Selector parser with built in methods for working with selector strings.
With npm do:
npm install postcss-selector-parser
const parser = require('postcss-selector-parser'); const transform = selectors => { selectors.walk(selector => { // do something with the selector console.log(String(selector)) }); }; const transformed = parser(transform).processSync('h1, h2, h3');
To normalize selector whitespace:
const parser = require('postcss-selector-parser'); const normalized = parser().processSync('h1, h2, h3', {lossless: false}); // -> h1,h2,h3
Async support is provided through parser.process and will resolve a Promise with the resulting selector string.
Please see API.md.
The parser walks the selector AST recursively, both when parsing and when serializing it back to a string (.toString()). In versions up to and including 7.1.1, a selector with extreme nesting — for example thousands of nested :not(...) — could recurse deeply enough to overflow the call stack and throw RangeError: Maximum call stack size exceeded, a potential denial-of-service when processing untrusted CSS.
This is now bounded by a maximum nesting depth (default: 256). Beyond that depth, parsing and serialization throw a regular, catchable Error at a predictable point instead of relying on the runtime hitting its stack limit. The default is far above any realistic selector, so it does not affect normal use.
Practical impact is low. The only attacker-controlled input is the selector string itself, which is now capped by the default limit. The limit is adjustable through the maxNestingDepth option, but that option is trusted configuration provided by the integrating code — it is never derived from the parsed CSS, so a malicious selector cannot change it:
// Tighten the limit when parsing untrusted input: parser().processSync(untrustedSelector, {maxNestingDepth: 128});
Raising maxNestingDepth to a very large value is an explicit, informed choice and can reintroduce the stack-overflow risk in environments with a small call stack (e.g. browser workers). The default is recommended unless you have a specific need.
MIT