Fix some raw types in std to match what's in libc

[dhduvall]

The pthread_t type, in particular, causes failures when building cargo 0.20.0 after rust-lang/libc@8304e06b5.

[cuviper]

Most of those raw types are deliberately wrong per #31551.

I think you can still make a case for pthread_t on its own though.

[binarycrusader]

@cuviper can you expound a bit on the "deliberately wrong" part? There's not enough reasoning in #31551 to explain why. Should rust itself no longer be using these at all? pthread_t is certainly used at the least which is causing correctness issues. The combination of them being simultaneously deprecated and yet still in use is confusing.

[sfackler]

Those type definitions are all stable and can't be changed.

[binarycrusader]

They were never correct to begin with and this only affects Solaris. Surely they can be fixed since they're causing build problems for cargo.

[malbarbo]

@binarycrusader see the comments in #41752.

[dhduvall]

This is the error that prompted this fix:

error[E0308]: mismatched types
   --> src/lib.rs:587:40
    |
587 |                     libc::pthread_kill(self.thread.as_pthread_t(), libc::SIGUSR1);
    |                                        ^^^^^^^^^^^^^^^^^^^^^^^^^^ expected u32, found usize
    |
    = help: here are some functions which might fulfill your needs:
            - .count_ones()
            - .count_zeros()
            - .leading_zeros()
            - .trailing_zeros()

error: aborting due to previous error

error: Could not compile `jobserver`.

As best I can tell, there's no issue in the jobserver code itself. The problem is a mismatch between libc (pthread_kill(), which takes libc's notion of pthread_t, which is c_uint/u32) and std (as_pthread_t(), which returns a RawPthread, which is std's notion of pthread_t, which is usize). I don't think there's any change to the jobserver code that can fix that error.

I could simply fix pthread_t, as @cuviper suggested, and that's certainly enough to solve that particular problem, but I'd like to know why that would be okay, but fixing the others is not. I'd also be perfectly okay with a solution that involved a fix somewhere else, but I'm not sure what that would be, other than defining RawPthread to be something other than std::os::unix::raw::pthread_t:

#[allow(deprecated)]
use os::unix::raw::pthread_t;

#[allow(deprecated)]
pub type RawPthread = pthread_t;

That use of a type from the raw module seems pretty rare (if not unique?), so maybe the thing to do is simply to fix that, but I'm not sure what knock-on effects that would have, so making changes to Solaris-only code seemed safest, since it's likely that this code simply never worked.

[alexcrichton]

I'd personally considered this OK to land in the sense that solaris is a pretty "low" tier in terms of guarantees right now. My guess is that this'll only fix code rather than break it, so I'd be ok landing.

@sfackler and @cuviper, do y'all disagree though?

[cuviper]

The filesystem-related types are deliberately wrong mainly because LFS is messy. At least on Linux, there's just not a good way to have types that are truly FFI compatible. So the decision was widen all the MetadataExt return values to 64-bit, except where fewer bits will definitely suffice, and to widen the raw types in kind. There's more discussion in RFC 1451 and its pull request.

But I don't feel too strongly about it, and I don't know the LFS story on Solaris.

Fixing pthread_t seems fine though. It did match libc::pthread_t before the commit linked in the OP, and it makes sense to me that std should get the same fix.

[sfackler]

My recollection is that we deprecated these types for exactly this reason - we couldn't change them since they're stable. Is this an exception because Solaris isn't a tier 1 platform?

[cuviper]

@sfackler #31551 did change those types while deprecating them.

[alexcrichton]

Ah yeah it's true, I'd mostly be worried about just fixing pthread_t which isn't deprecated and could be a problem, but I'd be fine if we backed out the other changes due to LFS business (and those types should be practically deprecated regardless)

[alexcrichton]

@dhduvall would you like to change this to just fixing pthread_t? After that I think we're good to go?

[dhduvall]

My git-fu is sad. I ended up opening #43597.