[lprnmns]

Attribute values are decoded with decodeHTML, the rules for text content. Inside attributes, HTML handles legacy references without a ; differently: if the reference is consumed as part of an attribute, isn't terminated by ;, and the next character is = or an ASCII alphanumeric, it is left as-is. This mostly matters for URLs with query strings:

const { parse } = require('node-html-parser');

const a = parse('<a href="?x=1&notify=1&lang=en" title="&copy=2">x</a>').firstChild;
a.getAttribute('href'); // before: '?x=1¬ify=1&lang=en'  after: '?x=1&notify=1&lang=en'
a.getAttribute('title'); // before: '©=2'                 after: '&copy=2'

Browsers return the value after the change. I checked the new test markup with getAttribute() in Chrome, and all values match.

entities (already a dependency, ^8.0.0) has decodeHTMLAttribute for this. attrs, attributes and the attrs cache update in setAttribute now use it. Text content still uses decodeHTML, so <p>&copyx</p> remains ©x.

Tests: test/tests/attribute-entities.js. The two assertions about the attribute rule fail without the change, and the other two check that complete references in attributes and text content still decode. TEST_TARGET=dist mocha --recursive ./test/tests has 309 passing. The 2 Module Import failures happen with and without the change in my environment (no bun). I built with npx tsdown, and eslint src/nodes/html.ts passes.