[3.13] gh-123880: Allow recursive import of single-phase-init modules (GH-123950)

[miss-islington]

(cherry picked from commit aee219f)

Co-authored-by: Petr Viktorin encukou@gmail.com
Co-authored-by: Shantanu 12621235+hauntsaninja@users.noreply.github.com
Co-authored-by: Brett Cannon brett@python.org

• Issue: Python 3.13 breaks circular imports during single phase init of extension module #123880

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @encukou for commit 58ba5ba 🤖

If you want to schedule another build, you need to add the 🔨 test-with-buildbots label again.

[Yhg1s]

FYI, pushing this to 3.13.1.

[hauntsaninja]

@Yhg1s good to know, is that at all open for discussion? This means that most mypyc compiled projects (like mypy and black and charset_normalizer) will be broken on 3.13.0 when they ship 3.13 mypyc-compiled wheels.

[Yhg1s]

Sure, it's open for discussion. Given that this showed up all the way at the tail end of the release candidate phase, I'm assuming it must not be that pressing (or it would've shown up before :). It sounds like it is pressing, and the actual change is not that complicated (although slightly more complicated than I would've liked between the release candidate and the final release).

Why didn't this show up before? How are we going to make similar issues show up earlier in the future?

[hauntsaninja]

It didn't show up before primarily because there were other issues running mypyc with 3.13 that showed up before this one. It looks like there was also a period of 3 weeks where it was known but not acted on (following #121489 (comment) ) due to a combination of classic open source volunteer time and communication issues + also I think there was some misplaced suspicion that the issue was something specific to mypyc's test suite.

Timeline on mypy's side is something like:

• beta1 I do some initial testing and fixes in beta1 on pure Python (non-mypyc) side
• beta2 first mypyc 3.13 fixes start landing
• beta3 mypy starts running against 3.13 in CI (allowing the failure), lots of fixes land, various issues get reported to CPython e.g. Mimalloc headers are not being installed #121467 "template with C linkage" error when including "pycore_interp.h" from C++ file #122584 [C API] Private functions used by mypyc #121489
• beta4 not much activity
• rc1 more mypyc fixes are landed
• rc2 I start looking into remaining issue, but am late by a day

3.12 was easier, I think we had things green before first release candidate.

Some things that could help:

• I think if Github offered a true nightly build of Python that would help. There was probably a bit of "oh well we're waiting for upstream"
• mypy could probably have started testing during alphas
• I'd planned to make sure things were greenable in rc1, but had some health issues that prevented me from doing so

[Yhg1s]

FWIW, we discussed this at the sprint as well, and it's clear that this should go into 3.13.0. We should try not to break mypyc (and similar users) quite so much, to make it easier for mypyc to find these issues for us in the future :)