inikulin / parse5 / 42259953270

96%

master: 96%

Pull #1922

github

fix(tokenizer): bound the duplicate-attribute check for wide tags

Checking every new attribute name against every earlier one makes a tag with many attributes quadratic (#1901). Tags with up to 64 attributes keep the direct scan; once a 65th distinct name is accepted, the names are collected in a Set that is used for the rest of the tag and dropped when the tag is emitted. Duplicate semantics (first wins, parse error) and attribute order are unchanged. The field starts as `null`, so a wide tag does not add a property to the tokenizer; adding one made later tokenizing in the same process 2–5% slower.

The test counts reads of earlier attributes for 2,000 names and checks that the next tag does not see the wide tag's names.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

553 of 556 branches covered (99.46%)

13 of 13 new or added lines in 1 file covered. (100.0%)

3925 of 4104 relevant lines covered (95.64%)

54266.42 hits per line