Unpin tox

[hynek]

It works locally for me.

[hynek]

no it doesn't

[webknjaz]

Wait, if it works locally but not in GHA, maybe it's the plugin's fault? I think I've seen a corresponding plugin for tox4 being separate from the one for tox3.

[hynek]

no, I ran tox -e mypy locally but the broken thing is the test suite. I've isolated & opened a bug: tox-dev/tox#2904

[webknjaz]

Ah, I see. But FTR there's https://github.com/tox-dev/tox-gh too.

[hynek]

What does that to do with anything? Our problem is that the package pins for Mypy are from PyPI instead of from the local package. I don't see how the plugins play into that.

[webknjaz]

It's unrelated to the mypy problem but is related to the tox4 migration, IIUC.

[hynek]

It's unrelated to the mypy problem but is related to the tox4 migration, IIUC.

According to https://pypi.org/project/tox-gh-actions/, tox-gh-actions>=3 works with tox 3 just fine.

I would switch if there's any upsides, but I can't see any rn?

---

Anyways, tox has fixed the bug 🎉 and we're waiting for a release now.

[webknjaz]

According to https://pypi.org/project/tox-gh-actions/, tox-gh-actions>=3 works with tox 3 just fine.

Looks like it's ~= 2.0 though.

[webknjaz]

FWIW here's some notes regarding the differences tox-dev/tox-gh#7 (comment)

[hynek]

what the flying fuck is going on now… it works perfectly fine on my machine!?

[webknjaz]

Looks related to the recent Hatch-scm integration. At least one of the logs failed on int('dev2').
Also, I saw a resource warning so there might be an unclosed file descriptor somewhere.

[hynek]

but but but… it used to work! otherwise i wouldn’t have merged it! :(

I’m currently in the process of setting up a new computer, once I’m fully functional I’ll look closer sigh

[webknjaz]

FWIW I checked this branch locally and it does work. Maybe somebody needs to stick https://github.com/marketplace/actions/debugging-with-tmate there and inspect the worker over SSH.

Oh, and I just remembered the most typical setuptools-scm mistake — the contributors usually have a full Git clone locally but CIs tend to do shallow clones. The fix is usually to set the fetch depth to 0 or run git unshallow.

Though, maybe this is not the only problem here, I haven't had time to dig deeper.

[hynek]

0c9b54c #genius

[hynek]

FTR on tox-gh: it seems like it has a very different way of working than tax-gh-actions. I tries using it with structlog and 3.11 = py311 means literally "only run py311" while tox-gh-actions essentially says -f py311. For attrs it doesn't matter for now, but that's a very significant different:

(using screenshot b/c logs get lost over time)