Improved parse error handling for CSSMQ.

Improved parse error handling for CSSMQ.

Implemented media query error handling as specified in CSS Media Queries.

As part of that, implemented a stack that keeps track of matching brackets to
do correct error handling for matching brackets/parentheses as specified in
CSS2.1 and Media Queries.

Matching brackets is only done for Media Queries error handling for now to
avoid the patch getting too big, but it should be simple to implement better
error handling for other parts of the grammar on top of this.

Also, the bracket stack emits unmatched brackets at the end of the token
stream to handle unexpected end-of-file as specified in CSS2.1.

Do recognize "and(" as function tokens. The existing code erroneously split
it into "and" and "(" inside media queries.

This patch does not address HTML5 compliant parsing of media attributes as
media query lists.

BUG=236302

[SeRya, 2013-04-29 16:32:45]

You introduce new method for handling unmatched braces while other parts of
grammar use another - the closing_brace nterm (in spite it's used
inconsistently).

Is there any reason why closing_brace wouldn't work here as well?

[rune, 2013-04-30 09:42:27]

On 2013/04/29 16:32:45, SeRya wrote:
> You introduce new method for handling unmatched braces while other parts of
> grammar use another - the closing_brace nterm (in spite it's used
> inconsistently).
> 
> Is there any reason why closing_brace wouldn't work here as well?

closing_brace doesn't handle balancing of parentheses/braces, so it doesn't
really work well. Also, I don't need to handle the EOF token, and closing_brace
probably won't need it either since unmatched parentheses/braces are emitted at
end of input with this patch. If this patch is accepted, I'll go ahead and fix
recovery for blocks and declarations too.

[SeRya, 2013-04-30 12:28:06]

On 2013/04/30 09:42:27, rune wrote:
> On 2013/04/29 16:32:45, SeRya wrote:
> > You introduce new method for handling unmatched braces while other parts of
> > grammar use another - the closing_brace nterm (in spite it's used
> > inconsistently).
> > 
> > Is there any reason why closing_brace wouldn't work here as well?
> 
> closing_brace doesn't handle balancing of parentheses/braces, so it doesn't
> really work well. Also, I don't need to handle the EOF token, and
closing_brace
> probably won't need it either since unmatched parentheses/braces are emitted
at
> end of input with this patch. If this patch is accepted, I'll go ahead and fix
> recovery for blocks and declarations too.

I feel that braces/parentheses matching should be fixed separately from other
issues.

Apparently closing_brace does handle balancing braces where it's used
consistently (like in declaration_list). If it is not I'd appreciate if you'd
find an example where it doesn't.

Of cause it doesn't handle balancing parentheses. However it could be easily
adjusted to do so. Something like "closing_parentheses: ')' | TOK_EOF";

[rune, 2013-05-02 07:58:38]

On 2013/04/30 12:28:06, SeRya wrote:
> On 2013/04/30 09:42:27, rune wrote:
> > On 2013/04/29 16:32:45, SeRya wrote:
> > > You introduce new method for handling unmatched braces while other parts
of
> > > grammar use another - the closing_brace nterm (in spite it's used
> > > inconsistently).
> > > 
> > > Is there any reason why closing_brace wouldn't work here as well?
> > 
> > closing_brace doesn't handle balancing of parentheses/braces, so it doesn't
> > really work well. Also, I don't need to handle the EOF token, and
> closing_brace
> > probably won't need it either since unmatched parentheses/braces are emitted
> at
> > end of input with this patch. If this patch is accepted, I'll go ahead and
fix
> > recovery for blocks and declarations too.
> 
> I feel that braces/parentheses matching should be fixed separately from other
> issues.

This patch is about getting the parse error handling for media queries right. I
need to balance braces/parentheses to get it done properly.

> Apparently closing_brace does handle balancing braces where it's used
> consistently (like in declaration_list). If it is not I'd appreciate if you'd
> find an example where it doesn't.
> 
> Of cause it doesn't handle balancing parentheses. However it could be easily
> adjusted to do so. Something like "closing_parentheses: ')' | TOK_EOF";

You still have the problem of unmatched braces/parentheses. It's probably
possible, but awkward, to try to solve that in the grammar. WebKit/Blink-based
browsers don't get this right:

<style>div { ( } div { color: red; }</style>
<div>This text should not be red</div>

[SeRya, 2013-05-05 10:10:41]

On 2013/05/02 07:58:38, rune wrote:
> On 2013/04/30 12:28:06, SeRya wrote:
> > On 2013/04/30 09:42:27, rune wrote:
> > > On 2013/04/29 16:32:45, SeRya wrote:
> > > > You introduce new method for handling unmatched braces while other parts
> of
> > > > grammar use another - the closing_brace nterm (in spite it's used
> > > > inconsistently).
> > > > 
> > > > Is there any reason why closing_brace wouldn't work here as well?
> > > 
> > > closing_brace doesn't handle balancing of parentheses/braces, so it
doesn't
> > > really work well. Also, I don't need to handle the EOF token, and
> > closing_brace
> > > probably won't need it either since unmatched parentheses/braces are
emitted
> > at
> > > end of input with this patch. If this patch is accepted, I'll go ahead and
> fix
> > > recovery for blocks and declarations too.
> > 
> > I feel that braces/parentheses matching should be fixed separately from
other
> > issues.
> 
> This patch is about getting the parse error handling for media queries right.
I
> need to balance braces/parentheses to get it done properly.
> 
> > Apparently closing_brace does handle balancing braces where it's used
> > consistently (like in declaration_list). If it is not I'd appreciate if
you'd
> > find an example where it doesn't.
> > 
> > Of cause it doesn't handle balancing parentheses. However it could be easily
> > adjusted to do so. Something like "closing_parentheses: ')' | TOK_EOF";
> 
> You still have the problem of unmatched braces/parentheses. It's probably
> possible, but awkward, to try to solve that in the grammar. WebKit/Blink-based
> browsers don't get this right:
> 
> <style>div { ( } div { color: red; }</style>
> <div>This text should not be red</div>

There is my prototype what handles this issue in the grammar. It doesn't look
awkward for me. Have a look please: https://codereview.chromium.org/14926003

[rune, 2013-05-06 09:23:07]

On 2013/05/05 10:10:41, SeRya wrote:
> There is my prototype what handles this issue in the grammar. It doesn't look
> awkward for me. Have a look please: https://codereview.chromium.org/14926003

Yeah, that looks ok. Can I take your approach into this patch then, and fix the
mq syntax at the same time?

I think I'll split out the "and(" change since it's a separate issue.

[rune, 2013-05-08 15:18:53]

On 2013/05/06 09:23:07, rune wrote:
> On 2013/05/05 10:10:41, SeRya wrote:
> > There is my prototype what handles this issue in the grammar. It doesn't
look
> > awkward for me. Have a look please: https://codereview.chromium.org/14926003
> 
> Yeah, that looks ok. Can I take your approach into this patch then, and fix
the
> mq syntax at the same time?
> 
> I think I'll split out the "and(" change since it's a separate issue.

Superseeded by 14620012