I think this is an impl-thread hit testing bug.

It reproduces on desktop devtools device emulation (randomly, when you resize it can start or stop reproducing).  The page content seems quite ordinary: it's a scrollable sublayer, the layout sizes are correct, there is no overflow: hidden nor touch listeners.  Touch and mousewheel scrolls don't work, but keyboard arrow-key scrolls do.

This appears to have been fixed in ToT, though I don't know what CL is responsible. A bisect would be helpful.

It's not fixed on ToT for me.  This is chrome_public_apk at r388892 on a Nexus 5X running MHC19J.

aelias: yep, I missed your comment #3 about resizing. If in emulation mode I switch to 125%, I can repro. Will dig in.

http://crrev.com/4bab6cab4df2a082f47729972d642afa022afc47 is the change that introduced this regression. If I'm sync'd to that change, I can reproduce the problem in my Chromium build. If I then revert that one change, the problem is not reproducible.

Thanks, assigning to jaydasika@ since the regression was bisected to his patch.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/0da498c027ed0b0d1c4596892c7666afaa5a98af

commit 0da498c027ed0b0d1c4596892c7666afaa5a98af
Author: jaydasika <jaydasika@chromium.org>
Date: Tue Apr 26 20:26:45 2016

cc: Fix sublayer scale bug in hit testing

In hit testing, we should use the render surface screen space
transform (which doesn't have sublayer scale) instead of the
screen_space_transform stored on the transform node (which
has the sublayer scale baked in) as the clips stored in clip
tree already account for the scale.

BUG= 604731 
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Review URL: https://codereview.chromium.org/1917043002

Cr-Commit-Position: refs/heads/master@{#389873}

[modify] https://crrev.com/0da498c027ed0b0d1c4596892c7666afaa5a98af/cc/trees/layer_tree_impl.cc
[modify] https://crrev.com/0da498c027ed0b0d1c4596892c7666afaa5a98af/cc/trees/layer_tree_impl_unittest.cc

This regression was introduced in Chrome 49. Is it worthy of merging to M51?

Seems reasonable to me - marking merge requested.

Your change meets the bar and is auto-approved for M51 (branch: 2704)

jaydasika@chromium.org, can you merge this?

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/97c769d0257b107238ab72175c74da80749d4778

commit 97c769d0257b107238ab72175c74da80749d4778
Author: Jayadev Dasika <jaydasika@google.com>
Date: Mon May 02 20:53:38 2016

cc: Fix sublayer scale bug in hit testing

In hit testing, we should use the render surface screen space
transform (which doesn't have sublayer scale) instead of the
screen_space_transform stored on the transform node (which
has the sublayer scale baked in) as the clips stored in clip
tree already account for the scale.

BUG= 604731 
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Review URL: https://codereview.chromium.org/1917043002

Cr-Commit-Position: refs/heads/master@{#389873}
(cherry picked from commit 0da498c027ed0b0d1c4596892c7666afaa5a98af)

Review URL: https://codereview.chromium.org/1939283002 .

Cr-Commit-Position: refs/branch-heads/2704@{#340}
Cr-Branched-From: 6e53600def8f60d8c632fadc70d7c1939ccea347-refs/heads/master@{#386251}

[modify] https://crrev.com/97c769d0257b107238ab72175c74da80749d4778/cc/trees/layer_tree_impl.cc
[modify] https://crrev.com/97c769d0257b107238ab72175c74da80749d4778/cc/trees/layer_tree_impl_unittest.cc

FYI the merge to m51 got reverted due to a crash in webview: crbug.com/614438 (sorry private bug)

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/8d0064fad7a0842a60a61a00ffc58775f8c8f982

commit 8d0064fad7a0842a60a61a00ffc58775f8c8f982
Author: jaydasika <jaydasika@chromium.org>
Date: Thu Jun 02 23:34:54 2016

cc : Fix hit testing bug in resourceless software draw mode

Hit testing assumes that the transform node corresponding to target id
of clip node always has a render surface. This is not true when we are
in resourceless software draw mode.

BUG=614438,  604731 
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Review-Url: https://codereview.chromium.org/2033163003
Cr-Commit-Position: refs/heads/master@{#397543}

[modify] https://crrev.com/8d0064fad7a0842a60a61a00ffc58775f8c8f982/cc/trees/layer_tree_host_common_unittest.cc
[modify] https://crrev.com/8d0064fad7a0842a60a61a00ffc58775f8c8f982/cc/trees/layer_tree_host_impl.h
[modify] https://crrev.com/8d0064fad7a0842a60a61a00ffc58775f8c8f982/cc/trees/layer_tree_impl.cc
[modify] https://crrev.com/8d0064fad7a0842a60a61a00ffc58775f8c8f982/cc/trees/layer_tree_impl.h

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/bb87edece9bacb94593ace23a2c36dc3f3439346

commit bb87edece9bacb94593ace23a2c36dc3f3439346
Author: Jayadev Dasika <jaydasika@google.com>
Date: Sat Jun 04 01:30:01 2016

cc : Fix hit testing bug in resourceless software draw mode

Hit testing assumes that the transform node corresponding to target id
of clip node always has a render surface. This is not true when we are
in resourceless software draw mode.

BUG=614438,  604731 
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Review-Url: https://codereview.chromium.org/2033163003
Cr-Commit-Position: refs/heads/master@{#397543}
(cherry picked from commit 8d0064fad7a0842a60a61a00ffc58775f8c8f982)

Review URL: https://codereview.chromium.org/2042573002 .

Cr-Commit-Position: refs/branch-heads/2743@{#226}
Cr-Branched-From: 2b3ae3b8090361f8af5a611712fc1a5ab2de53cb-refs/heads/master@{#394939}

[modify] https://crrev.com/bb87edece9bacb94593ace23a2c36dc3f3439346/cc/trees/layer_tree_host_common_unittest.cc
[modify] https://crrev.com/bb87edece9bacb94593ace23a2c36dc3f3439346/cc/trees/layer_tree_host_impl.h
[modify] https://crrev.com/bb87edece9bacb94593ace23a2c36dc3f3439346/cc/trees/layer_tree_impl.cc
[modify] https://crrev.com/bb87edece9bacb94593ace23a2c36dc3f3439346/cc/trees/layer_tree_impl.h

Moving Blink>Input>HitTesting to Blink>HitTesting