REGRESSION: Incomplete repaint when block with clipping grows

• Component: Layout and Rendering
• Status: RESOLVED FIXED

See the attached testcase. In this case, since the block with the border grows and repaints before its container (which is clipping) grows, so in RenderObject::repaintAfterLayoutIfNeeded it sees that its old bounds are equal to its new bounds and it doesn't repaint. When the container repaints, it paints only the delta, so the old border of the inner block isn't erased.

Created attachment 5926 [details]
Testcase

[Alice Liu]

<rdar://problem/4483838>

Created attachment 7750 [details]
Possible fix

I think this patch does not add a lot of painting, and what it does add can be divided between
1) what's needed to fix the bug: the layer did not need layout, but ended up resizing because of a child.
2) addressing the fact that the selfNeedsLayout() condition in repaintAfterLayoutIfNeeded() is always false when called from the layer.

Created attachment 7751 [details]
Possible fix

Please see my previous comment.

Created attachment 7752 [details]
Possible fix

Sorry, uploaded the wrong file.

[Darin Adler]

Comment on attachment 7752 [details]
Possible fix

This sure looks good to me, but I think it's something Hyatt will want to review himself.

[Dave Hyatt]

Won't this cause way too much repainting?  Turn on paint flashing and try loading a huge file like:

http://lxr.mozilla.org/seamonkey/source/layout/base/nsCSSFrameConstructor.cpp

Do you see the whole window repainting now with this fix as chunks of data come in?

It seems like this fix will cause the root block's layer to repaint whenever new chunks of data come in that make the root block's layer grow.

(In reply to comment #7)
> It seems like this fix will cause the root block's layer to repaint whenever
> new chunks of data come in that make the root block's layer grow.

Oh boy, I overlooked that! I think instead of 
+        m_fullRepaintOnResize = m_object->selfNeedsLayout() || m_object->normalChildNeedsLayout();
it should be
+        m_fullRepaintOnResize = m_object->selfNeedsLayout() || m_object->hasOverflowClip() && m_object->normalChildNeedsLayout();

[Dave Hyatt]

Comment on attachment 7752 [details]
Possible fix

Even that change will still cause overflow blocks to repaint over and over as they load content.

(In reply to comment #9)
> (From update of attachment 7752 [details] [edit])
> Even that change will still cause overflow blocks to repaint over and over as
> they load content.

In what case, other than when adding content causes their layer to resize (which is what this bug is about)?

Created attachment 7810 [details]
Patch, including change log and manual test

Added the hasOverflowClip() check and renamed the flag per Hyatt's suggestion.

[Dave Hyatt]

Comment on attachment 7810 [details]
Patch, including change log and manual test

r=me

[Timothy Hatcher]

Landed in r13982.