How to Write Custom Syntax

PostCSS can transform styles in any syntax, and is not limited to just CSS. By writing a custom syntax, you can transform styles in any desired format.

Writing a custom syntax is much harder than writing a PostCSS plugin, but it is an awesome adventure.

There are 3 types of PostCSS syntax packages:


A good example of a custom syntax is SCSS. Some users may want to transform SCSS sources with PostCSS plugins, for example if they need to add vendor prefixes or change the property order. So this syntax should output SCSS from an SCSS input.

The syntax API is a very simple plain object, with parse & stringify functions:

module.exports = {
  parse:     require('./parse'),
  stringify: require('./stringify')


A good example of a parser is Safe Parser, which parses malformed/broken CSS. Because there is no point to generate broken output, this package only provides a parser.

The parser API is a function which receives a string & returns a Root or Document node. The second argument is a function which receives an object with PostCSS options.

const postcss = require('postcss')

module.exports = function parse (css, opts) {
  const root = postcss.root()
  // Add other nodes to root
  return root

For open source parser npm package must have postcss in peerDependencies, not in direct dependencies.

Main Theory

There are many books about parsers; but do not worry because CSS syntax is very easy, and so the parser will be much simpler than a programming language parser.

The default PostCSS parser contains two steps:

  1. Tokenizer which reads input string character by character and builds a tokens array. For example, it joins space symbols to a ['space', '\n '] token, and detects strings to a ['string', '"\"{"'] token.
  2. Parser which reads the tokens array, creates node instances and builds a tree.


Parsing input is often the most time consuming task in CSS processors. So it is very important to have a fast parser.

The main rule of optimization is that there is no performance without a benchmark. You can look at PostCSS benchmarks to build your own.

Of parsing tasks, the tokenize step will often take the most time, so its performance should be prioritized. Unfortunately, classes, functions and high level structures can slow down your tokenizer. Be ready to write dirty code with repeated statements. This is why it is difficult to extend the default PostCSS tokenizer; copy & paste will be a necessary evil.

Second optimization is using character codes instead of strings.

// Slow
string[i] === '{'

// Fast
const OPEN_CURLY = 123 // `{'
string.charCodeAt(i) === OPEN_CURLY

Third optimization is “fast jumps”. If you find open quotes, you can find next closing quote much faster by indexOf:

// Simple jump
next = string.indexOf('"', currentPosition + 1)

// Jump by RegExp
regexp.lastIndex = currentPosion + 1
next = regexp.lastIndex

The parser can be a well written class. There is no need in copy-paste and hardcore optimization there. You can extend the default PostCSS parser.

Node Source

Every node should have source property to generate correct source map. This property contains start and end properties with { line, column }, and input property with an Input instance.

Your tokenizer should save the original position so that you can propagate the values to the parser, to ensure that the source map is correctly updated.

Raw Values

A good PostCSS parser should provide all information (including spaces symbols) to generate byte-to-byte equal output. It is not so difficult, but respectful for user input and allow integration smoke tests.

A parser should save all additional symbols to node.raws object. It is an open structure for you, you can add additional keys. For example, SCSS parser saves comment types (/* */ or //) in node.raws.inline.

The default parser cleans CSS values from comments and spaces. It saves the original value with comments to node.raws.value.raw and uses it, if the node value was not changed.


Of course, all parsers in the PostCSS ecosystem must have tests.

If your parser just extends CSS syntax (like SCSS or Safe Parser), you can use the PostCSS Parser Tests. It contains unit & integration tests.


A style guide generator is a good example of a stringifier. It generates output HTML which contains CSS components. For this use case, a parser isn't necessary, so the package should just contain a stringifier.

The Stringifier API is little bit more complicated, than the parser API. PostCSS generates a source map, so a stringifier can’t just return a string. It must link every substring with its source node.

A Stringifier is a function which receives Root or Document node and builder callback. Then it calls builder with every node’s string and node instance.

module.exports = function stringify (root, builder) {
  // Some magic
  const string = decl.prop + ':' + decl.value + ';'
  builder(string, decl)
  // Some science

Main Theory

PostCSS default stringifier is just a class with a method for each node type and many methods to detect raw properties.

In most cases it will be enough just to extend this class, like in SCSS stringifier.

Builder Function

A builder function will be passed to stringify function as second argument. For example, the default PostCSS stringifier class saves it to this.builder property.

Builder receives output substring and source node to append this substring to the final output.

Some nodes contain other nodes in the middle. For example, a rule has a { at the beginning, many declarations inside and a closing }.

For these cases, you should pass a third argument to builder function: 'start' or 'end' string:

this.builder(rule.selector + '{', rule, 'start')
// Stringify declarations inside
this.builder('}', rule, 'end')

Raw Values

A good PostCSS custom syntax saves all symbols and provide byte-to-byte equal output if there were no changes.

This is why every node has node.raws object to store space symbol, etc.

All data related to source code and not CSS structure, should be in Node#raws. For instance, postcss-scss keep in Comment#raws.inline boolean marker of inline comment (// comment instead of /* comment */).

Be careful, because sometimes these raw properties will not be present; some nodes may be built manually, or may lose their indentation when they are moved to another parent node.

This is why the default stringifier has a raw() method to autodetect raw properties by other nodes. For example, it will look at other nodes to detect indent size and them multiply it with the current node depth.


A stringifier must have tests too.

You can use unit and integration test cases from PostCSS Parser Tests. Just compare input CSS with CSS after your parser and stringifier.