1238534 - Assertion failure: IsAncestor(aClip1, aClip2) || IsAncestor(aClip2, aClip1) (one of the scroll clips must be an ancestor of the other)

Status: RESOLVED FIXED

(Core :: Graphics: Layers, defect)

Description

[Carsten Book [:Tomcat]]

Attachment: bughunter crash stack

found via bughunter on http://themes.goodlayers.com/thekeynote/ :

Assertion failure: IsAncestor(aClip1, aClip2) || IsAncestor(aClip2, aClip1) (one of the scroll clips must be an ancestor of the other)

i personally was not able to reproduce on a windows 7 vm trunk debug build based on mc-tip but i got on that page:

[4392] WARNING: '!temp', file c:/Users/mozilla/debug-build/mozilla-central/gfx/layers/basic/BasicCompositor.cpp, line 481
[GFX3-]: Surface width or height <= 0!
[GFX3-]: Surface width or height <= 0!
Crash Annotation GraphicsCriticalError: |[0][GFX1-]: Failed to allocate a surface due to invalid size (DSS) Size(0,0) (t=362.536)|[7941][GFX1-]: Failed to allocate a surf
ace due to invalid size (DSS) Size(0,0) (t=27153.1)|[7937][GFX1-]: Failed to allocate a surface due to invalid size (DSS) Size(0,0) (t=27146.7)|[7938][GFX1-]: Failed to a
llocate a surface due to invalid size (DSS) Size(0,0) (t=27148.1)|[7939][GFX1-]: Failed to allocate a surface due to invalid size (DSS) Size(0,0) (t=27148.7)|[7940][GFX1-
]: Failed to allocate a surface due to invalid size (DSS) Size(0,0) (t=27149.8)[GFX1-]: Failed to allocate a surface due to invalid size (DSS) Size(0,0)
[4392] WARNING: '!temp', file c:/Users/mozilla/debug-build/mozilla-central/gfx/layers/basic/BasicCompositor.cpp, line 481

Comment 1

[Nicolas Silva [:nical]]

I am not sure that the IsAncestor assertion and !temp are related. We haven't figured out the !temp assertion in BasicCompositor but we see it in various situations. There's bug 1065013. All of the logging about zero-sized surfaces always happens together with the !temp assertion.

Comment 2

[Matt Woodrow (:mattwoodrow)]

I can reproduce this fairly consistently on http://mrdoob.com/lab/javascript/threejs/css3d/

aClip1 is created by the root scrollframe of an iframe in the page, created by a call to InsertInactiveScrollClipForContentDescendants in ScrollFrameHelper.

aClip2 is the root scrollframe of the root content document, created by a call to TurnClipIntoScrollClipForContentDescendants in ScrollFrameHelper.

Both clips have no parent clip, nor a cross-stacking context parent which is why the assertion fails.

The frame that created the outer clip is still on the stack when we create the inner one, so I have no idea how we end up with no ancestors for the inner one.

I've confirmed that both are created on the same DisplayListClipState.

Comment 3

[Thinker Li [:sinker]]

Matt, is it possible coming from nsDisplayListBuilder::MarkFramesForDisplayList()?

Comment 4

[Markus Stange [:mstange]]

(In reply to Matt Woodrow (:mattwoodrow) from comment #2)
> The frame that created the outer clip is still on the stack when we create
> the inner one, so I have no idea how we end up with no ancestors for the
> inner one.

If the inner one is out-of-flow in some way, then the ancestor scroll clip will be the one that nsDisplayListBuilder::MarkOutOfFlowFrameForDisplay stored for it, as Thinker said.
But in the testcase I only see position:absolute, no position:fixed. So I'd expect the stored ancestor scroll clip to be the one created for the root scrollframe.

Has http://mrdoob.com/lab/javascript/threejs/css3d/ tripped this assertion ever since bug 1147673, or is it a regression from the recent 3d transform changes?

Comment 5

[Markus Stange [:mstange]]

I think the patches in bug 1238564 fix this.

Comment 6

[Carsten Book [:Tomcat]]

(In reply to Markus Stange [:mstange] from comment #5)
> I think the patches in bug 1238564 fix this.

yeah seems so, closing as wfm, thanks markus