[amjjbonvin]

Before submitting, read the contributing guidelines and the AI policy.

What does this PR do and why?

The step added in #1749, which compares the PR's code between pushes, runs with GitHub's default bash -e, without pipefail. A failing git merge-base or git diff inside patch_id was hidden by the pipe: both patch ids were empty, hence equal, and the tests were skipped whenever they had passed on the previous head, although the code may have changed.

• set -o pipefail in the script.
• Each patch id is computed in a checked command: any failure to fetch or compare the commits runs the tests.

How was this tested?

The step's script, extracted from the workflow, run locally with bash -e (as on GitHub) on commits of #1745:

• code changed but base commit unresolvable: the current version skips the tests (skip=true), the fixed one runs them (skip=false);
• merge of main after a fully tested push: still skipped;
• code change: tested;
• unknown commit: tested.

AI assistance

Claude Code found the problem while porting #1749 to disvis and powerfit, wrote the fix and checked it as described above. I reviewed the changes.

Checklist

• Tests cover the new and/or changed code (CI only; tested locally as described)
• Documentation updated if needed (also in the haddock3 user-manual) (not needed)
• CHANGELOG.md updated for user-facing changes (not user-facing)

Related issues

Closes #1750

Notes for reviewers

Same fix as in haddocking/disvis#38 and haddocking/powerfit#155.

🤖 Generated with Claude Code