[webkit-changes] [WebKit/WebKit] 372410: Invalidate scope-breaking :has(:is(...)) selectors...
Cameron McCormack
noreply at github.com
Fri Sep 15 14:16:57 PDT 2023
Branch: refs/heads/main
Home: https://github.com/WebKit/WebKit
Commit: 3724105055b3d155fee1da5604623c6200cf1f28
https://github.com/WebKit/WebKit/commit/3724105055b3d155fee1da5604623c6200cf1f28
Author: Cameron McCormack <heycam at apple.com>
Date: 2023-09-15 (Fri, 15 Sep 2023)
Changed paths:
M LayoutTests/imported/w3c/web-platform-tests/css/selectors/invalidation/is-pseudo-containing-complex-in-has-expected.txt
M LayoutTests/imported/w3c/web-platform-tests/css/selectors/invalidation/not-pseudo-containing-complex-in-has-expected.txt
M Source/WebCore/style/ChildChangeInvalidation.cpp
M Source/WebCore/style/ChildChangeInvalidation.h
M Source/WebCore/style/ClassChangeInvalidation.cpp
M Source/WebCore/style/RuleFeature.cpp
M Source/WebCore/style/RuleFeature.h
M Source/WebCore/style/StyleInvalidator.cpp
M Source/WebCore/style/StyleInvalidator.h
M Source/WebCore/style/StyleScopeRuleSets.cpp
M Source/WebCore/style/StyleScopeRuleSets.h
Log Message:
-----------
Invalidate scope-breaking :has(:is(...)) selectors more thoroughly
https://bugs.webkit.org/show_bug.cgi?id=253944
rdar://106768250
Reviewed by Antti Koivisto.
It's possible to write :has() selectors with arguments that can match
elements outside of the :has scope, such as
:has(:is(.x .y))
which could match the .x against an ancestor of the :has scope, and
:has(~ :is(.x .y))
which could match the .x against an ancestor of a later sibling of the
:has scope.
We currently handle scope breaking :has() selectors by generating
RuleFeatures based on the nested selectors' features and storing them
with MatchElement::HasNonSubjectOrScopeBreaking, which causes
invalidation to match the inner scope-breaking selector (the :is())
against all elements in the document.
Because we store these in InvalidationRuleSets keyed off the nested
selectors' features, it means that for the above examples, we will
process these document-wide invalidations whenever a class name changes
to or from x or y. But this is not sufficient to invalidate for
selectors like
:has(:is(.x + .y .z))
where if the .x and .y elements are outside the :has scope, an element
insertion or removal between them will not find any relevant
invalidation rule sets, and
:has(~ :is(.x ~ .y))
for similar reasons.
This PR changes how we collect scope breaking :has() rules, so that in
addition to generating entries in m_hasPseudoClassInvalidationRuleSets
on Style::ScopeRuleSet, we also record them in a new
m_scopeBreakingHasPseudoClassInvalidationRuleSet, which we look up for
every element insertion or removal. For simplicity, we do this more
drastic invalidation for both of the above cases.
MatchElement::HasNonSubjectOrScopeBreaking is split into two separate
values -- HasNonSubject and HasScopeBreaking -- where only the latter
causes an entry to be added to
m_scopeBreakingHasPseudoClassInvalidationRuleSet.
This change does not regress the performance of invalidation of :has()
selectors that do not contain logical combination pseudo-classes like
:is() and :not(), or which have do have one but do not have any selector
combinators inside them.
* LayoutTests/imported/w3c/web-platform-tests/css/selectors/invalidation/is-pseudo-containing-complex-in-has-expected.txt:
* LayoutTests/imported/w3c/web-platform-tests/css/selectors/invalidation/not-pseudo-containing-complex-in-has-expected.txt:
* Source/WebCore/style/ChildChangeInvalidation.cpp:
(WebCore::Style::ChildChangeInvalidation::invalidateForChangedElement):
(WebCore::Style::ChildChangeInvalidation::invalidateForChangeOutsideHasScope):
(WebCore::Style::ChildChangeInvalidation::invalidateForHasBeforeMutation):
(WebCore::Style::ChildChangeInvalidation::invalidateForHasAfterMutation):
(WebCore::Style::needsDescendantTraversal):
* Source/WebCore/style/ChildChangeInvalidation.h:
* Source/WebCore/style/ClassChangeInvalidation.cpp:
(WebCore::Style::ClassChangeInvalidation::computeInvalidation):
* Source/WebCore/style/RuleFeature.cpp:
(WebCore::Style::isSiblingOrSubject):
(WebCore::Style::isHasPseudoClassMatchElement):
(WebCore::Style::isScopeBreaking):
(WebCore::Style::computeNextHasPseudoClassMatchElement):
(WebCore::Style::computeHasPseudoClassMatchElement):
(WebCore::Style::computeSubSelectorMatchElement):
(WebCore::Style::RuleFeatureSet::recursivelyCollectFeaturesFromSelector):
(WebCore::Style::RuleFeatureSet::collectFeatures):
(WebCore::Style::RuleFeatureSet::add):
(WebCore::Style::RuleFeatureSet::clear):
(WebCore::Style::RuleFeatureSet::shrinkToFit):
* Source/WebCore/style/RuleFeature.h:
(WebCore::Style::RuleFeatureSet::usesMatchElement const):
(WebCore::Style::RuleFeatureSet::usesHasPseudoClass const):
* Source/WebCore/style/StyleInvalidator.cpp:
(WebCore::Style::Invalidator::invalidateStyleWithMatchElement):
(WebCore::Style::Invalidator::invalidateWithScopeBreakingHasPseudoClassRuleSet):
* Source/WebCore/style/StyleInvalidator.h:
* Source/WebCore/style/StyleScopeRuleSets.cpp:
(WebCore::Style::ScopeRuleSets::collectFeatures const):
* Source/WebCore/style/StyleScopeRuleSets.h:
(WebCore::Style::ScopeRuleSets::scopeBreakingHasPseudoClassInvalidationRuleSet const):
Canonical link: https://commits.webkit.org/268038@main
More information about the webkit-changes
mailing list