(No comment was entered for this change.)

 Issue 570343  might be related.

> It takes up 100% CPU and a while to come back to control.

Just to make sure, is it Chrome that consumes 100% CPU time?   Since IME in OS X is built on top of multi process architecture, understanding what component gets stuck would be important to figure out what's going on.

No, it's Inbox tab rather than Chrome itself.

	taskmanager.png 59.0 KB   View   Download

 Issue 570343  has been merged into this issue.

I'm suspecting this is caused by complex path r361474 but bisect would be helpful.

Tested on Mac OS 10.11 using chrome canary M49 - 49.0.2595.0 and observed that typing Japanese in Google Inbox takes up 100% of CPU usage. 

Bisect Information:
===================
Good build: 49.0.2573.0  
Bad Build : 49.0.2575.0 

Chromium builds are not invoking while running python tool bisect, so providing the manual changelog below as per omahaproxy.

Changelog URL: https://chromium.googlesource.com/chromium/src/+log/49.0.2573.0..49.0.2575.0?pretty=fuller&n=10000

From the above change log suspecting below
Suspecting:  r361474

Review URL: https://codereview.chromium.org/1474673003

eae@ - Could you please check whether this is caused with respect to your change, if not please help us in assigning it to the right owner.

Thanks!

Thanks you for the bisect, that I can confirm that this is the same issue as I'm working on.

Re #7:

Can we keep this bug open as  Issue 569572  and  Issue 570229  have different exit criteria?   As for this ( Issue 569572 ), addressing the performance regression is the primary goal.  For instance, just reverting that CL might be sufficient to close this issue and we can track in which revision the issue is addressed.

#8: Agreed, changed back to Assigned.

I'd appreciate if you could give us some time to try to fix this issue properly. This is our multi-years project, we've been trying to address and fix several performance issues in the past and finally enabled the flag in r361474. It was unfortunate that we missed this case before enabling it; actually, this has been a suspect for us, but I could not prove any significant performance hit in any other tests.

 Bug 565902  is yet another case we're seeing the performance regression for CJK, and my WIP patch for  bug 570229  fixes it.

Hopefully I can land the fix for  bug 570229  next week, it should not take long until we can figure out if it can resolve this issue too. Thank you for your understanding in advance.

Re #9:

Thank you for reopening this.  I'll defer to you guys development team and release team as I'm no longer actively working for Chromium project.

That said, please keep in mind that our dogfooders are still having troubles because of this.  If we were losing dogfooders because of this, it could be risking not only  Issue 404597  but also other on-going efforts in Chromium projects.  I'd still recommend reverting the change that flipped the flag until we address all the performance issues.

Thank you, and yes, I take this situation seriously. I deeply apologize for not finding and responding to this earlier.

eae@ should be back from off on Monday and the CL is ready for review, I'll make sure to contact him on this issue.

(No comment was entered for this change.)

Confirmed gmail issue was fixed in today Canary 49.0.2599.0 (r366505), so resolving as dup.

If you still see issues, or think that the improvement isn't enough, could you please comment or reopen? I'd be happy to investigate further.

Apologies again for taking long to fix and troubled your dogfooding.