Reverted suspected patch: https://chromium-review.googlesource.com/c/chromium/src/+/1318742

A new attempt which explicitly updates the AXObjectCache:

https://chromium-review.googlesource.com/c/chromium/src/+/1319610

Reduced pri since offender was reverted.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/18a9ef4599aa9e6cb6fdc27484ac50fcfdda9482

commit 18a9ef4599aa9e6cb6fdc27484ac50fcfdda9482
Author: Rune Lillesveen <futhark@chromium.org>
Date: Tue Nov 06 12:08:12 2018

Full AttachLayoutTree on a new Document not necessary.

We called AttachLayoutTree on a new Document which just got its
LayoutView created. The new Document should not have any child nodes,
it's not marked initially dirty, but trying to land this change the
first time we caused flakiness for a Window accessibility test.

Instead of the AttachLayoutTree call, do the AXObjectCache update which
would otherwise be called from Node::AttachLayoutTree.

This is a new attempt after the previous got reverted:

https://chromium-review.googlesource.com/c/chromium/src/+/1317889

Bug:  902050 
Change-Id: I4e366013549627b54f016d76b3dcbf12ea115f70
Reviewed-on: https://chromium-review.googlesource.com/c/1319610
Reviewed-by: Anders Ruud <andruud@chromium.org>
Commit-Queue: Rune Lillesveen <futhark@chromium.org>
Cr-Commit-Position: refs/heads/master@{#605657}
[modify] https://crrev.com/18a9ef4599aa9e6cb6fdc27484ac50fcfdda9482/third_party/blink/renderer/core/dom/document.cc

Let's see if this sticks.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/d941a8393d4275d5d4cfa69285e45436f43e25a1

commit d941a8393d4275d5d4cfa69285e45436f43e25a1
Author: Dmitry Titov <dimich@chromium.org>
Date: Tue Nov 06 22:02:34 2018

Revert "Full AttachLayoutTree on a new Document not necessary."

This reverts commit 18a9ef4599aa9e6cb6fdc27484ac50fcfdda9482.

Reason for revert: Broke the accessibility test (again): https://ci.chromium.org/p/chromium/builders/luci.chromium.ci/Win10%20Tests%20x64/29232

Please note this does not look like a flake, after this landed a dozen runs failed consistently. 

Original change's description:
> Full AttachLayoutTree on a new Document not necessary.
> 
> We called AttachLayoutTree on a new Document which just got its
> LayoutView created. The new Document should not have any child nodes,
> it's not marked initially dirty, but trying to land this change the
> first time we caused flakiness for a Window accessibility test.
> 
> Instead of the AttachLayoutTree call, do the AXObjectCache update which
> would otherwise be called from Node::AttachLayoutTree.
> 
> This is a new attempt after the previous got reverted:
> 
> https://chromium-review.googlesource.com/c/chromium/src/+/1317889
> 
> Bug:  902050 
> Change-Id: I4e366013549627b54f016d76b3dcbf12ea115f70
> Reviewed-on: https://chromium-review.googlesource.com/c/1319610
> Reviewed-by: Anders Ruud <andruud@chromium.org>
> Commit-Queue: Rune Lillesveen <futhark@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#605657}

TBR=futhark@chromium.org,andruud@chromium.org

Change-Id: Ied8ff9426940489ac4693abdb34001f44b13478c
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug:  902050 
Reviewed-on: https://chromium-review.googlesource.com/c/1321229
Reviewed-by: Dmitry Titov <dimich@chromium.org>
Commit-Queue: Dmitry Titov <dimich@chromium.org>
Cr-Commit-Position: refs/heads/master@{#605839}
[modify] https://crrev.com/d941a8393d4275d5d4cfa69285e45436f43e25a1/third_party/blink/renderer/core/dom/document.cc

Indeed, the revert fixed the build after several builds reliably failed since it was landed: https://ci.chromium.org/p/chromium/builders/luci.chromium.ci/Win10%20Tests%20x64/29236

Since test replaces innerHTML and then verifies DOM it might be affected somehow. Not sure but landing this patch twice, failing the same test and fixing it after 2 reverts definitely makes it related... 

I'll not try to reland the offending patch in any form at this point. Closing.