cssdalek

Exterminate unused CSS for major wins.

Why?

Often you will find yourself wanting to use a CSS framework, like Bootstrap or Materialize etc, only to discover that you are including large swaths of unused CSS.

cssdalek helps you drop these unused CSS bits, often making the CSS you end up serving your audience a whole lot smaller.

How?

Usage looks like this:

cssdalek \
  --css 'example/in-*.css' \
  --html 'example/*.html' > example/min.css

This uses the HTML extractor, and will work well if you can point it to all your markup (you can use the flags multiple times).

Words Extractor

If you're using dynamic templates, and/or JavaScript, then you can use the naive Words extractor, which assumes any word that exists can be a tag, classname or attribute. This will get you pretty far. For example:

cssdalek \
  --css 'example/in-*.css'
  --word 'example/*.tpl'
  --word 'example/*.js' > example/min.css

Includes

Finally, this tool can't recognize or detect dynamically created classnames etc. So you can explicitly tell it about such cases either by providing regular expressions that matches classnames or IDs, or by providing a full selector (which can include tag, attr etc). For example:

cssdalek \
  --css 'example/in-*.css' \
  --include-class '^foo' \
  --include-id '^bar.*baz$' \
  --include-selector '#bar .foo[type=text]' \
  --word 'example/*.tpl' \
  --word 'example/*.js' > example/min.css

Also remember all of these can be combined. Some HTML files, some using the word tokenizer, and others via the explicit includes.

Speed

There are alternatives to this tool that provide the same end result. Possibly a better, more accurate stripping of unused rules. It's very much possible that running your application in a browser will let you really see what rules can be dropped.

But that is slow. And so some of these other tools are also slow. So an important goal of this tool is not to be slow. We'll have to balance speed with accuracy.

Accuracy

Accuracy is the flip side of speed (and memory consumption). We're aiming for pretty good ™ accuracy. We're not going to store every HTML page in memory and run every selector like a browser, for example. But we want to drop as much as we can to actually make this tool useful.

FAQ

Descendant, child and sibling selectors are all considered the same: "an and set". For these selectors, if all the target nodes exist anywhere, we will include the selector. That is, the relationships are not actually checked for.
Attribute selectors are included if the attribute name is found. The value and type of operation is ignored.
Psuedo elements and children are essentially ignored, and only the rest of the selector determines usage.

TODO

Generate source mapping
Tables

daaku/cssdalek