fix: under pypy, when no dict args given, kwds is non-null

[27rabbitlt]

Not clearly whether it's bug on numexpr side or pypy side, but when using pypy, sometimes function Numexpr_Run(PyObject *self, PyObject *args, PyObject *kwds) gets no dictionary arguments but pointer kwds is still non-null.

So add a further check of the size of dictionary kwds. It won't affect the performance since it's only one-time check.

This should fix issue #463, although not all errors could be eliminated for pypy because sys module in pypy seems not to have attribute getrefcount which is used when testing.

[mattip]

It was an incompatibility in PyPy, I fixed it in this commit

[27rabbitlt]

It's fixed by https://foss.heptapod.net/pypy/pypy/-/commit/402204e196ac7a86975d8d38f6abf2c384f1dd9d, should be closed now.

[mgorny]

Well, I've noted in #463, PyPy wasn't fixed after all. Perhaps we could merge this for the time being? This fixes all of ex_uses_vml test failures, and it would improve compatibility with prior PyPy versions anyway.

[27rabbitlt]

Agree. I will reopen this or.

…

On Fri, Jan 5, 2024 at 20:13 Michał Górny ***@***.***> wrote: Well, I've noted in #463 <#463>, PyPy wasn't fixed after all. Perhaps we could merge this for the time being? This fixes all of ex_uses_vml test failures, and it would improve compatibility with prior PyPy versions anyway. — Reply to this email directly, view it on GitHub <#467 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A33BDH3GACGKQ5WGUU325C3YM7U7HAVCNFSM6AAAAABAXZVPUSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNZYGU3TGMJRGY> . You are receiving this because you modified the open/close state.Message ID: ***@***.***>

[mattip]

@27rabbitlt did you mean to reopen this?

[27rabbitlt]

@27rabbitlt did you mean to reopen this?

Yes, but I spent some time on this and still didn't catch up the latest situation. Does this problem still affects pypy3.10 user? Should we fix it on numexpr side?

[mgorny]

Yes, it still happens with the latest PyPy version and FWICS we haven't really figured out how to fix it on PyPy end. So yes, a workaround on numexpr side would be much appreciated.

[mgorny]

Could you merge it? I can confirm it fixes the issue for me. I'm going to file a PR for the other problem (sys.getrefcount()).

[FrancescAlted]

Thanks @27rabbitlt, @mattip and @mgorny