Warning: <type> is not FFI-safe, after recent changes involving `#[non_exhaustive]`

[neuschaefer]

After the recent commit (9ee2d2a) in which I added #[non_exhaustive] to kproc_info on FreeBSD 15, I get warnings like the following when I run libc-test:

warning: `extern` block uses type `libc::kinfo_proc`, which is not FFI-safe
      --> /home/jn/libc/target/debug/build/libc-test-9d66e31a05563bd6/out/ctest_output.rs:132002:62
       |
132002 |             fn ctest_field_ptr__kinfo_proc__ki_structsize(a: *const kinfo_proc) -> *mut u8;
       |                                                              ^^^^^^^^^^^^^^^^^ not FFI-safe
       |
       = note: this struct is non-exhaustive
       = note: `#[warn(improper_ctypes)]` on by default

This seems to point to a larger problem with using #[non_exhaustive] in the libc crate.

cc @tgross35

[tgross35]

I can’t think of why this would be when it’s passed indirectly. The warning would make sense for by-value, so maybe the lint just carried over by mistake.

I know somebody has been doing a lot of work on that lint so I’ll have to dig a bit.

[tgross35]

rust-lang/rust#132699 is the Rust issue (looks like @mbyx found it before I did)

[neuschaefer]

There is one instance in the tests where a type U=kinfo_proc is passed by value, it's the roundtrip test, ctest_roundtrip_kinfo_proc (Rust) / ctest_roundtrip__kinfo_proc (C) in this case. The C function takes a MaybeUninit<U> as its first argument, and returns a U.

If this is considered all right, I'd suggest #[allow]-ing improper_ctypes for the ctest_output module.