stty: Updating libc crate is blocked: 0.2.178->0.2.179

[oech3]

See comments at #8722 .
This blocks adding CI for cygwin to this repo in the future.
It would block pulling fix for #8474 from upstream.

[tgross35]

Do you know what exactly changed here? Or would it be possible to attach a debugger and see which call is failing?

rust-lang/libc#4694 updated readdir_r, and there are a few defs that would have changed from the 11->12 baseline change https://github.com/search?q=repo%3Arust-lang%2Flibc+%2F%5Cbfreebsd1%5B12%5D%5Cb%2F+path%3A%2F%5Esrc%2F&type=code / rust-lang/libc#2406. None of that jumps out to me as obvious, though. @asomers may have a better idea.

[oech3]

Myself does not have FBSD, but the log https://github.com/uutils/coreutils/suites/53178265891/logs?attempt=1 exists.

@cakebaker When will the log expired? Would you re-run the CI? (@tgross35 is upstream of libc crate)

[tgross35]

Fyi if you ever need a FreeBSD, OpenBSD, Solaris, etc machine or Linux on some less common arches, https://portal.cfarm.net/machines/list/ is a reasonably reliable way to do that. AUIU anyone working on a reasonably large open source project can request an account.

I just checked and cfarm240 (arm64 FreeBSD15) is up and running. Looks like it's cheri-enabled which is unusual, but that might still be fine to reproduce the failure.

[asomers]

As @tgross35 mentioned, the libc upgrade you made included switching from the FreeBSD 11+ ABI to FreeBSD 12+. That's a big difference, because it incluces 64-bit inodes. It also increases the size of struct statfs. That's a syscall that your df is probably using. I suggest looking there.

[oech3]

coreutils/src/uucore/src/lib/features/fsext.rs

Lines 449 to 454 in 81ee8d1

	// Rust on FreeBSD uses 11.x ABI for filesystem metadata syscalls.
	// Call the right version of the symbol for getmntinfo() result to
	// match libc StatFS layout.
	#[cfg(target_os = "freebsd")]
	#[link_name = "getmntinfo@FBSD_1.0"]
	fn get_mount_info(mount_buffer_p: *mut *mut StatFs, flags: c_int) -> c_int;

I experimentary changed it and running CI.

[oech3]

Any non-backword compat change about time for musl-i686 target at 0.2.178->0.2.179? I saw many undefined at

coreutils/src/uucore/src/lib/features/safe_traversal.rs

Lines 471 to 535 in 8cf88cd

	fn atime(&self) -> i64 {
	#[cfg(target_pointer_width = "32")]
	{
	self.stat.st_atime.into()
	}
	#[cfg(not(target_pointer_width = "32"))]
	{
	self.stat.st_atime
	}
	}
	fn atime_nsec(&self) -> i64 {
	#[cfg(target_pointer_width = "32")]
	{
	self.stat.st_atime_nsec.into()
	}
	#[cfg(not(target_pointer_width = "32"))]
	{
	self.stat.st_atime_nsec
	}
	}
	fn mtime(&self) -> i64 {
	#[cfg(target_pointer_width = "32")]
	{
	self.stat.st_mtime.into()
	}
	#[cfg(not(target_pointer_width = "32"))]
	{
	self.stat.st_mtime
	}
	}
	fn mtime_nsec(&self) -> i64 {
	#[cfg(target_pointer_width = "32")]
	{
	self.stat.st_mtime_nsec.into()
	}
	#[cfg(not(target_pointer_width = "32"))]
	{
	self.stat.st_mtime_nsec
	}
	}
	fn ctime(&self) -> i64 {
	#[cfg(target_pointer_width = "32")]
	{
	self.stat.st_ctime.into()
	}
	#[cfg(not(target_pointer_width = "32"))]
	{
	self.stat.st_ctime
	}
	}
	fn ctime_nsec(&self) -> i64 {
	#[cfg(target_pointer_width = "32")]
	{
	self.stat.st_ctime_nsec.into()
	}
	#[cfg(not(target_pointer_width = "32"))]
	{
	self.stat.st_ctime_nsec
	}
	}

[asomers]

coreutils/src/uucore/src/lib/features/fsext.rs

Lines 449 to 454 in 81ee8d1
// Rust on FreeBSD uses 11.x ABI for filesystem metadata syscalls.
// Call the right version of the symbol for getmntinfo() result to
// match libc StatFS layout.
#[cfg(target_os = "freebsd")]
#[link_name = "getmntinfo@FBSD_1.0"]
fn get_mount_info(mount_buffer_p: *mut *mut StatFs, flags: c_int) -> c_int;

I experimentary changed it and running CI.

If you're using the FreeBSD 12 version of struct statfs from libc, but using getmntinfo@FBSD_1.0, then that would definitely cause a problem. Why does coreutils define its own FFI binding for getmntinfo instead of using the one from libc?

[oech3]

I don't know. It was added by different people. libc was bumped to 0.2.178 anyway by removing it.
But I can't bump to 0.2.179 by i686-musl issue.

[oech3]

RUST_LIBC_UNSTABLE_MUSL_V1_2_3=1
RUST_LIBC_UNSTABLE_GNU_TIME_BITS=64

Is this something able to add to .toml instead of environment val?

https://github.com/uutils/coreutils/blob/main/.cargo/config.toml This file seems unrecommended.