Add pthread_self

[king6cong]

No description provided.

[Susurrus]

Thanks, @king6cong!

However I don't believe this is the right approach to this. If we can't support gettid directly on Mac, I don't believe we should emulate it for the sole reason of abstracting away platform differences. Nix only attempts to be a very small shim layer over unsafe libc interfaces. Any higher-level abstractions like this should be left to downstream crates. So instead of doing this, can you implement pthread_self() for all suitable platforms. Since that works on Linux and Mac someone could use that instead for cross-platform support.

[king6cong]

So instead of doing this, can you implement pthread_self() for all suitable platforms. Since that works on Linux and Mac someone could use that instead for cross-platform support.

To confirm , you mean that we should keep pthread_self out of nix or take it in in another form? 😃

[Susurrus]

I think we should provide a safe wrapper about pthread_self instead of pretending that pthread_self is gettid on Mac. It should be also noted that according to the man page:

The thread ID returned by pthread_self() is not the same thing as the kernel thread ID returned by a call to gettid(2)

Right now there's a bunch of pthread stuff is in src/sys/signal, but I'm not sure if this should be put there. I think putting it in a new src/sys/pthread.rs file is the correct way to do this. There's also previous conversation about this here.

[king6cong]

@Susurrus updated

[Susurrus]

That looks like a step in the right direction, but tests are failing.

[Susurrus]

@king6cong Can you also add an entry to the CHANGELOG for this?

[king6cong]

@Susurrus Yeah, haven't found the reason that tests fails, maybe tweak around ci setup to show failure details first.

[Susurrus]

Do the tests pass locally for you? If you have access to a mac and can test on one too, that might point us in the right direction.

[king6cong]

@Susurrus Yeah, it occurred to me that I could use my x86 mac to test i686 failure. Updated, Thanks!

[Susurrus]

A few minor issues, mostly the unix thing needs to get sorted out.

[Susurrus]

This should show nix::sys::pthread::pthread_self instead.

[Susurrus]

Aren't all the platforms covered by nix unix? This seems unnecessary.

[Susurrus]

Again, aren't all nix-supported platforms unix?

[Susurrus]

Please add an empty line below this one.

[Susurrus]

Delete one of these blank lines.

[Susurrus]

@king6cong I can't tell because the old code is gone, but what was the cause of the failures and what did you change to fix them?

[king6cong]

@Susurrus Updated. libc::pthread_self() should return pthread_t instead of pid_t, that's the reason that tests on several platforms failed

[Susurrus]

Great, thanks for clearing that up. Makes sense why it fixed things.

I wanted to ask, do you have a specific use case for this that this addition is all that's required for it to work? I'm going to block this on #590, so we can be sure that all platforms are passing before adding this, so we have a little time here in case there was any missing functionality you also needed. and wanted to add from pthread.h.

[Susurrus]

Can you make a new type Pthread, export it from nix, and also use that as the return value for pthread_self? I'd like nix to have all of its datatypes self-contained for better ergonomics for users.

[king6cong]

Updated

[Susurrus]

This looks good. We're going to wait until the new FreeBSD buildbot is integrated (#618) and this will be our first test.

[asomers]

Slight change of plans. We're no longer waiting for #618 .
bors r+

[bors]

Build succeeded

• continuous-integration/travis-ci/push

[Susurrus]

@King6Kong Thanks for your contribution!