Pass 1: Tokenize

First, we go through and parse the CSS document into a series of tokens internally. The tokenizing process does not check to make sure that you've got syntactically valid CSS, it just breaks up the text into a stream of tokens suitable for processing by the subsequent stages.

Pass 2: Collapse

We then march through the token list and collapse certain tokens down to their smallest possible representation. If they're still included in the final results we only want to include them at their shortest.

Whitespace

Runs of multiple whitespace characters are reduced down to a single whitespace character. If the whitespace contains any "end of line" (EOL) characters, then the end result is the first EOL character encountered. Otherwise, the result is the first whitespace character in the run.

Comments

Comments implementing the "Mac/IE Comment Hack" are collapsed down to the smallest possible comment that would still implement the hack ("/*\*/" to start the hack, and "/**/" to end it).

Zero Units

Zero Units (e.g. "0px") are reduced down to just "0", as the CSS specification indicates that the unit is not required when its a zero value.

Pass 3: Pruning

We then go back through the token list and prune and remove un-necessary tokens.

Whitespace

Wherever possible, whitespace is removed; before+after comment blocks, and before+after various symbols/sigils.

Comments

Comments that either (a) are needed to implement the "Mac/IE Comment Hack", or that (b) contain the word "copyright" in them are preserved. All other comments are removed.

Symbols/Sigils

Semi-colons that are immediately followed by a closing brace (e.g. ";}") are removed; semi-colons are needed to separate multiple declarations, but aren't required at the end of a group.

Everything else

We keep everything else; identifiers, quoted literal strings, symbols/sigils, etc.

Pass 4: Re-assembly

Lastly, we go back through the token list and re-assemble it all back into a single CSS string, which is then returned back to the caller.