[wozhenbang2004]

When a class name ends in an escaped backslash, a following . or # starts a new selector node. splitWord() currently treats every delimiter immediately preceded by a backslash as escaped and merges the nodes.

Count consecutive backslashes before the delimiter so an even number leaves it active and an odd number still escapes it. The regression tests exercise both delimiters, escaped delimiters after a backslash, and lossless round trips.

The two new delimiter cases fail on the unmodified parser. npm test passed all 905 tests on Node.js 24, including the coverage thresholds.

[MoOx]

I merged current main into your branch rather than asking you to rebase, because the conflict came from our side: 7.1.8 hoisted the @keyframes percent test out of the per-index filter, since evaluating it for every . made a digit-prefixed word quadratic. Resolving in favour of your version alone would have put that regression back.

Your isEscaped is kept as is. I checked it separately and it stays linear overall: the backslash runs do not overlap, so the total work is bounded by their sum. At a constant 200 KB of input the cost is flat from 10 to 50 000 delimiters.

npm test passes with 985 tests, and I confirmed both properties: .a\\.b now parses as two classes instead of one, and parsing a digit-prefixed word stays flat at 3 ms from 12 000 to 48 000 characters.

Tell me if you would rather resolve it yourself and I will drop my merge commit.