Repro video: https://storage.cloud.google.com/chromiumos-test-logs/bugfiles/cr/902157/IMG_2697.MOV?_ga=2.227832469.-967102564.1532641052

This is when we update from N to P 

Is this reproducible on any other devices or is it Eve specific?

Issue is currently marked as RBD. Next Dev is targeted for Thurs 11/8.

As noted in c#2, issued reproduced on Eve device when updating from M71 (Eve has ARC++ N) to M72 - 11231.0.0 (Eve has ARC++ P)

Issue not reproduced on kevin device. (has ARC++ N on both M71 and M72 - 11231.0.0)

Adding a couple people to CC to help find an owner.

My eve M72 is not very stable either (crashes), I didn't upgrade from one ARC version to other, just got an image from goldeneye. Cros 11238, do we have logs from your device sdantuluri@ ?

Debug logs: https://storage.cloud.google.com/chromiumos-test-logs/bugfiles/cr/902157/debug-logs_20181107-165341.tgz?_ga=2.166499766.-967102564.1532641052

comment #4
Are we sure this is caused by ARC++? It is possible for ARC++ to cause Chrome browser process crash, yes, but it's relatively rare.

+ARC constables

I currently don't have cycles so adding ARC++ constables o/

probably dup: https://bugs.chromium.org/p/chromium/issues/detail?id=889552

Took a quick look at the debug logs at #c7

In var/log/ui, there were two error messages (1) "Failed to retrieve primary account info" and (2) "Instance for arc.mojom.NotificationsInstance::SetNotificationConfiguration version mismatch" that appeared right before "Unexpected crash report id length". I doubt (2) could cause a crash, so may be (1) is the cause of the crash?

In arc.log, there were no fatal errors except for "Aborting setup flow because the parent died" (3)

---- var/log/ui/ui.20181107-165207
(1)
[7556:7556:1107/165259.406031:ERROR:service.cc(232)] Failed to retrieve primary account info.
Unexpected crash report id length
System crash_reporter failed to process crash report.

(2)
[8375:8375:1107/165309.787963:ERROR:connection_holder.h(291)] Instance for arc.mojom.NotificationsInstance::SetNotificationConfiguration version mismatch. Expected 22 got 21
Unexpected crash report id length
System crash_reporter failed to process crash report.

---- var/log/arc.log
(3)
2018-11-07T16:53:26.613547-08:00 ERR arc-networkd[10593]: [ERROR:ip_helper.cc(62)] Aborting setup flow because the parent died

Is anyone cc'ed here a possible owner for this issue?

+assistant folks for (1) since that's in their code

+yoshiki for (2)

+hugobenichi for (3).  I'm assuming that the problem isn't really arc-networkd and the browser and other scaffolding is falling down around it, but just in case

Eve devices haven't had a M72 dev channel release yet.  Branch is in two weeks and Beta very soon after.  This needs to get in someone's queue before too long or we'll have no soak time in dev channel.

As for (2), mitsuruo@ starts merging the CL:
https://chromium-review.googlesource.com/c/chromium/src/+/1333088

+linben fyi this will prevent moving eve-arcnext to master. 

We're about to release our 3rd M72 dev channel this week - without eve.  How do we get more attention on this and at what point does this become a P0?

I did a quick local test with dhaddock@:

1) flashed 11231.0.0 non-test image, skipped Google Assistant setup
2) dhaddock@ pushed 11261.0.0 / 72.0.3609.3 to test omaha
3) Turned on Dev Mode, followed steps in #1 to AU
4) Successfully AU'd to 11261.0.0, verified that Android Settings opens fine (so ARC++ is doing ok), and no Chrome browser crash loop

but -
5) Go to Settings, toggle Assistant on, instant Chrome crash

But, I don't get a crash loop. Everything's ok afterwards, and Assistant is still off. Everytime I toggle it on will crash it.

re: Assistant crash in #19 
We discovered a clang CFI error caused by a new class shared between libassistant.so and Chrome. The cl introduced the new class is reverted  https://chromium-review.googlesource.com/c/chromium/src/+/1335962

@vaandres and I verified that crash loop is not reproduced if assistant was not enabled on M71 before AU to M72

But with assistant enabled on M71, and then AU to M72, crash loop is still reproduced on RC build 	11264.0.0, 72.0.3609.3

Issue 905874 has been merged into this issue.

sdantuluri@ can you verify on build 11265.0.0, this is the first build has the revert CL.

Please do confirm that the 11265.0.0 72.0.3611.0 version on https://cros-goldeneye.corp.google.com/chromeos/console/listBuild?milestone=72#/ is good.

+xiaohuic, I don't understand why you can't verify build 11265.0.0 contain the reverted CL ?

I can, I assume my test won't be "final" as we need our QA to verify the bug. Also I already verified the revert locally. I can verify build 11265 again and let you know.

flashed to eve latest-beta R71-11151.33.0, turn on assistant, flashed to eve latest-dev R72-11267.0.0, no crash

Observing this issue on Nocturne too, after channel change from M71 stable (11151.33.0, 71.0.3578.57) to M72 dev (11264.0.0, 72.0.3609.3).

Verified that this issue is fixed on ToT 11265.0.0, 72.0.3611.0.

AU'd from M71 beta (with assistant enabled) to M72(11265.0.0) and there was no browser crash. Also enabling/disabling Assistant from Settings works fine without any crash.