Fixed vita libc definitions

[nikarh]

This fixes definitions that were incorrect in my initial PR (#3209). As with the previous one, this relies on open-source newlib implementation and doesn't use Sony internal API's or expose Sony internal data structures.

This fixes:

• fs API that uses dirent and stat
• sockets
• pthreads

A couple of explanations

• Unfortunately dirent in vita newlib is not very POSIX'y, before d_name it has a field with an internal data structure, which is of no use for std (no inodes, no file type inside), so I've just stubbed it as an __offset. Also, Vita doesn't expose inodes - dirent doesn't have it at all, and in stat it is always 0. I am not sure what would be the best approach here. Maybe I should move the POSIX dirent to generic.rs and reexpose it in all targets, and redefine it in vita/mod.rs?
• All pthread types on Vita are pointers, and the backing data structure is allocated by corresponding init functions, so their sizeof's are all 4. I also changed pthread_attr_t and pthread_rwlockattr_t to reflect their sizes and not be constant. May be in relation to [strict provenance] Fix System APIs That Are Liars rust#95496 it would be better to move existing thread definitions to generic, and for vita specifically make them pointer types instead of byte arrays.

The fixes in std will be in rust-lang/rust#111819

[rustbot]

r? @JohnTitor

(rustbot has picked a reviewer for you, use r? to override)

[JohnTitor]

Thanks! @bors r+

[bors]

📌 Commit 5f569dc has been approved by JohnTitor

It is now in the queue for this repository.

[bors]

⌛ Testing commit 5f569dc with merge 0d78fe1...

[bors]

☀️ Test successful - checks-actions, checks-cirrus-freebsd-12, checks-cirrus-freebsd-13, checks-cirrus-freebsd-14
Approved by: JohnTitor
Pushing 0d78fe1 to main...

[nikarh]

Hey @riking, thanks for reviewing this! In vita newlib dirent is weird and contains a 88 byte sized internal data structure at offset 0 (which we neither really want to expose nor care about here). And all other fields defined before this offset are filtered by a macro for vita. Plus this structure should be defined in s! macro which adds repr(C).

…

On Tue, Jun 20, 2023, 21:39 Kane York ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In src/unix/newlib/mod.rs <#3255 (comment)>: > pub d_type: ::c_uchar, + #[cfg(target_os = "vita")] + __offset: [u8; 88], This is a repr(Rust) struct, so this offset doesn't actually offset d_name. In fact, the compiler will preferentially place d_name at offset 0 because it has a higher natural alignment

[riking]

Yes, I saw the s! macro after writing that and deleted my comment.