Jelle Zijlstra

I can reproduce the nanobind message with your sample script, but the leak apparently no longer happens when I comment out `import pandas`. pandas is a big pile of code,...

I tried to narrow it down by removing parts of pandas's `__init__.py`, and found that the reported leak goes away when the `pandas._libs.json` module is no longer imported. That's a...

Good catch! So the leak chain is `markupsafe.Markup` -> one of its method has an `Optional[str]` annotation -> `Optional[str]` resolves to an instance of `typing._UnionGenericAlias` -> `typing._UnionGenericAlias.__getitem__` holds on to...

Note that the repro no longer works at the moment because nanobind 0.0.8 contains a workaround for this leak. I proposed wjakob/typing_repro#3 to "fix" it. In the meantime, to use...

pathlib.Path.\_\_new__ takes \*\*kwargs, but doesn't do anything with them (https://github.com/python/cpython/blob/master/Lib/pathlib.py#L979). This doesn't appear to be documented. This feature should presumably be either documented or removed (probably removed unless I'm missing...

Thanks, I'll add a PR. This doesn't need to be documented, right?

See #27635 and issues #88919 and #88986. Because of a similar change we ended up with Pablo's venv in the distributed 3.10 tarballs.

You do need a NEWS entry as the good knight Bedevere keeps telling you. Also, avoid force-pushing; it makes review harder. We'll squash-merge anywhere when we merge a PR.

I'd also want to hear @gpshead's opinion, and I haven't looked at #25774 in enough detail to understand the problem it's trying to solve. However, the other PR has been...

You're probably running the wrong version of Black somewhere.