bpo-1635741: Enhance _datetime error handling

[koubaa]

This helps prepare for multiphase init (PEP 489)

https://bugs.python.org/issue1635741

[koubaa]

@vstinner please review

[vstinner]

Please mention the modified module name in the PR title.

I suggest to canghe the PR title to "Enhance _datetime error handling" or "Refactor _datetime error handling", since this change is not directly related to multi-phase init. I'm not sure that we can soon convert this module to the multi-phase init because of the PyCapsule C API.

[vstinner]

You wrote the wrong bpo number in the PR title.

[pganssle]

@koubaa Can you provide a TL;DR of why this change is useful or necessary?

[vstinner]

@koubaa Can you provide a TL;DR of why this change is useful or necessary?

https://bugs.python.org/issue1635741 describes the overall goal: better embed Python in application. One side of the issue is to ensure that an extension module releases all its resources at Python exit.

[vstinner]

Be careful with preprocessor traps :-( https://gcc.gnu.org/onlinedocs/cpp/Macro-Pitfalls.html

[koubaa]

Be careful with preprocessor traps :-( https://gcc.gnu.org/onlinedocs/cpp/Macro-Pitfalls.html

Thanks again for sharing this, I easily forget them.

[vstinner]

Thanks again for sharing this, I easily forget them.

Oh, me tee! I got exactly the same issue than you a few days ago :-D

[vstinner]

Thanks, I merged your PR.

I checked manually for refleak using the following test run with "./python -m test (...) -R 3:3":

    def test_leak(self):
        support.run_in_subinterp("import _datetime")

I didn't notice any refleak.

[vstinner]

Hum, I'm not sure that my manual test is relevant, since _datetime doesn't use multiphase init yet. But it leaks, I don't expect issues with Refleaks buildbots.