mktemp: strip OS error code from permission-denied message

[MsfPablo]

What

Creating a temporary file or directory in an unwritable directory leaked the raw Permission denied (os error 13) at path "..." text from the underlying tempfile crate, instead of the clean Permission denied message used elsewhere in uutils:

$ coreutils mktemp -p /tmp/noperm   # chmod 000
mktemp: Permission denied (os error 13) at path "/tmp.haKXBI6xNU"

Why

tempfile wraps the underlying io::Error as io::Error::new(kind, "{<inner display>} at path {path:?}"). That wrapper has raw_os_error() == None, so UIoError::Display skips its kind-match normalization branch and prints self.inner.to_string() verbatim — which still contains the (os error 13) suffix and the internal at path "..." detail.

Routing the error through a dedicated MkTempError::Io(std::io::Error) variant rendered with strip_errno (as @sylvestre suggested in the issue) drops both the (os error N) suffix and the internal path detail, matching how perms.rs (common-write-error) and csplit render io errors:

$ coreutils mktemp -p /tmp/noperm
mktemp: Permission denied
$ coreutils mktemp -d -p /tmp/noperm
mktemp: Permission denied

Changes

• src/uu/mktemp/src/mktemp.rs: add MkTempError::Io(std::io::Error) rendered with strip_errno, and route the two catch-all Err(e) => Err(e.into()) arms in make_temp_file / make_temp_dir through it.
• tests/by-util/test_mktemp.rs: add test_permission_denied_clean_message (Unix, skipped under root) asserting the clean message for both the file and directory cases.

Closes #13720.

Verification

• cargo build --bin coreutils --no-default-features --features mktemp
• cargo test --no-default-features --features mktemp --test tests -- mktemp → 41 passed
• cargo fmt --all -- --check → clean
• cargo clippy --no-default-features --features mktemp → clean

AI disclosure

This change was authored with the assistance of Claude (Anthropic). The diagnosis of the UIoError::Display fall-through, the fix design (mirroring perms.rs / csplit), the test, and the verification were all produced with AI assistance and reviewed by me.

Note on prior claim

I saw that @AnandajithS mentioned working on this in the issue and @Devel08 followed up. No PR appeared after several days, so I went ahead. Happy to close/defer this if a PR from the original claimant is in flight — let me know.

[MsfPablo]

Both failing lint jobs (Style and Lint (unix) and Style/lint (ubuntu-latest, all, true)) report the same error:

error: unnecessary qualification
   --> src/uu/stat/src/stat.rs:1040:47

This is in uu_stat, which this PR does not touch — it's a pre-existing lint already fixed on main in #13814 (commit fb4904b8). This branch was based off just before that fix landed, so it surfaces here. uu_mktemp itself checks clean (the log shows Checking uu_mktemp v0.10.0 passing). Updating the branch against main resolves it; happy to rebase if that's preferred.

[sylvestre]

Could you please write the messages yourself? Llm comments are not helpful

[xtqqczze]

Suggested change

	fn test_permission_denied_clean_message() {
	fn test_permission_denied() {

The description should focus on testing the behavior being testing rather than lacking the specific (os error N) suffix (which is obvious).

[codspeed-hq]

Merging this PR will degrade performance by 4.91%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

❌ 2 regressed benchmarks
✅ 345 untouched benchmarks
⏩ 46 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	du_max_depth_balanced_tree[(6, 4, 10)]	63 ms	66.6 ms	-5.47%
❌	Simulation	du_summarize_balanced_tree[(5, 4, 10)]	16.3 ms	17 ms	-4.35%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing MsfPablo:mktemp-strip-os-error-v2 (dc41de7) with main (a8efa79)}

Footnotes

1. 46 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[Devel08]

Hi, it'd be good if the error message is mktemp: failed to create file via template '/tmp.XXXXXXXXXX': Permission denied (the path has to be path/tmp.XXXXXXXXXX), the error message that your branch gives is just Permission denied, which is worse than what we had in the main branch, that was the issue with using only strip_errno() on it, also the test can be as simple as just doing -p / as the runners don't have permission for it AFAIK :)

[github-actions]

GNU testsuite comparison:

Skip an intermittent issue tests/date/resolution (fails in this run but passes in the 'main' branch)

[xtqqczze]

Hi, it'd be good if the error message is mktemp: failed to create file via template '/tmp.XXXXXXXXXX': Permission denied

Agreed, I don't think stripping the path is an improvement.