gh-111062: Add nogil build test on GitHub Actions

[corona10]

• Issue: Add GitHub Action for checking free-threaded build #111062

[vstinner]

IMO NoGIL is too experimental to hold merging PRs as this point. I would prefer to rely on buildbots for now.

cc @colesbury

[bedevere-app]

When you're done making the requested changes, leave the comment: I have made the requested changes; please review again.

[vstinner]

What's the rationale for adding a pre-commit CI? This PR is related to an issue which is closed.

[corona10]

IMO NoGIL is too experimental to hold merging PRs as this point. I would prefer to rely on buildbots for now.

Well, In my memory, sub interpreter people suffered from the catch-up broken build.

Then what about adding CI that allows failure? (It's possible)

[corona10]

What's the rationale for adding a pre-commit CI? This PR is related to an issue which is closed.

Well, I just think that it will be great if we can check up the issue earlier in the PR round.
For example, we only able to check #110764 in the local environment or after merge status.

IMO NoGIL is too experimental to hold merging PRs as this point

That's why I am just adding checking build status not including test phase.

[vstinner]

Then what about adding CI that allows failure? (It's possible)

When I added a FreeBSD CI, people get annoyed by CI failures even if the CI was non-voting since FreeBSD is only a "Tier-3" buildbot.

[corona10]

When I added a FreeBSD CI, people get annoyed by CI failures even if the CI was non-voting since FreeBSD is only a "Tier-3" buildbot.

Did it allow pass status even if the CI itself failed?
I meant that we can add the CI that only check whether the build is successful or not. And if we also need to allow the broken build for other PR, we can do it.

[vstinner]

Did it allow pass status even if the CI itself failed?

People complained that the whole PR was marked as "failed" when the FreeBSD CI, whereas the FreeBSD CI was not mandatory.

[corona10]

People complained that the whole PR was marked as "failed" when the FreeBSD CI

In Github Action we can ignore the "failed" status as a false positive result. We did the same thing in pyperformance project.
python/pyperformance#274

[colesbury]

I think a GitHub action would be helpful, but it's not critical. If there is opposition to adding it, then it can wait.

[hugovk]

If people don't like seeing an (occasional?) red result, even if non-voting (i.e. doesn't block merge), another option is to append something like || true to the test command to always make it "pass".

The utility of this is debatable, as you have to click through to check if it really passed or failed.

[itamaro]

The utility of this is debatable, as you have to click through to check if it really passed or failed.

I agree with this. I don't see the value of signal that always passes and requires a human to dig in to check for the real result.
in that sense, the manual buildbot triggerring seems preferable to me.

[hugovk]

We're probably best off waiting for the official SC acceptance so we know more about how this should be tested. Hopefully it'll be ready soon.

(And if something like this PR, we should do some refactoring of build.yml, it's getting pretty big. See #110794 (comment).)