[deadstrobe5]

Purpose / Goal

Long attribute values are slow to parse. tagExpWithClosingIndex walks every character of a quoted value only to find its closing quote, and attrsRegx then reads the value with a lazy [\s\S]*? up to a backreference, which V8 runs much slower than a negated character class. SVG is the common case: path d attributes are often tens of KB.

This PR jumps over a quoted value with indexOf of its own quote character, and matches values with "([^"]*)"|'([^']*)' (the value now comes from group 3 or 4 through a small attrValue helper). Both stop at the first matching quote, as before, so the same strings reach entity decoding and the value processors. An unterminated quote still makes tagExpWithClosingIndex return undefined, and tabs inside a value are still left alone.

Same-process A/B on Node 24, ignoreAttributes: false, median of 7 alternating rounds:

input	before	after
SVG, 1 path of 2000 points (30 KB)	0.79 ms	0.16 ms
SVG, 20 such paths (592 KB)	14.5 ms	3.6 ms
2000 elements with 4 short attributes (122 KB)	38.9 ms	38.4 ms

The new spec covers long values that hold the other quote, >, /> and tabs, and passes on master too. A differential fuzz against master (random documents with mixed and unterminated quotes, tabs, boolean and unquoted attributes, CRLF, 9 option sets, 54,000 comparisons) gave identical output and identical errors.

Type

• Bug Fix
• Refactoring / Technology upgrade
• New Feature