Based on the bisect, I'm guessing it was this change:
https://chromium.googlesource.com/chromium/src/+/6c4cca504533b317b3b61e7417e4582a45f58881

+xlai

Adding reviewers also to analyze the severity/impact on M64 stable launch.

Unable to reproduce this on non-DPI Mac machine on Version 64.0.3282.119.

I have very little information to dig into debugging. What kind of canvas API are called in the process of using the eyedrop tool? Could you reproduce this
in a smaller test case like in JSFiddle?

xlai: Could you find a high-DPI Mac to test with, or perhaps configure macOS to do a 2x scale somehow? Given the bug involves the canvas size getting halved, it's likely that this only reproduces when the scale factor is > 1.

davidben@: My machine could not scale 2x (scaling down resolution cannot reproduce the bug either) but I will find one high-DPI to reproduce it tonight. 

The point I tried to make in Comment 5 is actually about the lack of
more information that could aid in debugging--what canvas API are called, and in what sequence of calls, in using the eyedrop in this example? 

I can still try debugging based on the information about high-DPI difference but
if more information can be provided by the user that could help me speed up the debugging time.

The eyedropper is running the following code on mouse down:
https://github.com/LLK/scratch-paint/blob/develop/src/helper/tools/eye-dropper.js#L38

By turning on "paint flashing" in the render tool of Chrome devtools, you can see that the miniature copy of the canvas appears immediately after the next render of the canvas.

I've made a JS Fiddle at https://jsfiddle.net/ts8wvpza/6/
It reproduces the bug as soon as you mouse over the canvas. It seems to be caused by calling context.getImageData and then rerendering.

Thanks liudi08@gmail.com for coming up with the JS Fiddle. I'm looking into it.

I am pretty sure that this CL will fix the problem in a high-DPI mac machine:
https://chromium-review.googlesource.com/c/chromium/src/+/899962

I will try to land it today and merge it to M64 and M65 asap (M64 merge CL is ready,
waiting for review here: 
https://chromium-review.googlesource.com/c/chromium/src/+/899913).

Sorry for the trouble.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/07062c952f820196c880cea5bb05158881662d45

commit 07062c952f820196c880cea5bb05158881662d45
Author: xlai <xlai@chromium.org>
Date: Fri Feb 02 21:14:31 2018

Set CanvasResourceHost when canvas replaces its canvas2d layer bridge

TBR=junov@chromium.org

Bug:  807636 ,  807427 
Change-Id: Ic5fe5abbf5deaed883c005280fd8418cf49ca9d2
Reviewed-on: https://chromium-review.googlesource.com/899962
Commit-Queue: Olivia Lai <xlai@chromium.org>
Reviewed-by: Mohammad Reza Zakerinasab <zakerinasab@chromium.org>
Cr-Commit-Position: refs/heads/master@{#534152}
[modify] https://crrev.com/07062c952f820196c880cea5bb05158881662d45/third_party/WebKit/Source/core/html/canvas/HTMLCanvasElement.cpp
[modify] https://crrev.com/07062c952f820196c880cea5bb05158881662d45/third_party/WebKit/Source/modules/canvas/canvas2d/CanvasRenderingContext2DTest.cpp

Requesting merge fix to M64, M65 to reduce negative impact on user.

This bug requires manual review: Request affecting a post-stable build
Please contact the milestone owner if you have questions.
Owners: cmasso@(Android), cmasso@(iOS), kbleicher@(ChromeOS), abdulsyed@(Desktop)

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

This just landed 4 hours ago, and doesn't have enough Canary coverage. Can you please explain why this is absolutely needed for M64? We're already rolling out M64 in stable. 

As discussed over chat, xlai@ will propose client side workarounds. And we will target the fix for M65. 

liudi08@gmail.com:
There is a cost to merge a fix to stable release; so we
will merge to M65 which is currently in its Beta.

As a temporary workaround to this problem in Chrome 64, you could do this:
1. type "chrome://flags" in a Chrome tab and enter.
2. Among the list of Chrome flags, find "Accelerated 2D Canvas", disable this flag.
3. Relaunch browser.

FWIW, I can also repro via the JSFiddle on a high-DPI Windows machine.

Is this the sort of thing we could get a layout test into Blink for future? (Do layout tests cover this case?) Looks like the unit test here wasn't quite of the final rendering and might get lost in a future refactor.

Your change meets the bar and is auto-approved for M65. Please go ahead and merge the CL to branch 3325 manually. Please contact milestone owner if you have questions.
Owners: cmasso@(Android), cmasso@(iOS), bhthompson@(ChromeOS), govind@(Desktop)

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

Pls merge your change to M65 branch 3325 before 2:00 PM PT tomorrow, Monday (02/25/18) so we can pick it up for last dev release on Tuesday. Thank you.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/160b0172e48798827282c024d0e61c8abbf5edaa

commit 160b0172e48798827282c024d0e61c8abbf5edaa
Author: xlai <xlai@chromium.org>
Date: Sun Feb 04 22:36:47 2018

Set CanvasResourceHost when canvas replaces its canvas2d layer bridge


Bug:  807636 ,  807427 
Change-Id: Ic5fe5abbf5deaed883c005280fd8418cf49ca9d2
Reviewed-on: https://chromium-review.googlesource.com/899962
Commit-Queue: Olivia Lai <xlai@chromium.org>
Reviewed-by: Mohammad Reza Zakerinasab <zakerinasab@chromium.org>
Cr-Original-Commit-Position: refs/heads/master@{#534152}
Reviewed-on: https://chromium-review.googlesource.com/900455
Reviewed-by: Olivia Lai <xlai@chromium.org>
Cr-Commit-Position: refs/branch-heads/3325@{#285}
Cr-Branched-From: bc084a8b5afa3744a74927344e304c02ae54189f-refs/heads/master@{#530369}
[modify] https://crrev.com/160b0172e48798827282c024d0e61c8abbf5edaa/third_party/WebKit/Source/core/html/canvas/HTMLCanvasElement.cpp

Verified the fix on Mac 10.12.6(retina) and Win-10 using Chrome dev version #65.0.3325.51 as per the comment #9.
Attaching screen cast for reference.
Observed that eyedropping did not affect what the canvas looks like.
Hence, the fix is working as expected. 
Adding the verified labels.

Thanks...!!

Deleted: 807427.mp4 1.6 MB

	807427.mp4 1.6 MB View Download

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/68f180c2597177c389641868a36c8c4e5208aa3e

commit 68f180c2597177c389641868a36c8c4e5208aa3e
Author: xlai <xlai@chromium.org>
Date: Tue Feb 06 18:05:24 2018

Set CanvasResourceHost when canvas replaces its canvas2d layer bridge

This CL is a hotfix to M64 branch based on
https://chromium-review.googlesource.com/c/chromium/src/+/899962. Due to code
structural change between M64 and M66, it is a bit different from the original
CL.

Bug:  807636 ,  807427 
Change-Id: I7f5826187e3deef2ceaeee796dc7fdf3baf71717
Reviewed-on: https://chromium-review.googlesource.com/899913
Reviewed-by: Justin Novosad <junov@chromium.org>
Cr-Commit-Position: refs/branch-heads/3282@{#652}
Cr-Branched-From: 5fdc0fab22ce7efd32532ee989b223fa12f8171e-refs/heads/master@{#520840}
[modify] https://crrev.com/68f180c2597177c389641868a36c8c4e5208aa3e/third_party/WebKit/Source/core/html/HTMLCanvasElement.cpp

 Issue 809403  has been merged into this issue.

We have landed the fix patch, which is the same fix to https://bugs.chromium.org/p/chromium/issues/detail?id=807636, into Chrome 64.

Once the latest release with this fix goes out in the coming week, we will update this issue again.

I'm marking this as fixed because the fix CL is already waiting in the next update of Chrome 64. But I'll continue verifying it after the update is released.

Verified the fix on Mac 10.12.6(retina) and Win-10 using Chrome version #64.0.3282.167 as per the comment #9.
Attaching screen cast for reference.
Observed that eyedropping did not affect what the canvas looks like.
Hence, the fix is working as expected. 
Adding the verified labels.

Thanks...!!

Deleted: 807427@M64.mp4 2.2 MB

	807427@M64.mp4 2.2 MB View Download