[theRizwan]

The bug

A | straight after a pseudo reuses the pseudo's name as its namespace, so the name is printed twice:

parser().processSync(":hover|a");    // ":hoverhover|a"
parser().processSync("a:hover|b");   // "a:hoverhover|b"
parser().processSync("::before|*");  // "::beforebefore|*"

Every parse and print adds another copy, so running the output back through the parser keeps growing it: :hovernshoverns|a, then :hovernshovernshovernshoverns|a.

pseudo() consumes the word hover through splitWord(). The loop then reaches the | and calls namespace() through combinator(), and namespace() looks back at the previous token, which is that same hover word, and uses it again as the prefix.

The fix

When word() or universal() see a | next, the token before it is still theirs, so the lookback is right. When the | is reached from the main loop instead, the token before it has already been consumed by something else, and only a skipped & (from nesting()) can still be a prefix. namespace() now takes a prefixed flag, and combinator() passes false.

:hover|a now parses as the pseudo followed by |a, a tag with an empty namespace. That is what [x]|a already gives, since ] is not a prefix token either.

Verification

Two tests in namespaces.mjs, one for |b after a pseudo-class and one for |* after a pseudo-element. All four of their checks fail on main.

I also ran a lossless round trip over 50,000 random selectors built from common fragments. Lossy output that changes when printed and parsed again went from 127 cases to 21, and none of the 21 involve a pseudo. They come from the Sass and Less interpolation spelling.

npm test is green: lint, type check, and coverage gates.