sed: support back-references in byte mode

[lKolabrodl]

Closes #543.

Problem

Byte mode rejected every regular expression containing a back-reference before attempting a match. This made address and substitution expressions such as \(a\)b\1 fail whenever the C locale selected byte-oriented matching.

User impact

Scripts using back-references could work in a UTF-8 locale but fail immediately in the C locale. Invalid UTF-8 input could not use back-references at all.

Root cause

The fast byte regex engine does not implement back-references, while the fallback fancy_regex engine accepts only UTF-8 text and reports UTF-8 offsets.

Fix

• add a byte-mode fancy_regex path that maps each input byte to one Unicode scalar
• translate matches and capture offsets back to the original byte buffer
• keep captures_iter streaming instead of collecting every match
• preserve captures from invalid UTF-8 input
• cover the reported address expression and an invalid UTF-8 substitution

Verification

• cargo test --lib — 360 passed
• cargo test backref — 9 related tests passed
• cargo clippy --lib --tests -- -D warnings
• cargo fmt --all -- --check

The full integration run on Windows reached 219 passing tests, 2 ignored tests, and 47 failures caused by existing CRLF-sensitive shell-script and fixture expectations. The focused integration tests for this change pass.

AI disclosure

AI assistance was used to help trace the byte/UTF-8 boundary, implement the offset mapping, and draft tests. The resulting code was reviewed and verified with the commands above.

[codecov]

Codecov Report

❌ Patch coverage is 51.12782% with 65 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.42%. Comparing base (f1c5813) to head (68b4e4e).

Files with missing lines	Patch %	Lines
src/sed/fast_regex.rs	51.12%	65 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #545      +/-   ##
==========================================
- Coverage   83.04%   82.42%   -0.62%     
==========================================
  Files          13       13              
  Lines        7046     7170     +124     
  Branches      401      404       +3     
==========================================
+ Hits         5851     5910      +59     
- Misses       1192     1257      +65     
  Partials        3        3              

Flag	Coverage Δ
macos_latest	83.02% <51.12%> (-0.65%)	⬇️
ubuntu_latest	83.24% <51.12%> (-0.64%)	⬇️
windows_latest	0.00% <0.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[codspeed-hq]

Merging this PR will degrade performance by 0.87%

⚡ 1 improved benchmark
❌ 2 regressed benchmarks
✅ 8 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	number_fix	1.1 s	1.2 s	-2.97%
❌	genome_subst	443.7 ms	453.1 ms	-2.09%
⚡	access_log_subst	2.4 s	2.3 s	+2.53%

Tip

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

---

_{Comparing lKolabrodl:fix/543-byte-backrefs (68b4e4e) with main (f1c5813)}

[sylvestre]

could you please add more tests? the code coverage isn't ideal:
https://app.codecov.io/gh/uutils/sed/pull/545?dropdown=coverage&src=pr&el=h1

[lKolabrodl]

Added five focused unit tests in 51aaf56 covering non-ASCII byte offset mapping, optional capture groups, multiple invalid UTF-8 matches, zero-width iteration, and no-match paths. Locally, cargo fmt --all -- --check, cargo test fancy_byte --lib (7 passed), and cargo test backref (9 passed) all pass. Thanks!