CI failures are unrelated issues:

• broken CI: Windows code coverage always fails #6534
• flaky CI Android: sshd doesn't start? #6570 (both)
• And some new MacOS issue? Similar to test_stdin_pipe_fifo1 flakes with "weird file" on MacOS #6145, but not identical? See below.

---- test_stat::test_stdin_pipe_fifo1 stdout ----
run: /Users/runner/work/coreutils/coreutils/target/aarch64-apple-darwin/debug/coreutils stat -
thread 'test_stat::test_stdin_pipe_fifo1' panicked at 'Expected stderr to be empty, but it's:
stat: cannot stat '-': Bad file descriptor (os error 9)
', tests/by-util/test_stat.rs:279:10

Changes since initial push:

• Added a test (doink!)
• Changed the output method so that non-UTF-8 filenames are correctly output (without Unicode replacement encoding or other loss)
• Unified the "OsStr(ing) to bytes" conversion with cut, as already suggested in tr: Accept non UTF-8 arguments for sets #6563

Changes since last push description:

• Formatting (looks like something disabled my pre-commit hook, whoops)
• Added missing #[cfg(unix)] on use
• Fix the unified definition to work on windows – turns out, a if cfg!() branch must successfully compile even if it is not taken.
• Added some spellcheck exceptions. (These should really be added to the pre-commit hook)
• Restrict the test to linux-only, because MacOS doesn't allow non-UTF-8 filenames in the first place

Changes since last push description:

• Now that the test is restricted, the initial import must only happen on linux. Aaargh!
• Both of them.

Remaining CI failures are unrelated:

• broken CI: Windows code coverage always fails #6534
• flaky CI Android: sshd doesn't start? #6570 (both)
• Timeout while downloading a 877 byte(!) file, probably because there have been so many restarts

Changes since last push description:

• Rebase (also solves most CI failures, except broken CI: Windows code coverage always fails #6534 of course)
• Use the shared code from tr: Accept non UTF-8 arguments for sets #6563 for converting a OsStr to &[u8]
• Remove the .to_os_string() that's now unnecessary (This took two attempts – why didn't my clippy catch that?!)

GNU testsuite comparison:

Congrats! The gnu test tests/rm/rm2 is no longer failing!

Changes since (my?) last push:

• Remove notfile from the match structure, reducing code duplication
• Rename print_filename to should_print_filename
• Reduce filename copying in the test

Changes since last push:

• Turn if !not_file { " " } else { "" } into if not_file { "" } else { " " } to make clippy happy (why wasn't this caught by pre-commit?)

Thanks :)