This is not a user-visible feature, but a large internal change for IME pipeline. (See  issue 551193  for details).

Here's the current status:
1) CL has been landed for M50 behind a switch (a big change)
2) Migrating the switch to feature API: https://codereview.chromium.org/1771223002/ (a small change to enable finch experiment)

We're trying to ship it for M51 or later, but 
I'd like to try this out on M50 to get preliminary result.

Grace suggested offline that we run a small experiment code on the first activation for each user on M50.

But I realized that there won't be a lot of information we can get from the failure when it happens. UMA does not seem to give all the information about the framework such as OS version and build, manufacturer, and the stack trace leading to it. (At least not available on uma.googleplex.com and Dremel seems to require special access right to access some of it) I'm worried that there won't be actionable items if we see failures.

In this sense, I prefer to keep the crash path and run finch experiments. What do you think?

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/7ded37535ac881713223cf71d9168887ec60caf3

commit 7ded37535ac881713223cf71d9168887ec60caf3
Author: changwan <changwan@chromium.org>
Date: Wed Mar 09 23:25:12 2016

Migrate *-ime-thread switches to feature API

The new Feature API is the simplest way to run Finch experiments with.
Also, it gets propagated to the renderer.

BUG= 551193 ,  591149 

Review URL: https://codereview.chromium.org/1771223002

Cr-Commit-Position: refs/heads/master@{#380256}

[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/chrome/browser/about_flags.cc
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/browser/renderer_host/ime_adapter_android.cc
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/browser/renderer_host/ime_adapter_android.h
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/browser/renderer_host/render_process_host_impl.cc
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/public/android/java/src/org/chromium/content/browser/input/ImeAdapter.java
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/public/android/java/src/org/chromium/content/common/ContentSwitches.java
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/public/common/content_features.cc
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/public/common/content_features.h
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/public/common/content_switches.cc
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/public/common/content_switches.h
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/content/renderer/render_widget.cc
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/testing/variations/fieldtrial_testing_config_android.json
[modify] https://crrev.com/7ded37535ac881713223cf71d9168887ec60caf3/tools/metrics/histograms/histograms.xml

Requesting merge of the above CL into M50 for the following reasons:
1) Switches used in M50 will no longer apply to M51. If they opt in on M50, they need to opt in again on M51, which could be a pain.
2) For potential finch experiment on M50.

Your change meets the bar and is auto-approved for M50 (branch: 2661)

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21

commit 9d4eb204f3cdf8f282f258aaf1770d6d3727bb21
Author: Changwan Ryu <changwan@google.com>
Date: Thu Mar 10 01:54:36 2016

Migrate *-ime-thread switches to feature API

The new Feature API is the simplest way to run Finch experiments with.
Also, it gets propagated to the renderer.

BUG= 551193 ,  591149 

Review URL: https://codereview.chromium.org/1771223002

Cr-Commit-Position: refs/heads/master@{#380256}
(cherry picked from commit 7ded37535ac881713223cf71d9168887ec60caf3)

Review URL: https://codereview.chromium.org/1784743002 .

Cr-Commit-Position: refs/branch-heads/2661@{#162}
Cr-Branched-From: ef6f6ae5e4c96622286b563658d5cd62a6cf1197-refs/heads/master@{#378081}

[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/chrome/browser/about_flags.cc
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/browser/renderer_host/ime_adapter_android.cc
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/browser/renderer_host/ime_adapter_android.h
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/browser/renderer_host/render_process_host_impl.cc
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/public/android/java/src/org/chromium/content/browser/input/ImeAdapter.java
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/public/android/java/src/org/chromium/content/common/ContentSwitches.java
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/public/common/content_features.cc
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/public/common/content_features.h
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/public/common/content_switches.cc
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/public/common/content_switches.h
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/content/renderer/render_widget.cc
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/testing/variations/fieldtrial_testing_config_android.json
[modify] https://crrev.com/9d4eb204f3cdf8f282f258aaf1770d6d3727bb21/tools/metrics/histograms/histograms.xml

How do you plan to monitor the crash if ever happen due to this? The rate can be low and not surface up to stability sheriff.

We still record UMA before crash:
https://uma.googleplex.com/histograms?dayCount=28&histograms=InputMethod.RegisterProxyView&fixupData=true&showMax=true&filters=isofficial%2Ceq%2CTrue&implicitFilters=isofficial

Actually I'm not sure whether UMA will be recorded when crashing immediately.

We can track crashes from the following link and compare the value against UMA:

https://crash.corp.google.com/browse?q=product.name%3D%27Chrome_Android%27%20%20AND%20custom_data.ChromeCrashProto.magic_signature_1.name%20contains%20%27ThreadedInputConnectionFactory%27&ignore_case=false

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

Grace, there are some number of crashes, but Alex and I could not find any issue when locally testing. The crash report does not contain any user comment so far: https://crash.corp.google.com/dremel_query_ui?q=SELECT%20reportid%2C%20custom_data.ChromeCrashProto.magic_signature_1.name%2C%20usercomments%20FROM%20crash.prod.latest%0AWHERE%20product.name%3D%27Chrome_Android%27%0Aand%20product.name%3D%27Chrome_Android%27%20%20AND%20custom_data.ChromeCrashProto.magic_signature_1.name%20contains%20%27ThreadedInputConnectionFactory%27%20AND%20custom_data.ChromeCrashProto.magic_signature_1.name%3D%27%5BAndroid%20Java%20Exception%5D%20java.lang.AssertionError%20at%20org.chromium.content.browser.input.ThreadedInputConnectionFactory.onRegisterProxyViewFailed(ThreadedInputConnectionFactory.java)%27%0AGROUP%20BY%201%2C2%2C3%0AORDER%20BY%202%2C1%20DESC%0ALIMIT%20100%0A

Also, there is no issue report from the testing team so far. Would it be possible to experiment this on beta, dev, canary to get more user feedback?

Do you want to expand the current experiment? What is the percentage now?

We know before hand that we can't repro the crash. Do we have the idea whether crash happens on certain type of device or OS?

Thanks for the quick response. Currently it is enabled only for those who opted in on "Run IME on its own thread" flag from about:flags. There is no experiment running, and this is the bug to enable the experiment. The crash happens on various OSes and devices.

https://crash.corp.google.com/browse?q=product.name%3D%27Chrome_Android%27%20%20AND%20custom_data.ChromeCrashProto.magic_signature_1.name%20contains%20%27ThreadedInputConnectionFactory%27%20AND%20custom_data.ChromeCrashProto.magic_signature_1.name%3D%27%5BAndroid%20Java%20Exception%5D%20java.lang.AssertionError%20at%20org.chromium.content.browser.input.ThreadedInputConnectionFactory.onRegisterProxyViewFailed(ThreadedInputConnectionFactory.java)%27&ignore_case=false&enable_rewrite=true&omit_field_name=&omit_field_value=&omit_field_opt=%3D

According to the above link, it seems that we have more crashes on M, but I think it just reflects the fact that early adopters and googlers tend to use M more than average users.

ping. Can we get approval for experimenting IME thread on dev and beta? Also, I'd like to experiment 100% on WebView beta as well since there is no finch experiment there.

Sorry for the delay.

Please file a L-R bug to request beta, dev, canary finch trial. Please specify what metrics you are monitoring.

If it goes well, let us do webview beta next.

Thanks, I've filed issue 620662.