Under the hood, the software processes are split into a number of stages:
The input HTML is loaded into the tokeniser as a string, and a regular expression splits it up into categorised tokens.
The tokens are passed to the parser which then loops through and consumes each token through an object based Finite State Machine to create an internal object structure that represents the document. This enables irregular tokens to be ignored and the document to be parsed with more reliability.
Once parsed, the document will contain an array of child objects each representing the node type in the original document. The tag type can contain its own child objects and so on.
Minification is performed by each object on its own structure, the command passed down each level from the one above.
Each object has its own minification process, and sometimes (Such as the text object) references it siblings or parent through its parent object.
As an example of a process, when whitespace is removed from the document, each text object will remove non significant whitespace from its content property.
The compilation process reconstructs the HTML from its object representation. Each object generates itself as a string, and then requests its children generate themselves. The result is all concatenated together and either output as a string or saved to the requested location.