[qodelicious]

In a POSIX regex a quantifier right after another one applies to the whole repeated part. So b+? is (b+)? and b*+ is (b*)+, both greedy. The regex engines read a ? or + there as a lazy or possessive modifier instead, so sed gave different results from GNU.

$ printf 'abbc\n' | sed -E 's/ab+?/X/'
Xbc        # GNU sed gives Xc
$ printf 'bbc\n' | sed -E 's/b+?/X/'
Xbbc       # GNU sed gives Xc
$ printf 'abba\n' | sed -E 's/(a)b*+b\1/X/'
abba       # GNU sed gives X

BREs hit it too with \+\?.

Now compile_regex puts the repeated part in a non-capturing group before a stacked quantifier, so ab+? becomes a(?:b+)?. Escapes, bracket expressions and groups count as the thing being repeated. ensure_dotall read a leading (?: as a flag group and turned (?:b+)? into (?:b+s)?, so it now puts (?s) in front of a group like that.

Added test_quantifier_after_quantifier_stays_greedy (all cases checked against GNU sed 4.9) and unit tests for the rewrite and for ensure_dotall. The integration test fails on main and passes with the change.

[codecov]

Codecov Report

❌ Patch coverage is 98.61111% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 95.16%. Comparing base (e154c18) to head (17532c6).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
src/sed/compiler.rs	98.54%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #642      +/-   ##
==========================================
+ Coverage   95.07%   95.16%   +0.08%     
==========================================
  Files          14       14              
  Lines        7459     7634     +175     
  Branches      441      460      +19     
==========================================
+ Hits         7092     7265     +173     
- Misses        366      368       +2     
  Partials        1        1              

Flag	Coverage Δ
macos_latest	96.63% <98.61%> (+0.05%)	⬆️
ubuntu_latest	96.57% <98.61%> (+0.05%)	⬆️
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]

Merging this PR will improve performance by 4.17%

1 warning

⚠️ 9 benchmarks with a significant change ran on different hardware

Their base and head results were measured on machines with a different CPU, core count or memory, so the changes might not come from the code. Run base and head on the same runner type.

Learn more about Macro Runners

See affected benchmarks

⚡ 9 improved benchmarks
✅ 2 untouched benchmarks

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	no_op_short	2.2 s	2 s	+7.87%
⚡	access_log_append	1,054.7 ms	995.8 ms	+5.92%
⚡	access_log_no_op	1.6 s	1.5 s	+5.4%
⚡	access_log_translit	991.9 ms	949.6 ms	+4.46%
⚡	remove_cr	376.5 ms	364.4 ms	+3.31%
⚡	access_log_no_subst	861.9 ms	835 ms	+3.22%
⚡	genome_subst	371.1 ms	359.7 ms	+3.18%
⚡	access_log_subst	2 s	2 s	+2.21%
⚡	number_fix	584.6 ms	572.4 ms	+2.14%

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing qodelicious:stacked-quantifiers (17532c6) with main (273dcfb)}

[sylvestre]

in byte mode these bytes are whole characters, no?
a\xa9+? becomes (?:a\xa9+)? here, GNU with LC_ALL=C reads it as a(?:\xa9+)?
please pass the character mode and add a test

[qodelicious]

You're right, in byte mode those are whole characters. It now gets the character mode and only skips 0x80-0xbf as continuation bytes in UTF-8 mode, so with LC_ALL=C a\xa9+? becomes a(?:\xa9+)?. Added a unit test for both modes and a LC_ALL=C test in test_sed.rs (s/a\xa9+?/X/ on ab gives Xb, same as GNU).

[sylvestre]

could you please run the GNU testsuite too?
git clone https://github.com/mirror/sed.git ../gnu.sed
./util/run-gnu-testsuite.sh

[qodelicious]

Ran it with the branch on top of current main. 19 pass / 30 fail / 16 skip, same as main test by test, both with LC_ALL=C and in a UTF-8 locale. So no regressions, but nothing new passes either (the skips are the multibyte locale tests, I don't have those locales here).