Branch: refs/heads/main
Home: https://github.com/WebKit/WebKit
Commit: ddf3655422ded3195d1f6f598a504cfcb2b1370c
https://github.com/WebKit/WebKit/commit/ddf3655422ded3195d1f6f598a504cfcb2b1370c
Author: Sosuke Suzuki <[email protected]>
Date: 2026-08-15 (Sat, 15 Aug 2026)
Changed paths:
A JSTests/microbenchmarks/regexp-lookbehind-unicode-bmp-body.js
A JSTests/microbenchmarks/regexp-lookbehind-unicode-fixed.js
A JSTests/microbenchmarks/regexp-lookbehind-unicode-variable.js
A JSTests/microbenchmarks/regexp-u-flag-dotall-greedy.js
A JSTests/stress/regexp-lookbehind-jit-unicode.js
A JSTests/stress/regexp-unicode-any-character-inside-surrogate-pair.js
M Source/JavaScriptCore/runtime/RegExp.cpp
M Source/JavaScriptCore/yarr/YarrJIT.cpp
Log Message:
-----------
[YARR] Compile lookbehinds in unicode patterns in the JIT
https://bugs.webkit.org/show_bug.cgi?id=321656
Reviewed by Daniel Liu.
319067@main still sent every pattern with a lookbehind and the u or v flag to
the interpreter.
Two things were missing. reverseDisjunctionForBackward() now counts a fixed
non-BMP character or
class as two code units, like setupAlternativeOffsets() does. A Backward term
points at the last
code unit of its character, so on 16-bit subjects readCharacter() decodes it
with the new
tryReadUnicodeCharBackward(). Inside a surrogate pair it reads errorCodePoint,
like the forward
read does, so a lookbehind fails there as in V8; the interpreter is wrong at
such positions and
in several other unicode lookbehinds (bug 317275), so the new test runs on the
JIT only.
The any-character class (. with the s flag, [\s\S]) did not reject
errorCodePoint, so it matched
half of a pair even before this patch, e.g.
/(?!\u{1F600})./msu.exec("\u{1F600}") gave
[1, "\uDE00"]. It now gets the same test as an inverted class.
Baseline Patched
regexp-lookbehind-unicode-fixed 309.45+-1.91 ^ 47.49+-0.66 ^
definitely 6.5161x faster
regexp-lookbehind-unicode-bmp-body 215.05+-3.01 ^ 36.89+-0.69 ^
definitely 5.8299x faster
regexp-lookbehind-unicode-variable 186.04+-1.55 ^ 35.13+-0.75 ^
definitely 5.2958x faster
regexp-lookbehind-variable 34.26+-0.19 ? 34.58+-0.54 ?
regexp-u-flag-dotall-greedy 40.03+-1.08 38.17+-3.94
might be 1.0488x faster
Tests: JSTests/microbenchmarks/regexp-lookbehind-unicode-bmp-body.js
JSTests/microbenchmarks/regexp-lookbehind-unicode-fixed.js
JSTests/microbenchmarks/regexp-lookbehind-unicode-variable.js
JSTests/microbenchmarks/regexp-u-flag-dotall-greedy.js
JSTests/stress/regexp-lookbehind-jit-unicode.js
JSTests/stress/regexp-unicode-any-character-inside-surrogate-pair.js
* JSTests/microbenchmarks/regexp-lookbehind-unicode-bmp-body.js: Added.
* JSTests/microbenchmarks/regexp-lookbehind-unicode-fixed.js: Added.
* JSTests/microbenchmarks/regexp-lookbehind-unicode-variable.js: Added.
* JSTests/microbenchmarks/regexp-u-flag-dotall-greedy.js: Added.
* JSTests/stress/regexp-lookbehind-jit-unicode.js: Added.
(shouldBe):
(matchOf):
(indicesOf):
(shouldBe.matchOf):
* JSTests/stress/regexp-unicode-any-character-inside-surrogate-pair.js: Added.
(shouldBe):
(matchOf):
(shouldBe.matchOf):
* Source/JavaScriptCore/runtime/RegExp.cpp:
(JSC::RegExp::compile):
(JSC::RegExp::compileMatchOnly):
* Source/JavaScriptCore/yarr/YarrJIT.cpp:
Canonical link: https://commits.webkit.org/319226@main
To unsubscribe from these emails, change your notification settings at
https://github.com/WebKit/WebKit/settings/notifications
