fix segmentation fault when datetime module is invalid

[davidhewitt]

Reported downstream in pydantic/pydantic-core#1171

We aren't checking if PyDatetime_IMPORT() succeeds, which can lead to dereference of a null pointer and a crash if the import fails.

This PR fixes that crash by adding error checking. Where possible I return the error, otherwise I just panic.

[codspeed-hq]

CodSpeed Performance Report

Merging #3818 will degrade performances by 16.95%

_{Comparing davidhewitt:datetime-segv (5b11041) with main (fa53d81)}

Summary

⚡ 2 improvements
❌ 3 regressions
✅ 74 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	main	davidhewitt:datetime-segv	Change
❌	mapping_from_dict	300 ns	355.6 ns	-15.63%
❌	list_via_downcast	157.2 ns	185 ns	-15.02%
❌	sequence_from_list	272.2 ns	327.8 ns	-16.95%
⚡	extract_float_extract_success	472.8 ns	417.2 ns	+13.32%
⚡	extract_float_downcast_success	481.1 ns	425.6 ns	+13.05%

[adamreichold]

Two nits. Otherwise LGTM.

I guess an invalid module on the path is weird enough that this does not need an immediate backport and 0.20.x maintenance release? This module did change significantly since 0.20.x IIRC.

[davidhewitt]

Thanks for the review 👍

I guess an invalid module on the path is weird enough that this does not need an immediate backport and 0.20.x maintenance release? This module did change significantly since 0.20.x IIRC.

Yes, I think it is a weird edge case. I think it's also very likely that this will always just crash immediately without doing any harm, so it's just ugly UX rather than dangerous.

That said, I'm about to prepare a patch release for #3619, so if it's trivial to cherry-pick this I will.