Implement CSS parser

[kristoff-it]

We want to also provide autoindentation and diagnostics for CSS content.

To do so correctly, we need to implement a CSS parser which would need to follow the same principles of the main HTML parser:

• Spec compliant whenever possible, but break away whenever it makes sense (eg HTML parser marks implicitly closed tags as errors)
• Should be structured in away that makes it as easy as possible to validate its behavior against the CSS spec (eg the HTML tokenizer references the spec in every state)
• Keeps in mind performance and avoids unnecessary parts of the spec

If anybody is interested in taking this on, let's first discuss the idea a little.

[Arnau478]

I've written a CSS tokenizer and parser before. The one I wrote is not really suitable for SuperHTML, as it was intended for a web browser engine. However, I am willing to address this if you let me, making a new one. A few things I wanted to say:

• The CSS spec is not as weird and backwards-compatible as the HTML one, so having to break the spec shouldn't be a recurrent problem. In fact, the CSS grammar is designed to be compatible across all levels (versions), without the need to explicitly support it.
• As I mentioned, there is a core CSS grammar that is respected by all levels, but then each level can define its own (strictly subset) grammar. So, every valid CSS 2.1 stylesheet is valid according to the CSS core grammar, but not necessarily the other way around. Following the CSS 2.1 grammar can basically lead to more detailed parse errors that, with the core grammar, could be less descriptive or not even considered an error. Personally, I would go with CSS 2.1, as it's also a subset of CSS 1 (and thus backwards-compatible by design). As an example, the core CSS grammar knows about at-rules and can parse them all right, but it doesn't report errors for things like @meida instead of @media (@meida might exist in the future). The only real advantage of going with the core CSS grammar as opposed to CSS 2.1 is performance, but I doubt it makes a significant difference.

[kristoff-it]

Thank you for your interest. Yes I would concur that it's a good idea to support CSS2.1 and more in general I would be interested, over time, to have support for all the fancy stuff that can be done with CSS.

Here's how I would proceed:

• make a branch dedicated to the CSS parser
• implement a small part of the spec to support a small initial test case (eg h1 { color: red; })
• send a PR so I can review and make adjustments to the main structure