Confirmed. Waiting on bisect.

Bisects to
https://chromium.googlesource.com/chromium/src/+log/57fd39f44a3af5474f75cca7bd9870d4e26fc6d2..d290cbee660c449245346abe4a9f2c7948df4c9e

Specifically, https://chromium.googlesource.com/chromium/src/+/d290cbee660c449245346abe4a9f2c7948df4c9e

I don't think that the static/animating behaviro change is the issue. I think that the perspective transform changed such that the static case now has us looking directly onto the front face of the die. The previous behavior, which matches Firefox now, is to have larger die with a very obvious non-centered projection.

Marking as P1 regression until we know whether the new behavior is right, and until we can reduce the test case to understand the underlying importance of the issue.

Passing on to Chris to find an owner for investigating the spec questions raised in #3.

Reverting https://chromium.googlesource.com/chromium/src/+/d290cbee660c449245346abe4a9f2c7948df4c9e doesn't make any difference to the behavior on ToT. Maybe the bisect is wrong or maybe something else changed in conjunction with it.

This demo doesn't seem to hit any known spec bug. I think the right behavior is to match M57 & Firefox.

The simplified page looks like this:
<div style="perspective:60em">
  <div>
  .... deep nested div, no fix-pos container or grouping stuff
    <div>
      <div style="position:fixed; transform-style:preserve-3d;">
        <div style="transform:...">face of the dice</div>
        <div style="transform:...">face of the dice</div>
        <div style="transform:...">face of the dice</div>
      </div>
    </div>
  </div>
</div>

It does look like having something to do with r449448, though I don't know exactly how. :/

Ah, there's the change in cc/layers/layer.cc

 bool Layer::IsContainerForFixedPositionLayers() const {
-  if (!inputs_.transform.IsIdentityOrTranslation())
-    return true;
-  if (parent_ && !parent_->inputs_.transform.IsIdentityOrTranslation())
-    return true;
   return inputs_.is_container_for_fixed_position_layers;
 }

Thus the perspective layer (CLM::m_childTransformLayer) is no longer considered fixed-pos container automatically.

Just verified locally the following patch fixes the bug:

diff --git a/third_party/WebKit/Source/core/layout/compositing/CompositedLayerMapping.cpp b/third_party/WebKit/Source/core/layout/compositing/CompositedLayerMapping.cpp
index 445b2ad8364f..a2cd3b82e32c 100644
--- a/third_party/WebKit/Source/core/layout/compositing/CompositedLayerMapping.cpp
+++ b/third_party/WebKit/Source/core/layout/compositing/CompositedLayerMapping.cpp
@@ -1613,7 +1613,7 @@ void CompositedLayerMapping::RegisterScrollingLayers() {
       owning_layer_.GetLayoutObject().CanContainFixedPositionObjects() &&
       !owning_layer_.IsRootLayer();
   scrolling_coordinator->SetLayerIsContainerForFixedPositionLayers(
-      graphics_layer_.get(), is_container);
+      child_transform_layer_ ? child_transform_layer_.get() : graphics_layer_.get(), is_container);
 }

 void CompositedLayerMapping::UpdateInternalHierarchy() {

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/25eaaeb522126f998e9bd83056f96a5ce3cc78c2

commit 25eaaeb522126f998e9bd83056f96a5ce3cc78c2
Author: trchen <trchen@chromium.org>
Date: Wed Apr 26 00:17:10 2017

[blink] Perspective layer should be considered fixed-pos container layer

This CL fixes a bug that fixed-position descendants skip its containing
block's perspective transform. Perspective layer (CompositedLayerMapping::
child_transform_layer_) should be considered fixed-pos container if
present.

BUG= 714001 
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2

Review-Url: https://codereview.chromium.org/2833233002
Cr-Commit-Position: refs/heads/master@{#467175}

[add] https://crrev.com/25eaaeb522126f998e9bd83056f96a5ce3cc78c2/third_party/WebKit/LayoutTests/transforms/perspective-fixed-pos-descendant-expected.html
[add] https://crrev.com/25eaaeb522126f998e9bd83056f96a5ce3cc78c2/third_party/WebKit/LayoutTests/transforms/perspective-fixed-pos-descendant.html
[modify] https://crrev.com/25eaaeb522126f998e9bd83056f96a5ce3cc78c2/third_party/WebKit/Source/core/layout/compositing/CompositedLayerMapping.cpp

Should we bring this back to M58 & M59?

Your change meets the bar and is auto-approved for M59. Please go ahead and merge the CL to branch 3071 manually. Please contact milestone owner if you have questions.
Owners: amineer@(Android), cmasso@(iOS), gkihumba@(ChromeOS), Abdul Syed@(Desktop)

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/425fee0e59a4cbaa0ff67a9f2702a269bcd1ac18

commit 425fee0e59a4cbaa0ff67a9f2702a269bcd1ac18
Author: Tien-Ren Chen <trchen@chromium.org>
Date: Fri Apr 28 18:49:43 2017

[blink] Perspective layer should be considered fixed-pos container layer

This CL fixes a bug that fixed-position descendants skip its containing
block's perspective transform. Perspective layer (CompositedLayerMapping::
child_transform_layer_) should be considered fixed-pos container if
present.

BUG= 714001 
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2

Review-Url: https://codereview.chromium.org/2833233002
Cr-Commit-Position: refs/heads/master@{#467175}
(cherry picked from commit 25eaaeb522126f998e9bd83056f96a5ce3cc78c2)

Review-Url: https://codereview.chromium.org/2846313002 .
Cr-Commit-Position: refs/branch-heads/3071@{#299}
Cr-Branched-From: a106f0abbf69dad349d4aaf4bcc4f5d376dd2377-refs/heads/master@{#464641}

[add] https://crrev.com/425fee0e59a4cbaa0ff67a9f2702a269bcd1ac18/third_party/WebKit/LayoutTests/transforms/perspective-fixed-pos-descendant-expected.html
[add] https://crrev.com/425fee0e59a4cbaa0ff67a9f2702a269bcd1ac18/third_party/WebKit/LayoutTests/transforms/perspective-fixed-pos-descendant.html
[modify] https://crrev.com/425fee0e59a4cbaa0ff67a9f2702a269bcd1ac18/third_party/WebKit/Source/core/layout/compositing/CompositedLayerMapping.cpp

A bit late for M58. Only merged to M59 for now.

Verified this issue on Mac OS 10.12.4, Windows-10 and Ubuntu 14.04 using chrome latest beta M59-59.0.3071.36 by following steps mentioned in the original comment. Observed able to view the dice in perspective view as expected. Hence adding TE-Verified label.

Thanks! 

Deleted: Screenshot from 2017-05-03 13:28:50.png 39.0 KB

	Screenshot from 2017-05-03 13:28:50.png 39.0 KB View Download

Thanks guys! Everything works well after the last update :)