better build locks for bootstrap

[onur-ozkan]

This PR improves the way bootstrap handles the build locks by supporting Solaris and removing many dependencies coming with fd-lock.

cc @psumbera

[rustbot]

r? @Mark-Simulacrum

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

[psumbera]

This PR improves the way bootstrap handles the build locks by supporting Solaris and removing many dependencies coming with fd-lock.

Thank you for doing this! I can confirm it builds on Solaris.

[Mark-Simulacrum]

fs2 and fd-lock look basically the same to me dependency wise. (fd-lock actually looks more modern, e.g., windows-sys vs. winapi). fd-lock is also actively maintained, while fs2 isn't (last release 6 years ago).

Can we stick with fd-lock rather than fs2 and submit a PR for fd-lock support for Solaris? That seems like the better solution here, rather than moving to a mostly unmaintained crate.

[bors]

☔ The latest upstream changes (presumably #118069) made this pull request unmergeable. Please resolve the merge conflicts.

[psumbera]

Can we stick with fd-lock rather than fs2 and submit a PR for fd-lock support for Solaris? That seems like the better solution here, rather than moving to a mostly unmaintained crate.

fd-lock support for Solaris is problematic. Solaris doesn't support flock syscall. And emulation via fcntl isn't easy where there is different semantic (flock works per file descriptor and fcntl per process).

[onur-ozkan]

fd-lock support for Solaris is problematic. Solaris doesn't support flock syscall. And emulation via fcntl isn't easy where there is different semantic (flock works per file descriptor and fcntl per process).

I was planning to apply this workaround https://github.com/danburkert/fs2-rs/blob/9a340454a8292df025de368fc4b310bb736f382f/src/unix.rs#L56-L92 on fd-lock. It shouldn't be problem I think as they are sharing almost the same implementation.

[onur-ozkan]

@psumbera Can you test this https://github.com/onur-ozkan/rust/tree/demo-for-build-locks branch (to see if locking works as expected or not on solaris) by any chance?

[psumbera]

@psumbera Can you test this https://github.com/onur-ozkan/rust/tree/demo-for-build-locks branch (to see if locking works as expected or not on solaris) by any chance?

It builds just fine. But still prints:

WARNING: file locking not supported for target, not locking build directory

I think it's now printed always:

onur-ozkan@0c2b090#diff-5624b915604aca92c82fe2af07e555adfd958a3247eada8300061e6fe6c3dd6eR55

[onur-ozkan]

@psumbera Can you test this https://github.com/onur-ozkan/rust/tree/demo-for-build-locks branch (to see if locking works as expected or not on solaris) by any chance?

It builds just fine. But still prints:

WARNING: file locking not supported for target, not locking build directory

I think it's now printed always:

onur-ozkan@0c2b090#diff-5624b915604aca92c82fe2af07e555adfd958a3247eada8300061e6fe6c3dd6eR55

Oh, I forgot to remove that log (not really important anyway, this is just for debugging). Does the locking work? What happens when you run x concurrently?

[psumbera]

Oh, I forgot to remove that log (not really important anyway, this is just for debugging). Does the locking work? What happens when you run x concurrently?

Yes, seems to work:

WARNING: build directory locked by process 2647, waiting for lock

[onur-ozkan]

Created #119413 PR and closing this.