[ondrejmirtes]

TypeParser::isHtml() runs for every Name< to tell a generic from an HTML tag such as <br>. For list<string>, the text <string> looks like a tag, so it walked all remaining tokens looking for </string>. That's one walk to the end of the PHPDoc per generic, so PHPDocs with many generics took quadratic time.

TokenIterator now records, once per PHPDoc, the last token containing each /name>. isHtml() returns early when no such text is written at or after the current token. Every end tag the walk looks for contains /name>, so the early return can't change a result. The walk itself is unchanged. The new TokenIterator::hasEndTagCandidateAhead() is @internal.

Parsing a PHPDoc with this line repeated N times: * @return array{a: int, b?: list<string>}|(callable(int): void)|null

Lines	Before	After
100	32 ms	3.2 ms
200	124 ms	6.4 ms
400	507 ms	13 ms

Typical PHPDocs are unchanged within noise: a corpus of 5371 real-world PHPDocs takes 92–97 ms on both versions.

Verification:

• The ASTs are byte-identical to the previous parser on 10,742 corpus inputs (whole and truncated PHPDocs, with and without attributes).
• They're also identical on 60,000 random PHPDocs mixing generics with <br>/</br>-style tags, end tags before and after the <, and broken tags.
• make check passes.