Note: discovered via this LayoutTest: https://codereview.chromium.org/2733633002/diff/230001/third_party/WebKit/LayoutTests/compositing/overflow/composited-nested-sticky-inline.html

Small update: this doesn't actually need nested sticky, the sticky element just needs to be the second inline element. See updated repro at http://output.jsbin.com/muticiy/quiet

[1:1:0321/112540.030105:6126257109689:INFO:property_tree.cc(434)] 7: StickyPositionOffset: [12.000000 1.000000] - [-12.000000 1.000000] - [0.000000 0.000000] - [12 0] = [12.000000 0.000000]


This is sticky_offset - layer_offset - node->source_to_parent - constraint.scroll_container_relative_sticky_box_rect.OffsetFromOrigin()

I suspect the layer_offset is wrong (-12 seems wrong), but haven't thought about it for long yet.

[1:1:0321/150435.065508:6139392145118:INFO:property_tree_builder.cc(764)] 7: main_thread_offset = [12.000000 1.000000] - [24 0]

[1:1:0321/150808.838570:6139605918174:INFO:CompositedLayerMapping.cpp(326)] "LayoutInline (sticky positioned) DIV class='inline sticky'": parentRelativeStickyBoxOffset init as "12,0"
[1:1:0321/150808.838950:6139605918437:INFO:CompositedLayerMapping.cpp(334)] "LayoutInline (sticky positioned) DIV class='inline sticky'": enclosingLayerOffset init as "0,0"
[1:1:0321/150808.839126:6139605918576:INFO:CompositedLayerMapping.cpp(337)] "LayoutInline (sticky positioned) DIV class='inline sticky'": enclosingLayerOffset += "0x0"
[1:1:0321/150808.839248:6139605918692:INFO:CompositedLayerMapping.cpp(353)] "LayoutInline (sticky positioned) DIV class='inline sticky'": contentOffsetInCompositingLayer() is "-12x0"

The parentRelativeStickyBoxOffset is [12,0] - [0,0] - [-12,0], so this ends up as [24,0].

contentOffsetInCompositingLayer() seems to be wrong from a first thought.

[1:1:0321/151356.236570:6139953316118:INFO:CompositedLayerMapping.cpp(1736)] "LayoutInline (sticky positioned) DIV class='inline sticky'": updateContentsOffsetInCompositingLayer, setting m_contentOffsetInCompositingLayer to "0,1" - ("12,1" + "0,0")

Here we have:

snappedOffsetFromCompositedAncestor = [0, 1]
m_graphicsLayer->position() = [12, 1]
graphicsLayerParentLocation = [0, 0]

It feels like snappedOffsetFromCompositedAncestor should be [12, 1] not [0, 1]. Having trouble spotting where that might actually have gone wrong though. So far:

[1:1:0321/152509.263292:6140626342726:INFO:CompositedLayerMapping.cpp(1072)] "LayoutInline (sticky positioned) DIV class='inline sticky'": after computeBoundsOfOwningLayer, compositingContainer is "LayoutView #document", localCompositingBounds is "12,0 8x17", relativeCompositingBounds is "12,1 8x17", offsetFromCompositedAncestor is "0,1", snappedOffsetFromCompositedAncestor is "0,1"


[1:1:0321/152509.263039:6140626342489:INFO:CompositedLayerMapping.cpp(890)] "LayoutInline (sticky positioned) DIV class='inline sticky'": localRawCompositingBounds is "12,0 8x17"
[1:1:0321/152509.263178:6140626342617:INFO:CompositedLayerMapping.cpp(871)] "LayoutInline (sticky positioned) DIV class='inline sticky'": computeOffsetFromCompositedAncestor: offset is "12,1", subpixelAccum is "0x0", localRepresentativePointForFragmentation is "12,0"

I wonder if convertToLayerCoords is expected to return [12,1] here (note that the initial [12,0] is passed into it though.)

+chrishtr, +ajuma in the hope they may have some insight since they touched the code from #7 last (a long time ago admittedly).

Chris, Ali: When http://output.jsbin.com/muticiy/quiet is composited (press the button), the sticky positioned element is being positioned incorrectly. The root cause appears to be that its 'snappedOffsetFromCompositedAncestor' is [0,1] which to me seems to be incorrect (its composited ancestor is the root document).

Rough blame stack is:

1. contentOffsetInCompositingLayer() == [-12, 0]

2. updateContentsOffsetInCompositingLayer sets m_contentOffsetInCompositingLayer to [0,1] - ([12,1] + [0,0]) = [-12, 0]

3. snappedOffsetFromCompositedAncestor is set from computeBoundsOfOwningLayer, and turns out to be [0, 1]. I think it should be [12, 1].


Do either of you have any idea as to whether that code 'should' work for inline elements or not?

Yes it's supposed to work. However, see https://codereview.chromium.org/2738943004
for an example of subtlety in this area that might help you. TL;DR is that I fixed
a problem resulting from the fact that an inline PaintLayer has a
boundingBoxForCompositing that does not start at 0,0 because it is positioned in the
space of its containing block. Blocks OTOH start at 0,0 and have a location()
that is the offset from its containing block.

This same situation may be happening to you, not sure.

No current work is happening by me on sticky, dropping to Available.

Thanks to chrishtr@ who pointed out to me that this no longer reproduces. (A lesson to me to check that when cleaning up bugs!)

I ran a bisect:

You are probably looking for a change made after 477877 (known bad), but no later than 477882 (first known good).
CHANGELOG URL:
  https://chromium.googlesource.com/chromium/src/+log/cfa975069fd5856fdc4c494276a3a9af6be3da79..0867b7646e170b2685bb47ed938fb4562f55c0df

This was likely fixed by yigu's change in https://chromium.googlesource.com/chromium/src/+/65784430dea11bf2e12e947c77dd08ce2c543c6f

Cool