csplit: report the size of the split the input failed in

[arbelonson-source]

Fixes #13139.

csplit prints the size of each split as it closes it, but a read error returns straight out of the split loop, so the split that was open when the input went wrong is never accounted for:

$ csplit a_directory '/^a/'    # ours, before
csplit: read error: Is a directory

$ csplit a_directory '/^a/'    # GNU 9.11
csplit: read error: Is a directory
0

The two other ways a split can end early — MatchNotFound and LineOutOfRange — already call finish_split before returning their error, which is why csplit f '/^zzz/' prints its count today and this one does not. Read errors now take the same path, through a small helper on the writer that closes the split before handing the error on. That keeps the count subject to the same options as any other: --quiet silences it, -z drops the empty split instead of counting it, -k keeps the file it counted, and a %REGEXP% section that writes nowhere still prints nothing.

One difference remains, in ordering rather than content: GNU prints the count after the error, because it closes the split during its cleanup; we print it before, because the error is reported by the caller once csplit returns. The streams are separate, so this is only visible when both are pointed at the same terminal.

Testing

• 24-case differential check against GNU 9.11 comparing stdout, stderr, exit code and the files left behind: a directory and /proc/self/mem as input, crossed with -k, -z, -s, -b, line patterns, regex patterns, {N} repeats and %REGEXP%, plus the ordinary successful splits and the no-match and out-of-range cases that had to keep their existing counts. All 24 match.
• New test_read_error_reports_the_size_of_the_open_split covering the count under -k, -z and -s.
• test_directory_input_file used stderr_only, which asserted the missing count; it now checks both streams and the GNU-matching output.
• cargo test --features csplit --test tests -- test_csplit: 93 passed, 0 failed; cargo test -p uu_csplit: 41 passed.
• cargo fmt --check and cargo clippy -p uu_csplit --all-targets: 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.

[codspeed-hq]

Merging this PR will degrade performance by 4.66%

❌ 2 regressed benchmarks
✅ 20 untouched benchmarks
⏩ 390 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	csplit_line_number	56 ms	59.1 ms	-5.19%
❌	Simulation	csplit_regex_pattern	73 ms	76.2 ms	-4.13%

Tip

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

---

_{Comparing arbelonson-source:fix/csplit-count-on-read-error (d20a2d3) with main (2228600)²}

Footnotes

1. 390 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. ↩

2. No successful run was found on main (3251833) during the generation of this report, so 2228600 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[github-actions]

GNU testsuite comparison:

Skipping an intermittent issue tests/date/date-locale-hour (passes in this run but fails in the 'main' branch)
Congrats! The gnu test tests/tail/tail-n0f is now passing!