Ugh the build wheels jobs have been failing again starting today, across multiple PRs. E.g. macOS 10.15 intel saying that Pantsd is timing out while initializing https://github.com/pantsbuild/pants/actions/runs/3267360934/jobs/5372439004

what is the reason for even running the build wheels job for a PR ?

usually it's supposed to be skipped for PRs. Several times we've been bitten though by breaking builds on main from PRs changing things. For example, just today Josh's new PR to add a new backend would have broken main because we didn't update

bin/BUILD

- and CI in the PR caught that

Will building for a single platform instead of 4 would have caught that ? Also, maybe add some more checks (linter, a script that will check for stuff) in order to prevent those regressions w/o needing to run those time & cpu consuming jobs.

yeah agreed it's overkill

the way I think about this: If there is a regression (essentially a bug, even if it just breaks some CI process, it is still a bug) then part of fixing it should be adding a test so it doesn't happen again. some times it is a unit tests code and sometimes it is a script.

As of my recent CI changes, we should only be building wheels in PRs that touch these files: https://github.com/pantsbuild/pants/blob/20ac158e6dc50821fb465a0d3f67f96c8079a19a/build-support/bin/classify_changed_files.py#L25