[arbelonson-source]

fold's handle_obsolete scanned every argument for -DIGITS and pulled the first match out, without checking whether that argument was already spoken for. Five GNU incompatibilities followed from that one function.

Found by differential testing against GNU coreutils 9.11; each row below was confirmed by running both binaries.

Command	GNU	uutils before
fold -w -1 FILE	invalid number of columns: '-1'	invalid number of columns: 'FILE'
fold -w3 -5	folds at 5	folds at 3
fold -3 -5	folds at 5	error: unexpected argument '-5' found
fold -w 3 -w 5	folds at 5	error: ... cannot be used multiple times
fold -- -3	fold: -3: No such file or directory	reads stdin

The first row also applies to -bw -1, -sw -1 and --width -1: -1 was taken as the obsolete form, so -w bound to the file name and reported that as the invalid width.

Approach

Rather than lifting one width out of the argument list, the obsolete form is now rewritten into --width=N in place. That keeps the original left-to-right order, so clap can apply the same last-one-wins rule GNU gets from processing options in sequence — enabled here with args_override_self(true), matching what cut already does.

An argument is only eligible for that rewrite when it is not the value of -w/--width and has not been placed after a -- terminator. Short clusters are handled by position: -sw takes the next argument, while -w3 and -wb carry their value inline.

Testing

Three regression tests added. I verified they actually catch the bugs by reverting the change to fold.rs alone — all three fail, then pass again with it restored.

• cargo test --features fold --no-default-features test_fold: 99 passed, 0 failed (96 pre-existing, 3 new)
• cargo fmt --check and cargo clippy -p uu_fold --all-targets: clean
• 26-case differential sweep against GNU: 24 match exactly

The 2 that still differ are fold -0 / fold -w 0, where GNU says Numerical result out of range and uutils says Result too large. That is a pre-existing wording difference in the error string, unrelated to argument parsing, so I left it alone rather than widen this PR.

Partially related to #5998 (allowing repeated flags), though this only changes fold.

Disclosure

Prepared with AI assistance (Claude Opus 5, via Claude Code), per the AI policy in CONTRIBUTING.md. On the GPL concern raised there specifically: the expected behavior was derived by running the installed GNU fold binary as a black box and recording its output. I did not read GNU coreutils source while writing this, and the implementation approach (rewrite-in-place + args_override_self) follows the existing pattern in this repo's cut, not anything from GNU. All testing above was run locally.

[github-actions]

Binary size comparison:

Individual binary size comparison VS main (threshold: >=5% AND >=4 KB).

Total size of compared binaries: 152.20 MB (+1016 KB, +0.66%)

Significant per-binary changes:
  comm     1.12 MB ->    2.33 MB  (+1.21 MB, +107.29%)

[github-actions]

GNU testsuite comparison:

Skip an intermittent issue tests/tail/tail-n0f (fails in this run but passes in the 'main' branch)
Skip an intermittent issue tests/pr/bounded-memory (was skipped on 'main', now failing)

[sylvestre]

sorry, i broke it

[arbelonson-source]

No apology needed — and thank you for pushing the fix rather than waiting on me. short.find('w') == short.len().checked_sub(1) is neater than what I'd have written; comparing the two Options directly makes the empty case fall out instead of needing a guard in front of it.

I've added a test on top so it stays fixed: - as an operand on its own, alongside a width, and as the value of -w (where GNU rejects it as a width rather than reading it as the operand). It reproduces the abort on the commit before yours.

100 fold tests pass, cargo fmt --check clean.

Prepared with AI assistance (Claude Opus 5, via Claude Code); comparisons run locally against GNU coreutils 9.11.

[cakebaker]

Thanks!

[xtqqczze]

test_fold::test_dash_operand_reads_stdin (introduced here) is failing: #15004