Get rid of the existing bison-based parser, implement our own parser, fix bugs and improve extensibility in the process
Is there a spec for how a CSS parser should work? Can we produce a spec at the same time as we produce the implementation?
One major win here would be better error reporting.
There is nothing apart from the grammar plus prose combination the current specifications have. Having an HTML5-like specification for this would be really great.
I can already see a bug filed three years later: get rid of messy-crashy-buggy-slow manually maintained CSS parser code, replace it with autogenerated one :)
Turned on (see bug 165213).