Fixes #1910

Problem

Complex CSS selectors with deeply nested pseudo-classes were being parsed incorrectly, causing wrong query results. For example:

.emXRrt:not(:has(> tbody > tr:not([data-ignore-row])))>thead>tr>th:nth-child(2)

The issue was that the custom regex-based parser only handled one level of nested parentheses.

Solution

Replaced the custom regex-based CSS selector parser with parsel-js, a battle-tested CSS selector parser by Lea Verou.

Benefits:

• Handles arbitrarily nested parentheses correctly
• More spec-compliant CSS selector parsing
• Well-maintained external library (~5KB minified)
• Reduces custom parsing code complexity
• Handles edge cases we may not have considered

Implementation Details

The new parser:

1. Pre-processes selectors to handle escaped characters that parsel-js can't handle (e.g., \:, \#, \&, \[, \])
2. Uses parsel-js to parse the selector into an AST
3. Converts the AST to the existing SelectorItem format
4. Restores escaped characters in the final output

Edge cases handled:

• Escaped characters in class names, IDs, and attribute names
• Empty attribute values ([attr=""])
• Brackets in attribute values ([attr="bracket[]bracket"])
• Space before combinators in :has() (e.g., :has( +h2))

Testing

Added two test cases for issue #1910:

1. The exact selector from the issue report with a table structure
2. A simpler test case demonstrating :not(:has(> .child:not([attr]))) parsing

Updated tests for stricter CSS compliance:

• Invalid selectors (.before:, .before&after, button:not([type]) now throw errors instead of returning empty results
• Fixed test for escaped colon in attribute selector to use correct attribute name

Changes

• packages/happy-dom/package.json: Added parsel-js dependency
• packages/happy-dom/src/query-selector/SelectorParser.ts: Rewrote to use parsel-js for parsing
• packages/happy-dom/test/query-selector/QuerySelector.test.ts: Added tests for issue Error when parsing complex CSS has expression #1910 and updated edge case tests