd-kit's C++ built successfully (~86s, needs cc1plus)
• 04:38:47: dp --version works -> install fully succeeded
• 04:38:49: pytest starts
• ~05:32:16 (~53 min later): torch inductor's g++ ... main.cpp fails with cannot execute 'cc1plus'

So the toolchain was healthy at install and only failed deep into the test run. A reinstall-at-start (this PR) runs while the toolchain is still fine and does nothing for a mid-run degradation, so it would not reliably prevent this. Only 1 of 24 shards failed and a re-run on a fresh VM passed, i.e. a transient single-VM event. Closing as misdiagnosed; investigating the real mid-run cause separately.