The image is an <iron-image> that's being transformed with translateY up and down.

<iron-image id="material-hero-image" preload="true" fade="true" sizing="cover" position="top" class="x-scope iron-image-0" style="transform: translateY(-118px);">
...
  </iron-image>

Looking at the paint invalidation flashing this is a non-composited transform moving the background image up and down. We're not issuing the paint invalidation rects in the right spots so the previous location in the layer is left with old content.

Deleted: Screenshot 2016-12-11 at 6.29.41 PM.png 1.2 MB

	Screenshot 2016-12-11 at 6.29.41 PM.png 1.2 MB View Download

I didn't reproduce the issue on Pixel 1 (55.0.2883.87 beta) and Linux (bisecting from r414002) with or without --enable-prefer-compositing-to-lcd text.

I saw a different UI than yours. See the attachment. Is the difference because play music changed their web site?

Deleted: Screenshot 2016-12-11 at 8.06.15 PM.png 1.9 MB

	Screenshot 2016-12-11 at 8.06.15 PM.png 1.9 MB View Download

Anyone have a Pixel 2 to verify on?

I also wasn't able to reproduce, but I also don't have a pixel 2. I can try on my retina macbook later today.

@esprehn, can you check chrome://gpu to see if you have any flags enabled?

@pdr are you seeing the same UI as mine (with an album cover on top of the background image, and "Other artists on this station") or esprehn's (with a song list)?

I am getting roughly the UI Elliott is. I think you may need to click play and possibly click on a specific song. You may need an account to do this :/

I was able to repro this with this url: https://play.google.com/music/listen?u=0#/ps/Iy7ltzd4ghvrrmilysh7y52qzkq

Steps to reproduce:
1) go to url
2) scroll up

There were flags set on this device but I disabled them and this still reproduces.

Assigning to myself to look into this further on a retina mac.

I was able to confirm this on a Retina mac, even in dev. This regression exists as far back as at least M48.

@Testers, can you please bisect this on a retina mac?

To bisect this, I used "--enable-prefer-compositing-to-lcd-text" on linux. I manually bisected this to:
https://chromium.googlesource.com/chromium/blink/+log/0ee3fcb..8f8d2bb

I think this is one of these, likely the first:
https://chromium.googlesource.com/chromium/blink/+/8f8d2bb1e41a8779f9e15d0e8d56b95a5c969831
https://chromium.googlesource.com/chromium/blink/+/b5f009ecf7a71738f8b121738c9ea8a07d3e135b

I also managed to get my chromium account locked due to re-logging in for every bisect point. Don't do that.

--user-data-dir=/tmp/a might help to avoid re-logging-ins for every bisect point.

I logged in but still saw the UI like in #2. Do I need to subscribe Google Music?

I am not sure how https://chromium.googlesource.com/chromium/blink/+/8f8d2bb1e41a8779f9e15d0e8d56b95a5c969831 caused the bug, but the logic seems incomplete. We should update not only previous paint invalidation rects but also previous locations, for not only layers but also all scrolled objects. I think we should just force a subtree paint invalidation here instead of trying to adjust the previous paint invalidation rects.

Lowering priority because it's an old regression.

Did you try this url? https://play.google.com/music/listen?u=0#/ps/Iy7ltzd4ghvrrmilysh7y52qzkq

Good point about using a tmp user data dir.

This URL works for me to reproduce the bug. Thanks!

Note this bug was fixed by Music (http://b/33548244) by adding will-change: transform which removes the repainting. If you remove that it'll come back. Could someone make a reduction and attach it to this issue?

Almost there on a reduction..

Reduced testcase attached.

Deleted: testcase.html 1.0 KB

testcase.html 1.0 KB View Download

Reproduces on 57.0.2954.0 / Mac Retina.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/03444642622ffcecb8197bf40d6a21495d694240

commit 03444642622ffcecb8197bf40d6a21495d694240
Author: wangxianzhu <wangxianzhu@chromium.org>
Date: Tue Dec 27 18:02:15 2016

Replace adjustment of previous visual rects on scroll with normal paint invalidation

Previously the adjustment of previous visual rects on scroll was not
complete: it ignored non-layered objects, and didn't adjust location
in backing.

Now set paint invalidation flag on scroll if needed, and all required
adjustments will be done during paint invalidation.

BUG= 673197 
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2

Review-Url: https://codereview.chromium.org/2605543002
Cr-Commit-Position: refs/heads/master@{#440763}

[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/LayoutTests/paint/invalidation/overflow-hidden-yet-scrolled-with-custom-scrollbar-expected.txt
[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/Source/core/frame/FrameView.cpp
[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/Source/core/frame/FrameView.h
[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/Source/core/layout/LayoutObject.cpp
[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/Source/core/layout/LayoutObject.h
[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/Source/core/paint/PaintLayer.cpp
[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/Source/core/paint/PaintLayer.h
[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/Source/core/paint/PaintLayerScrollableArea.cpp
[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/Source/core/paint/PaintLayerTest.cpp
[modify] https://crrev.com/03444642622ffcecb8197bf40d6a21495d694240/third_party/WebKit/Source/platform/scroll/ScrollAnimatorBase.h

Fixed?