Branch: refs/heads/main
Home: https://github.com/WebKit/WebKit
Commit: a458a6c1f0a71390a7641f8fbb5a5485c6900a1e
https://github.com/WebKit/WebKit/commit/a458a6c1f0a71390a7641f8fbb5a5485c6900a1e
Author: Sosuke Suzuki <[email protected]>
Date: 2026-08-01 (Sat, 01 Aug 2026)
Changed paths:
M Source/JavaScriptCore/yarr/YarrPattern.cpp
Log Message:
-----------
[YARR] Stop the chunked v-mode class set operation loop as soon as it can
produce nothing more
https://bugs.webkit.org/show_bug.cgi?id=320619
Reviewed by Yusuke Suzuki.
CharacterClassConstructor::nonLatin1OpSorted() computes v-mode class set
operations over the non-Latin-1 range by walking 0x100..0x10FFFF in
2048-codepoint chunks until every LHS/RHS index is consumed. It had an
early-exit for intersection with an exhausted RHS, but that compared
rhsRangeIndex with '>' against rhsRanges32.size(), which can never be true,
so /[\p{Any}&&[a-c]]/v scanned all ~540 chunks even though the RHS has no
non-Latin-1 elements at all.
Replace the loop condition with an op-aware canProduceMore() predicate:
union continues while either side has input left, intersection while both
do, and subtraction while the LHS does. This subsumes the dead early-exit and
the empty-input bail-out, and additionally stops early when the LHS of an
intersection or a subtraction is exhausted. The resulting character sets are
unchanged.
Microbenchmark (500 constructions of new RegExp("(?:x" + i + ")?" + pattern,
"v")):
pattern baseline patched
[\p{Any}&&[a-c]] 1876.6 ms 2.9 ms
[[a-c]&&\p{Any}] 1866.6 ms 2.3 ms
[\p{L}&&\p{Any}] 2458.2 ms 981.7 ms
[\p{Any}--[a-c]] 4862.2 ms 4932.5 ms
[[\u{100}-\u{200}]--\p{Any}] 1774.9 ms 6.8 ms
[\p{L}[0-9]] 645.5 ms 696.4 ms
* Source/JavaScriptCore/yarr/YarrPattern.cpp:
(JSC::Yarr::CharacterClassConstructor::nonLatin1OpSorted):
Canonical link: https://commits.webkit.org/318392@main
To unsubscribe from these emails, change your notification settings at
https://github.com/WebKit/WebKit/settings/notifications
