Re: [CSSWG] Minutes and Resolutions 2009-08-05

On Thursday 2009-08-06 18:41 +0200, Bert Bos wrote:
> I don't think the comment parsing is a problem for a modern computer 
> either. If the thing you're parsing is indeed a style sheet with a 
> typo, rather than a stream of random bytes whose CSS parse tree doesn't 
> interest anybody anyway, then buffering the unclosed comment will maybe 
> cost you a few tens of kilobytes of memory. Not something to worry 
> about. (If you already don't have enough memory to store the style 
> sheet as a text string, you're unlikely to have enough for its DOM...)

The real issue isn't the buffering; it's getting consistent
behavior (including making different parts of the spec say the same
thing).

> I (reluctantly) agreed to the change, but I think I only agreed to 
> change the grammar of comments. That was also where some browsers we 
> tested differed from the spec. Most browsers did not exhibit the same 
> bug for unclosed URLs. (Which makes sense: Zack's change allows you to 

For what it's worth, after implementing a fix for Zack's testcase
yesterday, I'm no longer confident that it's really testing
tokenization; in Mozilla's case what it was really testing was
whether we were doing proper parenthesis-matching in the parser, not
whether the tokenizer backtracked or not.

(That said, we actually implement the tokenization using a strategy
rather like what I described in
http://lists.w3.org/Archives/Public/www-style/2009Aug/0098.html .)


I'm not sure I want the url() change anymore either (though I think
the comments part is right).  But I'm also not confident the current
spec is internally consistent, though I'd like to think about it
further.

-David

-- 
L. David Baron                                 http://dbaron.org/
Mozilla Corporation                       http://www.mozilla.com/

Received on Thursday, 6 August 2009 17:14:18 UTC