[gravitypdf-bot]

:has() never matched in the current engine (CSS\DOMTraverser), which find(), filter() and the factories use. PseudoClass::has() ran its argument in initialized mode against the element itself, so ul:has(li) asked whether the ul is an li.

$xml = '<?xml version="1.0"?><root><ul id="a"><li/><li/></ul><ul id="b"><li/><li/><li/></ul></root>';

qp($xml)->find('ul:has(li)');               // before: (none)   after: a, b
qp($xml)->find('ul:has(li:nth-child(3))');  // before: (none)   after: b
qp($xml)->find('ul:not(:has(li))');         // before: a, b     after: (none)

What changed

• has() now reads its argument as a CSS 4 relative selector anchored to the element with :scope. It matches a descendant by default, a child for > li, the next sibling for + li, and a later sibling for ~ li. Each group is parsed and anchored once per argument and then cached.
• :not() and :matches() keep the old "is this element a match?" check, now in is(). isNot() used to call has().
• applySetFilter() had the same initialized-mode bug for :has() with a positional argument. has is no longer in Util::SELECTOR_ARGUMENT_PSEUDO_CLASSES, so :has() is never a set filter. Its argument is searched separately around each element, as in jQuery, so ul:has(li:eq(2)) matches every ul that has a third li. This matches remove().
• DOMTraverser::findGroups() runs a selector that is already parsed, and find() now calls it. SimpleSelector::isBlank() is new.

Supported relative forms

li, > li, + li, ~ li, chains such as + ul ~ q and ~ ol > li > span, and comma lists without a leading combinator (x, span). Parser limitations I did not fix:

• A leading combinator after the comma (li, > span) throws "two combinators in sequence", or parses wrongly.
• A comma list that contains parentheses (li:nth-child(3), span) loses its comma when the argument is rebuilt. This is close to Doubly nested parentheses in a selector throw "Recursion error detected" (e.g. li:not(:not(:nth-child(1)))) #120.

Behaviour change

Selectors using :has() used to match nothing, and now they match. :not(:has(...)) used to match everything. On self-matching, this engine and remove() now differ: the legacy engine counts the element itself, so remove('*:has(li.k)') includes the li.k. find() follows the spec and does not.

Performance

A descendant search is one traversal per candidate. + checks only the next sibling. ~ searches the later siblings one at a time and stops at the first match. When ~ has no match it costs O(siblings) for each candidate: li:has(~ b) over 2000 siblings takes about 4 s.

Sharing #119's sibling index

#119 (now on main) indexes sibling lists once per query and shares the index with nested traversers through DOMTraverser's 4th constructor argument. is() and findsAny() both pass $this, so every per-element and per-sibling search inside :has() reuses one index. Without that, li:has(~ li:nth-child(1)) over 400 siblings took about 9.2 s. With it, it takes about 0.26 s. PseudoClassTest::testHasQueriesShareTheSiblingIndex pins this for +, ~ and > arguments. Each case indexes exactly one sibling list, and all three fail without $this.

Rebased onto #72, #73 and #119

findsAny() searched each later sibling with an uninitialized find() that used to count the sibling itself. #73 made find() search descendants only, which broke 9 sibling-relative tests. findsAny() now takes a $self flag. A later sibling is tested itself first, and searched below only when the group continues to a child or descendant (~ ol > li > span). relativeSelector() now records whether a group starts at a sibling, rather than returning the anchor combinator.

Tests

• DOMQueryTest: a data provider of 24 find() cases, plus html5qp(), filter(), agreement with remove() on the issue's selectors, and :not() regression checks.
• PseudoClassTest: unit checks of has, matches and not.

Every new :has() test fails on main. vendor/bin/phpunit: 630 tests, 1797 assertions, 0 failures (1 skipped). lint and lint:min-php pass.

Fixes #121.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 98.30508% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 91.17%. Comparing base (6ae6acf) to head (9fa8255).
⚠️ Report is 13 commits behind head on main.

Files with missing lines	Patch %	Lines
src/CSS/DOMTraverser/PseudoClass.php	97.95%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##               main     #125      +/-   ##
============================================
+ Coverage     90.69%   91.17%   +0.48%     
- Complexity     1424     1459      +35     
============================================
  Files            25       26       +1     
  Lines          3170     3231      +61     
============================================
+ Hits           2875     2946      +71     
+ Misses          295      285      -10     

☔ 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.