725664 - UpdateOverflow handling is wrong for transforms

Status: RESOLVED WORKSFORME

(Core :: Layout, defect)

Description

[Mats Palmgren (inactive)]

Attachment: Testcase

The wallpaper in bug 722325 is wrong.  It causes the overflow areas
to monotonically grow until there is a reflow.

Comment 1

[Mats Palmgren (inactive)]

I think the problem is related to IsIgnoringViewportClipping() being true.
When scrolling the viewport I'm seeing transform style changes with a 1px diff,
they seem synthetic in the sense they don't appear to correspond to the scroll
position but just used to invalidate, or something.


I/Gecko   (14954): ScrollVisual:  blitting  active  IsIgnoringViewportClipping 
D/GeckoLayerController(14954): scrollBy: v=RectF(0.0, 6284.0894, 720.0, 7322.0933) p=(720.0,9704.0) z=2.0 o=0.0,0.0
D/GeckoLayerController(14954): scrollBy: v=RectF(0.0, 6295.204, 720.0, 7333.208) p=(720.0,9704.0) z=2.0 o=0.0,0.0
I/Gecko   (14954): mSpecifiedTransform '
I/Gecko   (14954): translate(0px, 101px) scale(2)' 
I/Gecko   (14954):  !=  aOther.mSpecifiedTransform '
I/Gecko   (14954): translate(0px, 100px) scale(2)'
I/Gecko   (14954): mSpecifiedTransform '
I/Gecko   (14954): translate(0px, 101px) scale(2)' 
I/Gecko   (14954):  !=  aOther.mSpecifiedTransform '
I/Gecko   (14954): translate(0px, 100px) scale(2)'


I/Gecko   (14954): AndroidGeckoEvent: 0x419334d0 : 20
I/Gecko   (14954): ScrollVisual:  blitting  active  IsIgnoringViewportClipping 
I/Gecko   (14954): mSpecifiedTransform '
I/Gecko   (14954): translate(0px, 100px) scale(2)' 
I/Gecko   (14954):  !=  aOther.mSpecifiedTransform '
I/Gecko   (14954): translate(0px, 99px) scale(2)'
I/Gecko   (14954): mSpecifiedTransform '
I/Gecko   (14954): translate(0px, 100px) scale(2)' 
I/Gecko   (14954):  !=  aOther.mSpecifiedTransform '
I/Gecko   (14954): translate(0px, 99px) scale(2)'

Comment 2

[Mats Palmgren (inactive)]

... and ViewportFrame / SubdocumentFrame has special overflow stuff too:
http://mxr.mozilla.org/mozilla-central/source/layout/generic/nsSubDocumentFrame.cpp#613
http://mxr.mozilla.org/mozilla-central/source/layout/generic/nsViewportFrame.cpp#281

Comment 3

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

The patch in bug 720987 that was partially backed out is also wrong, as I describe in bug 984226.

Comment 4

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I think this should be fixed as I describe in bug 984226 comment 3.

Comment 5

[Mats Palmgren (inactive)]

We should land the testcase here as a reftest if it wasn't done in some other bug.

Comment 6

[Daniel Holbert [:dholbert]]

What was the original buggy rendering of the testcase here? (comment 0 said the overflow areas
monotonically grow until there's a reflow, but that doesn't sound like something that would be visible in a reftest)

I'm asking both w.r.t. comment 5 and w.r.t. bug 1131371 comment 10. (I'm not sure how to tell if I'm regressing this testcase, aside from using GDB and trying to inspect the overflow areas.)

Comment 7

[Daniel Holbert [:dholbert]]

(oh, maybe the buggy rendering was that scrollbars were triggered? I can see how too-large overflow areas would cause that, & that does seem reftest-able.)

Comment 8

[Mats Palmgren (inactive)]

Oh, it was very visible, the outline was repainted with an offset outward for each run
in the interval until it hit the viewport at which point a reflow was triggered and the
overflow areas were reset and it started all over again.  So, just a couple of doTest()
calls instead of the setInterval() would catch it in a reftest I would think.