The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/20f1a16e0ed6c595f50df84cb829e7f4fe842fd3

commit 20f1a16e0ed6c595f50df84cb829e7f4fe842fd3
Author: bsep <bsep@chromium.org>
Date: Fri Feb 10 01:25:34 2017

Disable ImageWriterUtilityClientTest.VerifyCancel and [etc].WriteCancel.

They're flaking a lot on the Win7 Debug bot.

TBR=haven@chromium.org
BUG= 690717 

Review-Url: https://codereview.chromium.org/2684103004
Cr-Commit-Position: refs/heads/master@{#449504}

[modify] https://crrev.com/20f1a16e0ed6c595f50df84cb829e7f4fe842fd3/chrome/browser/extensions/api/image_writer_private/image_writer_utility_client_browsertest.cc

Built a DEBUG chrome on WIN10
 - was not able to reproduce the current  bug 690717  reliably.
 - failed about 1 / 1000 browser test runs sometimes.
 - unreliable reproduction.

Changed the Cancel() code to read:

void ImageWriterUtilityClient::Cancel(const CancelCallback& cancel_callback) {	 
  DCHECK_CURRENTLY_ON(content::BrowserThread::IO);

  ResetRequest();
  task_runner_->PostDelayedTask(FROM_HERE, cancel_callback, kDelay);
}

Used a kDelay >= 600ms, this to simulate a really slow DEBUG bot, and I was able to reproduce this  bug 690717  reliably (in all runs, I did 100) in DEBUG Win10.

Problem seems to do with the task_runner_->PostTask that ImageWriterUtilityClient uses to callback its clients.

Why are the callers and ImageWriterUtilityClient code using a task_runner anyway?
 - callers bounce requests rx-ed on FILE thread to the IO thread, which
   ImageWriterUtilityClient expects.
 - seems that is a hang over from an earlier time when ImageWriterUtilityClient
   used UtilityProcessHostClient to control the utility process (needed IO
   thread for that).
 - UtilityProcessHostClient responses are bounced back from IO thread to FILE
   using task_runner_->PostTask() (Bug seems to be here).

Hmmmm.

What if we moved all caller and ImageWriterUtilityClient work to FILE thread?
  - callers are on FILE thread already.
  - mojo utility process client works fine with FILE thread callers.
  - change ImageWriterUtilityClient to expect FILE thread callers.
  - and ImageWriterUtilityClient responses will be on FILE thread, so we ...
  - can remove task_runner_->PostTask from ImageWriterUtilityClient.

Easy to test locally in my WIN10 DEBUG build: make these changes.
  - retain the kDelay >= 600ms
  - re-run the browser test with these changes.
  - can no longer reproduce this  bug 690717  in my DEBUG Win10.

Bug reproduction correctly fingers task_runner_->PostTask() as a problem.
 - Move all work onto the FILE thread, bug no longer reproduces.
  - uploaded a patch to do that.
 - Latent bug maybe, the existing unit test does not test the Cancel case.
 - Browser test I added does test the Cancel case, hence the current bug.
 - Filed bug 703514 about Cancel, production code does not seem to use it.

Other things noted:
 - this component (Platform>Apps>Image Recovery Tool) has no owners.
 - no response from past owners in bug pings [1] or in review [2].
 - maybe just go ahead and delete it all, assign new owners, etc?

[1] https://bugs.chromium.org/p/chromium/issues/detail?id=352442#c18
[2] https://codereview.chromium.org/2734773002

>- Move all work onto the FILE thread, bug no longer reproduces.
  - uploaded a patch to do that.

https://codereview.chromium.org/2756393002

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/54af805eac68ac00ef7a6a9f3ee40fb49894792f

commit 54af805eac68ac00ef7a6a9f3ee40fb49894792f
Author: noel <noel@chromium.org>
Date: Thu Mar 23 01:05:26 2017

Deflake ImageWriterUtilityClient browsertest and clients

The image writer browser tests WriteCancel and VerifyCancel added on
crrev.com/449296 flaked on the Windows DEBUG bots,  issue 690717 .

Re-enable these browser tests. Move all work to the FILE thread both
in the callers (operation.cc, operation_nonchromeos.cc) of the image
writer interface, and in the browser test (to match caller behavior)
to see if this removes the flake on the Windows DEBUG bots (analysis
per  crbug.com/690717#c2  suggests that it should).

The browsertest can't test the low-level USB driver code on any port
since our bots don't have USB sticks. So I manually tested using the
the ChromeOS Recovery Tool [1], and I was able to write and verify a
ChromeOS image to my USB stick using this patch on Windows 10.

Covered by existing tests:
  image_writer_utility_client_browsertest, image_writer_unittest

[1] http://bit.ly/2jFtsZZ

BUG= 690717 , 352442 

Review-Url: https://codereview.chromium.org/2756393002
Cr-Commit-Position: refs/heads/master@{#458965}

[modify] https://crrev.com/54af805eac68ac00ef7a6a9f3ee40fb49894792f/chrome/browser/extensions/api/image_writer_private/image_writer_utility_client.cc
[modify] https://crrev.com/54af805eac68ac00ef7a6a9f3ee40fb49894792f/chrome/browser/extensions/api/image_writer_private/image_writer_utility_client.h
[modify] https://crrev.com/54af805eac68ac00ef7a6a9f3ee40fb49894792f/chrome/browser/extensions/api/image_writer_private/image_writer_utility_client_browsertest.cc
[modify] https://crrev.com/54af805eac68ac00ef7a6a9f3ee40fb49894792f/chrome/browser/extensions/api/image_writer_private/operation.cc
[modify] https://crrev.com/54af805eac68ac00ef7a6a9f3ee40fb49894792f/chrome/browser/extensions/api/image_writer_private/operation_nonchromeos.cc

> Other things noted:
 - this component (Platform>Apps>Image Recovery Tool) has no owners.
 - no response from past owners in bug pings [1] or in review [2].
 - maybe just go ahead and delete it all, assign new owners, etc?
 - maybe create a native app users could download for win / osx

@devlin I'm not sure if these issues are best dealt with on bugs, so maybe add them to your notes.

Cancel tests passing reliably (no evidence of flake) on the WIN DEBUG bots for the last 24 hours.