Rewrite

[oliverklee]

While trying to cover the code with tests and to add native type declarations, I find the existing code increasingly hard to understand and to work with. Also, we're seeing feature requests and bug reports like #1170 that are hard to solve with the current architecture.

So I'd like to tackle a rewrite after the release of version 9 in these stages, closely following the algorithms, nomenclature and data structures as outlined in the CSS specs:

1. create a preprocessor to remove BOM, unify line endings etc.
2. use it in the legacy parser
3. create a charset detector
4. use it in the legacy processor
5. deprecate the functionality to manually set/override the charset
6. create a lexer that takes a preprocessed CSS string and returns a token stream
7. create a new parser and classes for the CSS "things"
8. create a new renderer
9. deprecate the old classes
10. remove the old classes in V10

What do you think, @sabberworm @JakeQZ ?

[JakeQZ]

I'm with you up to about item 4½. Beyond that, without knowing more detail or reading your mind, I have no idea.

I would generally suggest an Agile approach: small incremental changes rather than a wholesale redesign all at once.

[oliverklee]

I'd prefer that as well - and currently don't see how …

[oliverklee]

Note (mainly) to myself: We might want to copy/implement the CSS Object Model( CSSOM) when we rewrite the CSS classes.

[JakeQZ]

We might want to copy/implement the CSS Object Model( CSSOM) when we rewrite the CSS classes.

Perhaps not fully.

It seems that the official names of the various entities have morphed over the years. E.g. CSSStyleDeclaration represents what is now called a "declaration block". There's also the misnomer of a "ruleset" which until CSS3 didn't actually contain "rules". Or maybe in CSS2 it was named differently, or the now-called "declarations" within it were previously known as "rules".

This article explains some of the background and current issues.

This is a minefield of naming. Changing class names to represent what the object is currently officially known as may be a wasted journey, because it will change again in future. In CSSOM, CSSStyleDeclaration has not been deprecated in favour of a new CSSDeclarationBlock interface.

We should consider carefully whether renaming things is actually a beneficial change.

[JakeQZ]

6. create a lexer that takes a preprocessed CSS string and returns a token stream

I don't think the spec is clear enough to do this. It seems to be a make-it-up-as-you-go from the tech giants.