616092 - position: absolute breaks opaque shadowThebesLayer state in remote case

Status: RESOLVED WORKSFORME

(Core :: Layout, defect)

Description

[Oleg Romashin (:romaxa)]

on google.com we have a bit special layer which makes shadow thebes layer to be non-opaque, that cause making ShadowableThebes buffer with offscreen big enough and 2x bigger on mobile because opaque layer usually means 16bpp 2 bpp surface.

I think that is wrong and we should not create transparent layer when it is not needed.

Comment 1

[Oleg Romashin (:romaxa)]

Attachment: simple testcase

Comment 2

[Oleg Romashin (:romaxa)]

Attachment: Layers dump

Comment 3

[Oleg Romashin (:romaxa)]

Not sure probably testcase could be more simple, but when I removing anything from this testcase style, then it is started create opaque layers... I think that is related to http://mxr.mozilla.org/mozilla-central/source/layout/base/FrameLayerBuilder.cpp#847

Comment 4

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

This testcase results in an opaque layer on desktop

-1220061184[122aab0]: BasicLayerManager (0x7fc89c0af950)
-1220061184[122aab0]:   BasicContainerLayer (0x2506bd0) [visible=< (x=0, y=0, w=821, h=809); >] [opaqueContent] [metrics={ viewport=(w=821, h=809) viewportScroll=(x=0, y=0) displayport=(x=0, y=0, w=0, h=0) }]
-1220061184[122aab0]:     BasicThebesLayer (0x2505e60) [visible=< (x=0, y=0, w=821, h=120); >] [opaqueContent] [noTextOverTransparent] [valid=< (x=0, y=0, w=821, h=32); (x=0, y=32, w=821, h=88); >]
-1220061184[122aab0]:     BasicContainerLayer (0x2507170) [clip=(x=0, y=120, w=821, h=689)] [visible=< (x=0, y=120, w=821, h=689); >] [opaqueContent]
-1220061184[122aab0]:       BasicThebesLayer (0x250c9f0) [transform=[ 1 0; 0 1; 0 120; ]] [visible=< (x=0, y=0, w=821, h=689); >] [opaqueContent] [noTextOverTransparent] [valid=< (x=0, y=0, w=821, h=689); >]

Looks like a problem with fennec's setup in the content process.

Comment 5

[Oleg Romashin (:romaxa)]

On desktop google giving us different page... have you tried to setup fennec UA in desktop? and yeah.. .something wrong with content process setup.

Comment 6

[Oleg Romashin (:romaxa)]

Oh, I'm wrong... you are running simple testcase.... and with non-remote fennec that non-shadowTHebesLayer is opaque... only problem with remote setup.

Comment 7

[Oleg Romashin (:romaxa)]

Tatiana did short investigation and found that 
x=0, y=-12, w=980, h=539 - clipped correctly in non-remote case, and y-negative part of rectangle recognized in remote case as transparent region.

Sounds like some clip or other value like effectiveVisible region not applied correctly on shadow Layers root tree...

Comment 8

[Tatiana Meshkova (:tatiana)]

This -12 non-opaque "text2" part becomes visible because remote Fennec uses displayport rendering and visibleRegion has negative part in this case:

PRBool nsDisplayList::ComputeVisibilityForRoot(nsDisplayListBuilder*, nsRegion*):363
--aVisibleRegion: < (x=-10800, y=-45000, w=69600, h=120000); >
builder: 0xbf8edbe4, Type: 44, Name: Text
--rectOnCompute: (x=480, y=-720, w=2340, h=1140)--aVisibleRegionOnCompute: < (x=-10800, y=-45000, w=69600, h=120000); >--boundsOnCompute: (x=480, y=-720, w=2340, h=1140)--aListVisibleBoundsOnCompute: (x=0, y=-720, w=48000, h=30720)
builder: 0xbf8edbe4, Type: 44, Name: Text
--rectOnCompute: (x=480, y=480, w=1740, h=1140)--aVisibleRegionOnCompute: < (x=-10800, y=-45000, w=69600, h=120000); >--boundsOnCompute: (x=480, y=480, w=1740, h=1140)--aListVisibleBoundsOnCompute: (x=0, y=-720, w=48000, h=30720)
builder: 0xbf8edbe4, Type: 11, Name: CanvasBackground
--rectOnCompute: (x=0, y=0, w=48000, h=30000)--aVisibleRegionOnCompute: < (x=-10800, y=-45000, w=69600, h=120000); >--boundsOnCompute: (x=0, y=0, w=48000, h=30000)--aListVisibleBoundsOnCompute: (x=0, y=-720, w=48000, h=30720)

Layer: 0xb07b0560 isOpaque: 0

Comment 9

[Daniel Holbert [:dholbert]]

The string thebeslayer no longer exists in the tree. I think this bug is obsoleted by WebRender and subsequent code-removal in bug 1724935 ("kill-layers").

(I ran across this bug in a random search for "shadow position absolute" when looking to find if a particular bug about absolute positioning with custom-elements & shadow-root happens to already be filed or not.)