All commit authors signed the Contributor License Agreement.

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

Please open an issue in this tracker.

I see some similar issues elsewhere:

cpython/Include/internal/pycore_freelist.h

Line 88 in 642e5df

PyObject *op = _PyFreeList_PopNoStats(fl);

cpython/Include/internal/pycore_weakref.h

Line 63 in 642e5df

PyObject *obj = FT_ATOMIC_LOAD_PTR(ref->wr_object);

cpython/Include/internal/pycore_weakref.h

Line 90 in 642e5df

PyObject *obj = FT_ATOMIC_LOAD_PTR(ref->wr_object);

Theoretically, we could test for this by including internal headers in test_cppext, but I haven't decided whether I think that's a good idea yet. I'm worried that we're going to be encouraging people to use the private API. Which APIs does Greenlet need?

This:

https://github.com/python-greenlet/greenlet/blob/b24e39c992e1758032bd913bf7c03d239c29480b/src/greenlet/TGreenlet.hpp#L30-L37

Ok. For 3.15, please open an issue requesting that those private APIs be exposed as unstable APIs.

Ok. For 3.15, please open an issue requesting that those private APIs be exposed as unstable APIs.

They don't need to open a new issue. This is a known issue with Cython reporting it as well #130931

I would be happy to merge this (I won't block it). However, we should have some way to make sure we don't regress. pycore_interpframe.h seems to be used by greenlet and Cython which makes it a big enough use case.

I think including pycore_interpframe.h (or pycore_interp_structs.h) in test_cppext would work as a good regression test.

I think including pycore_interpframe.h (or pycore_interp_structs.h) in test_cppext would work as a good regression test.

I wrote test_cppext to test the public C API (headers). Please don't include the internal C API in test_cppext, or add new tests.

Hm, why not? Wouldn't it be easier if we were able to catch regressions immediately instead of Cython or Greenlet complaining to us later?

I mean that I would prefer to leave existing test_cppext unchanged. But you can add new tests on the internal C API if you want.

I wrote #136247 to test the internal C API in test_cext.

So far, I failed to test the internal C API with C++, I get many compiler errors from mimalloc headers and I'm not sure how to handle them.

I think we can ignore mimalloc. Nobody--even Cython--is using that with the internal headers.

I think we can ignore mimalloc. Nobody--even Cython--is using that with the internal headers.

mimalloc is included indirectly by many internal headers such as pycore_interp.h.