Recalc styles on resize if we have viewport units.

Recalc styles on resize if we have viewport units.

This patch forces style recalcs after resize operations if we have any viewport units.
This fixes the issue of having font-size and line-height properties not being updated
when we resize, but more importantly sets us up to resolve all viewport units at style
calc time and thus remove the various viewport unit types from Length objects, which
will fix various interactions with viewport units.

BUG=124331

[Timothy Loh, 2013-04-08 05:26:05]

Hi Eric,

Could you review this patch for me? You've looked at it previously at
https://bugs.webkit.org/show_bug.cgi?id=87846.

Thanks.

[eseidel, 2013-04-09 08:12:06]

This seems reasonable to me.  CCing other rendering reviewers as well. lgtm.

[esprehn, 2013-04-09 08:20:27]

This doesn't resolve all viewport units at recalc time yet right?

I'm concerned this is going to mean awful performance if your viewport changes
size, but it's probably okay to start. Perhaps add a fixme about this.

https://codereview.chromium.org/13771002/diff/1/Source/WebCore/page/FrameView...
File Source/WebCore/page/FrameView.cpp (right):

https://codereview.chromium.org/13771002/diff/1/Source/WebCore/page/FrameView...
Source/WebCore/page/FrameView.cpp:1115: document->scheduleForcedStyleRecalc();
This is a big hammer. It's going to mean pretty crushing performance if you have
a big page with viewport units and resize it. ex. resizing the HTML5 spec window
if it used even a single vh or vw unit.

https://codereview.chromium.org/13771002/diff/1/Source/WebCore/page/FrameView...
Source/WebCore/page/FrameView.cpp:1124: // the layout beats any sort of style
recalc update that needs to occur, and if we
"or if" is probably more clear. I had to read this a few times to understand.

[esprehn, 2013-04-09 08:34:22]

eseidel and I discussed this and I think we want to not do this change until we
can figure out a way to avoid discarding the whole style resolver. It's a nasty
foot gun where adding a viewport unit suddenly makes your entire page super
expensive on window resize, or even worse, on orientation change on mobile.

not lgtm.

[Timothy Loh, 2013-04-10 01:10:08]

Yes, resolving viewport units at recalc time would be a separate (much larger
but mostly straightforward) change. Doing it this way seems to me to be the
cleanest way of getting this done, and looks much nicer than my first approach
of resolving all the units at layout time (see
https://bugs.webkit.org/show_bug.cgi?id=91554) -- and in any case, we'd still
need to do at least some sort of partial style recalc to fix the linked bug
about viewport dependent font-size.

I tend to think that it's better to have things correct and slow rather than
incorrect and fast, but if this isn't the approach that you guys think is right
then perhaps you could make some suggestions on how to make it better? :)

[esprehn, 2013-04-12 03:14:25]

On 2013/04/10 01:10:08, Timothy Loh wrote:
> .. 
> I tend to think that it's better to have things correct and slow rather than
> incorrect and fast, but if this isn't the approach that you guys think is
right
> then perhaps you could make some suggestions on how to make it better? :)

The issue is one of having big foot guns. Right now we're wrong sometimes, but
every page that uses vh and vw isn't super slow.

I'd want to see a long term solution so this "make my page slow" button wasn't
in the tree forever before allowing this kind of change.