Bug 88171 – CSS 2.1 failure: floats-wrap-top-below-inline-* fail

• Component: CSS
• Status: RESOLVED FIXED

[Robert Hogan]

..

[Robert Hogan]

Created attachment 148009 [details]
Patch

[WebKit Review Bot]

Comment on attachment 148009 [details]
Patch

Attachment 148009 [details] did not pass chromium-ews (chromium-xvfb):
Output: http://queues.webkit.org/results/12975223

New failing tests:
css1/formatting_model/floating_elements.html

[WebKit Review Bot]

Created attachment 148017 [details]
Archive of layout-test-results from ec2-cr-linux-03

The attached test failures were seen while running run-webkit-tests on the chromium-ews.
Bot: ec2-cr-linux-03  Port: <class 'webkitpy.common.config.ports.ChromiumXVFBPort'>  Platform: Linux-2.6.35-28-virtual-x86_64-with-Ubuntu-10.10-maverick

[Robert Hogan]

Created attachment 148019 [details]
Patch

[WebKit Review Bot]

Comment on attachment 148019 [details]
Patch

Attachment 148019 [details] did not pass chromium-ews (chromium-xvfb):
Output: http://queues.webkit.org/results/12978108

New failing tests:
css1/formatting_model/floating_elements.html

[WebKit Review Bot]

Created attachment 148023 [details]
Archive of layout-test-results from ec2-cr-linux-04

The attached test failures were seen while running run-webkit-tests on the chromium-ews.
Bot: ec2-cr-linux-04  Port: <class 'webkitpy.common.config.ports.ChromiumXVFBPort'>  Platform: Linux-2.6.35-28-virtual-x86_64-with-Ubuntu-10.10-maverick

[Robert Hogan]

Created attachment 149588 [details]
New result in floating_elements

[Robert Hogan]

Created attachment 149589 [details]
Current rendering of change to floating_elements

[Robert Hogan]

Created attachment 149591 [details]
Patch

[Robert Hogan]

Comment on attachment 149591 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=149591&action=review

> LayoutTests/ChangeLog:64
> +        * platform/chromium-win/css1/formatting_model/floating_elements-expected.txt:
> +        This is a progression due to lineboxes now avoiding floats they overlap.

The change in rendering that causes the change in results is viewable here: https://bugs.webkit.org/attachment.cgi?id=149588
You can compare that with the current rendering: https://bugs.webkit.org/attachment.cgi?id=149589

[Eric Seidel (no email)]

Comment on attachment 149591 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=149591&action=review

I think the change is OK.  Need answers to a couple questions first:

> Source/WebCore/rendering/RenderBlock.cpp:3958
> +    if ((m_lowValue >= interval.low() && m_lowValue < interval.high())
> +        || (m_lowValue < interval.low() && m_highValue > interval.high())
> +        || (m_highValue > m_lowValue && m_highValue > interval.low() && m_highValue <= interval.high())) {

This is just a non-intersecting range check, right?  Don't we have a cleaner way to write this?  Maybe with a helper?  I guess we don't really have a "range" object.

> Source/WebCore/rendering/RenderBlock.cpp:3963
> +        ASSERT(r->isPlaced() && ((m_renderer->pixelSnappedLogicalTopForFloat(r) <= m_lowValue && m_renderer->pixelSnappedLogicalBottomForFloat(r) > m_lowValue)
> +            || (m_lowValue < m_renderer->pixelSnappedLogicalTopForFloat(r) && m_highValue > m_renderer->pixelSnappedLogicalBottomForFloat(r))
> +            || (m_highValue > m_lowValue && m_renderer->pixelSnappedLogicalTopForFloat(r) < m_highValue && m_highValue <= m_renderer->pixelSnappedLogicalBottomForFloat(r))));

I'm not sure I understand this, but it looks similar to above, and likely could use the same helper function?

> Source/WebCore/rendering/RenderBlock.cpp:4018
> +        FloatIntervalSearchAdapter<FloatingObject::FloatLeft> adapter(this, roundToInt(logicalTop), roundToInt(logicalTop + logicalHeight), left, heightRemaining);

Should these be rounded?  Do we need to snap these?  Have you asked Emil/Levi or consulted http://trac.webkit.org/wiki/LayoutUnit ?

> Source/WebCore/rendering/RenderBlock.h:132
> +    LayoutUnit availableLogicalWidthForLine(LayoutUnit position, bool firstLine, RenderRegion* region, LayoutUnit offsetFromLogicalTopOfFirstPage, LayoutUnit logicalHeight = 0) const

A height of 0 seems like an odd default.  Why does this need a default?

> Source/WebCore/rendering/RenderBlockLineLayout.cpp:134
> +    m_left = m_block->logicalLeftOffsetForLine(height, m_isFirstLine, logicalBottom);
> +    m_right = m_block->logicalRightOffsetForLine(height, m_isFirstLine, logicalBottom);

Now your passing bottom where other places you pass top, height.  Is this intentional?

[Robert Hogan]

Created attachment 150336 [details]
Patch

[Robert Hogan]

(In reply to comment #11)
> (From update of attachment 149591 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=149591&action=review
> 
> I think the change is OK.  Need answers to a couple questions first:
> 
> > Source/WebCore/rendering/RenderBlock.cpp:3958
> > +    if ((m_lowValue >= interval.low() && m_lowValue < interval.high())
> > +        || (m_lowValue < interval.low() && m_highValue > interval.high())
> > +        || (m_highValue > m_lowValue && m_highValue > interval.low() && m_highValue <= interval.high())) {
> 
> This is just a non-intersecting range check, right?  Don't we have a cleaner way to write this?  Maybe with a helper?  I guess we don't really have a "range" object.

I've improved the code but not the quality of the check. I'm sure there's a way to do it more concisely but I can't see it.

> 
> > Source/WebCore/rendering/RenderBlock.cpp:4018
> > +        FloatIntervalSearchAdapter<FloatingObject::FloatLeft> adapter(this, roundToInt(logicalTop), roundToInt(logicalTop + logicalHeight), left, heightRemaining);
> 
> Should these be rounded?  Do we need to snap these?  Have you asked Emil/Levi or consulted http://trac.webkit.org/wiki/LayoutUnit ?

roundToInt(logicalTop) is current code - so I followed it here. From my reading of the wiki page roundToInt() seems right here.

> 
> > Source/WebCore/rendering/RenderBlock.h:132
> > +    LayoutUnit availableLogicalWidthForLine(LayoutUnit position, bool firstLine, RenderRegion* region, LayoutUnit offsetFromLogicalTopOfFirstPage, LayoutUnit logicalHeight = 0) const
> 
> A height of 0 seems like an odd default.  Why does this need a default?

All of the prototypes here have a bunch of callers that don't (at the moment care) about the logical height of the child - so the default gets set to zero and this gets added to the logical top of the child when checking its top and 'bottom' against floats.

> 
> > Source/WebCore/rendering/RenderBlockLineLayout.cpp:134
> > +    m_left = m_block->logicalLeftOffsetForLine(height, m_isFirstLine, logicalBottom);
> > +    m_right = m_block->logicalRightOffsetForLine(height, m_isFirstLine, logicalBottom);
> 
> Now your passing bottom where other places you pass top, height.  Is this intentional?

Yeah - logicalHeight is a slightly misleading name in a lot of RenderBlock stuff - here it means the logicalTop of the current child or line, which is why I went for logicalBottom to express the height/bottom of the line. But I think you're right that that just confuses things a bit more. I've changed it back so it's the same color of confusing everywhere.

[Eric Seidel (no email)]

Comment on attachment 150336 [details]
Patch

LGTM.  I think we should consider adding a LineRange class in a follow-up patch.  I suspect with a little object orientation we could make this read a lot nicer. :)

[WebKit Review Bot]

Comment on attachment 150336 [details]
Patch

Clearing flags on attachment: 150336

Committed r121789: <http://trac.webkit.org/changeset/121789>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.