⚠️ Potential issue | 🟠 Major

Major: Workflow fails if generate script produces no changes.

The workflow detects output via git diff --name-only (line 43) and uploads artifacts with if-no-files-found: error (line 49). If the generated output is already up-to-date (no changes), the job fails, which contradicts the use case where this workflow may be idempotent.

Additionally, line 39 runs the generate script with no explicit error handling—if it fails, the workflow continues to the artifact upload step.

Consider:

1. For idempotent behavior: Use if-no-files-found: warn instead of error to allow the workflow to complete gracefully when no changes are detected.
2. For explicit error handling: Add set -e or explicit checks after the generate step.

Apply this diff:

      - name: Generate (${{ matrix.os }})
        run: python crates/rustpython_doc_db/generate.py

      - name: Detect output file
        id: diff
        run: echo "diff=$(git diff --name-only)" >> $GITHUB_OUTPUT

      - uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4 # v5.0.0
        with:
          name: doc-db-${{ env.PYTHON_VERSION }}-${{ matrix.os }}
          path: ${{ steps.diff.outputs.diff }}
-         if-no-files-found: error
+         if-no-files-found: warn

Based on learnings, the maintainers prefer simplicity and accept workflows that may fail gracefully in edge cases; however, the intent here is unclear.

.github/workflows/update-doc-db.yml lines 38-50: the workflow will fail when the
generate script produces no git-diff (upload-artifact uses if-no-files-found:
error) and also doesn't stop early if the generate script itself fails; change
the upload step to use if-no-files-found: warn to make idempotent runs succeed
with no changes, and make the generate step fail fast by running the script
under a strict shell (e.g., add set -e before running the python command or
check the script exit status and exit non-zero on failure) so the job correctly
surfaces generation errors.

I ain't sure if we want to open a PR, or just push the generated artifacts

we need to update it only when target Python version is changed.

Yeah, so probably not worth the effort in automating a PR