musl: introduce musl_stat_timespec cfg for struct stat

[heiher]

Description

Some musl targets use out-of-line timespec fields in struct stat. This property is independent of the musl_v1_2_3 switch.

Add a new musl_stat_timespec cfg and use it to control the presence of inline time fields in musl struct stat definitions, avoiding incorrect API assumptions based on musl version alone.

Sources

• https://github.com/kraj/musl/tree/v1.2.5

Checklist

• Relevant tests in libc-test/semver have been updated
• No placeholder or unstable values like *LAST or *MAX are
  included (see #3131)
• Tested locally (cd libc-test && cargo test --target mytarget);
  especially relevant for platforms that may not be checked in CI

[JohnTitor]

The key idea sounds good to me, @tgross35 any thoughts on this? I haven't catched up the time64 change deeply and would like to hear your views.

[tgross35]

Thanks for the fix, sorry that this broke.

Quite honestly I'm feeling like we should just delete the timespec fields and leave st_atime/st_atime_nsec (et. al.) with a comment explaining why we don't exactly match musl source, i.e. more or less revert 55fa65b. Musl allows using the st_atime names via macros https://github.com/kraj/musl/blob/ff441c9ddfefbb94e5881ddd5112b24a944dc36c/include/sys/stat.h#L32C18-L34 so we'd just have a mismatch on the _nsec fields which I think is fine. I'm assuming musl keeps _nsec a bit more hidden since they're not part of posix https://pubs.opengroup.org/onlinepubs/009695299/basedefs/sys/stat.h.html.

That's also nicer for our users who won't need to do a different config for musl and glibc.

Cc @xbjfk for your thoughts as you authored the commit in question.

[xbjfk]

I would personally push for using timespec to:

• Closer match the libc source
• Be POSIX correct and match the correct C usage / docs / man pages.
• Have more compatibility with other architectures (hermit, AIX and hurd are already)
• Make struct definitions easier to read (especially with bi-endian architectures)
• We can remove a lot of test skips.

I personally think the best path forward would be to revert 55fa65b so that people can try the new experimental musl with less breaking changes (and fixing loongarch) and reintroduce the name change without a feature gate for all platforms in the 1.0 branch only for the reasons listed above (virtually every UNIX platform supported by libc uses timespec).

IIRC this was changed with time64 changes the because having conditional definitions for the padding was a bit unwieldy.

Will look into making a PR for this.