[sherenkov]

A sed replacement written the unix way, s/([0-9]+)/\\1!/g, put the digit 1 in place of the group, so the rule of #2255 turned "Hello, 2025" into "Hello, 1!", with no error. nextUntilSlash reads the replacement with escape on and dropped the backslash before any character (src/Functions.hs:171), so \1 reached substituteGroups of src/Regexp.hs as a plain 1, and there only $ followed by digits named a group.

Now nextUntilSlash keeps the backslash in the replacement when a digit or another backslash follows it, and substituteGroups reads \1 to \9 as the group of that one digit, as sed does, so \10 is group one followed by 0, and reads \\ as one backslash, so \\1 stays a literal \1. An escaped slash or brace is read as before, and so is $1. A back reference to a group the pattern does not have is kept as it is, the way $9 already is. The README now says that \1 to \9 and $1 name the groups.

FunctionsSpec runs the rule of the issue through sed and expects "Hello, 2025!", and checks that \\1 stays a backslash and a 1; RegexpSpec checks the back reference, the single digit it takes, one out of bounds and the escaped backslash in substituteGroups. The first of them gives "Hello, 1!" without the change.

Fixes #2255

[sherenkov]

@yegor256 CI is green and the branch is up to date with master, could you take a look when you have a moment? It makes sed read \1 to \9 in a replacement as the groups they name and \\ as one backslash, instead of dropping every backslash.

[github-actions]

Benchmark comparison

Median of 5 round(s) per side, interleaved on the same runner against the PR merge-base, with the bench fixture frozen across both builds. Noise is the spread between the fastest and the slowest round, taken from the noisier side.

Bench	base (μs)	head (μs)	delta	noise
morph/symbolic/accum/0	113.89	109.11	-4.2%	±1.9%
morph/symbolic/accum/400	134.06	128.16	-4.4%	±24.0%
morph/symbolic/demo/e1	132.89	131.18	-1.3%	±3.1%
morph/symbolic/demo/e2	132.73	128.88	-2.9%	±2.4%
morph/symbolic/demo/e3	133.91	129.76	-3.1%	±1.8%
morph/symbolic/demo/e4	134.19	129.69	-3.4%	±2.0%
morph/symbolic/demo/e5	135.71	131.71	-3.0%	±1.6%
morph/symbolic/native/e5	1158.04	1131.37	-2.3%	±26.3%
parse/phi	234777.47	235534.27	+0.3%	±2.1%
parse/xmir	947819.74	931013.63	-1.8%	±1.3%
print/salty/multiline	1750699.97	1807544.54	+3.2%	±2.9%
print/sweet/flat	66647.89	66878.76	+0.3%	±5.7%
print/sweet/multiline	538542.21	560425.49	+4.1%	±3.3%
rewrite/normalize	2.23	2.17	-2.7%	±5.6%

Scaling

Each sized case against its /0 sibling, the same work over nothing extra: a ratio near 1 means the extra size is free.

Bench	against	base	head
morph/symbolic/accum/400	morph/symbolic/accum/0	1.18×	1.17×

No case moved by more than 20% beyond its own noise.
This report is informational, a delta never fails the build.