Expose mknodat/mkfifoat on Apple targets

[mattsu2020]

Summary

• Expose rustix::fs::{mknodat, mkfifoat} on Apple targets, with support starting in macOS 13.0, iOS 16.0, tvOS 16.0, and watchOS 9.0.
• Weak-link mknodat on Apple and return NOSYS when unavailable on older systems. mkfifoat delegates to the same implementation. Other targets retain direct libc calls.
• Document the supported Apple OS versions and behavior on older systems for both public APIs.

Validation

• cargo doc --no-deps --features all-apis
• git diff --check

The documentation build succeeded on macOS with 10 unresolved-link warnings in unchanged documentation; none were in the updated API documentation.

[xtqqczze]

I think a fallback is needed for macOS ≤ 13 since Apple macOS targets are supported as far back as macOS 10.12, and there are also additional Apple platforms to consider such as tvOS and watchOS.

It would also be good to add references for the newly exposed functions on Apple targets. For example, https://github.com/apple-oss-distributions/xnu/blob/main/bsd/man/man2/mkfifoat.2

[oech3]

Or add non-*at too for macOS?

[xtqqczze]

Right, the POSIX mknod and mkfifo are currently absent from all platforms.

[mattsu2020]

I'll revise this PR to restore weak-linking/runtime detection. When the symbol is unavailable, I'll only fall back to mknod/mkfifo for the AT_FDCWD case, and otherwise return an unsupported/error result. I'll also add Apple man page references for these functions.

[xtqqczze]

On second thoughts, we probably don't need to implement a fallback for the libc backend. We can just return the appropriate Err and let callers handle it.

[badeend]

@mattsu2020 The fix for the CI failure #1620 has been merged. Could you update this PR?

[mattsu2020]

@mattsu2020 The fix for the CI failure #1620 has been merged. Could you update this PR?

I’ve updated it.

[badeend]

@sunfishcode Could you take a look at this? The code change seems relatively straightforward, and it passes CI.

[oech3]

Resolve conflicts

[mattsu2020]

Resolve conflicts

fix it

[xtqqczze]

@mattsu2020 please squash your commits and write a new commit description in your own words

[mattsu2020]

@mattsu2020 please squash your commits and write a new commit description in your own words

I performed a squash merge and added a comment.
Do you think we can merge it?

[xtqqczze]

Can we clarify the platform support for mknodat and mkfifoat in src/fs/at.rs?

The macro-expanded declaration in sys/stat.h shows that it was introduced in macOS 13.0, iOS 16.0, tvOS 16.0, and watchOS 9.0.

Macro-expanded declaration

int mkfifoat(int, const char *, mode_t)
__attribute__((availability(macos,introduced=13.0)))
__attribute__((availability(ios,introduced=16.0)))
__attribute__((availability(tvos,introduced=16.0)))
__attribute__((availability(watchos,introduced=9.0)));

Also, please amend the commit message to summarize the changes rather than describing every implementation detail. For example:

Expose mknodat and mkfifoat on Apple targets.

Weak-link mknodat on Apple and return NOSYS when unavailable on older systems. Keep direct libc calls on other targets.

This will make the PR easier to review when the maintainer gets to it.

[mattsu2020]

Thank you for your feedback. I have incorporated your comments accordingly.

[xtqqczze]

@mattsu2020 Please don't include PR number in commit.

Also, please update the PR description to remove unnecessary details, was that generated by an LLM?

I opened an issue for the rustdoc warnings: #1685

[mattsu2020]

I added it because I thought it was necessary, but I'll remove it.

[mattsu2020]

Should the rustdoc warnings be addressed in this PR?

I'm also fine with handling them in a separate PR.