mkdir, install, mkfifo, mknod: fix -m's invalid-mode message and an empty-clause bug

[arbelonson-source]

Follow-up to #14301 (chmod). mkdir/install/mkfifo/mknod share the same mode parser chmod uses, and — as flagged in that PR's body — have the identical class of bug in their own -m/--mode handling: each leaked a message shaped like whatever failed to parse internally, rather than GNU's own wording.

$ mkdir -m 999 d          # ours, before: a raw ParseIntError
mkdir: invalid digit found in string
$ mkdir -m 999 d          # GNU 9.11
mkdir: invalid mode '999'

Three different messages, verified individually — not assumed

GNU's wording is not identical across these four, so I checked each rather than assuming chmod's fix applies as-is:

• mkdir/install name the operand, but unlike chmod use no colon before the quote and add no "Try --help" hint: mkdir: invalid mode '999', install: invalid mode '999'.
• mkfifo/mknod don't name the operand at all, regardless of what was wrong with it: just mkfifo: invalid mode / mknod: invalid mode.

A second, more serious bug

Investigating turned up something beyond message wording, in the parser these four share but chmod's own inline loop does not use (which is why it didn't affect chmod): an empty clause in a comma-separated mode was silently skipped as a no-op instead of erroring.

$ mkdir -m ',' d                       # ours, before: silently created d with a default mode
$ mkdir -m 'u+rwx,' d                  # ours, before: same
$ mkdir -m ',' d                       # GNU
mkdir: invalid mode ','

Checked against GNU with a leading, trailing, and doubled comma — none are tolerated, so parse_chmod's if mode_part.is_empty() { continue; } was simply wrong; removing it lets parse_symbolic's existing empty-string handling report it the way it already does for chmod.

Also touches describe()

mode.rs's describe() needed the same fix chmod's PR made: now that the headline never explains which operator was invalid or expected, that detail lives in the caret-diagram label instead. This function is shared, so both PRs touch the same three lines with the same final text — whichever of the two merges second will need a trivial, content-identical rebase.

Testing

• Differential sweep against GNU 9.11 for all four utilities across bad digits, too-large octal, missing/invalid operators, empty mode, and the empty-clause cases (leading/trailing/doubled comma): all match.
• New regression tests in each of the four utilities' test files, each verified to fail without its respective fix (5 failures reproduced by stashing the fix and re-running).
• cargo test --features mkdir,install,mkfifo,mknod,chmod --test tests -- test_mkdir test_install test_mkfifo test_mknod test_chmod: 257 passed, 0 failed (all pre-existing tests unaffected; chmod's own suite included since it shares mode.rs).
• moz-fluent-lint on all touched locale directories: no errors.
• cargo fmt --check and cargo clippy -p uu_mkdir -p uu_install -p uu_mkfifo -p uu_mknod -p uucore --features mode --all-targets -- -D warnings: clean.

Disclosure

Prepared with AI assistance (Claude Opus 5, via Claude Code), per the AI policy in CONTRIBUTING.md. GNU's behaviour was established by running the installed GNU binary as a black box; I did not read GNU coreutils source. All testing was run locally.

[github-actions]

GNU testsuite comparison:

Skip an intermittent issue tests/date/resolution (fails in this run but passes in the 'main' branch)
Skipping an intermittent issue tests/cut/bounded-memory (passes in this run but fails in the 'main' branch)
Note: The gnu test tests/tail/pipe-f is now being skipped but was previously passing.
Congrats! The gnu test tests/pr/bounded-memory is now passing!