+xiyuan - can you think of recent login changes that might do this?

Chrome should start in stub user mode. IIRC, that part of code is not changed recently. I can run in this mode on my dev box. Maybe it just happens on the device.

Any theories why is this happening/what code to look at in order to try to fix this?

Has this worked before on a real device?

Looking at code in ChromeBrowserMainPartsChromeos::PreEarlyInitialization [1]. The stub user switches are only added when NOT running on a real device.

[1]: https://cs.chromium.org/chromium/src/chrome/browser/chromeos/chrome_browser_main_chromeos.cc?rcl=19c5a8a7ff6d04e4688a1a88a3d57ef620fab6ab&l=536-550

I've never used it, but it's in our "Getting Started" guide for deploying builds to a chromebook, so I am assuming it did work at some point. James might know for sure.

Regarding the code you mentioned - if I remove !base::SysInfo::IsRunningOnChromeOS() - should the switch start working on device? (I am not suggesting landing it, just want to get around login).

It appears that is not enough. The current stub user is "stub-user@example.com", which fails policy code when it tries to get user policy from session_manager daemon. I got the following:

[10660:10827:0614/094933.510146:ERROR:object_proxy.cc(573)] Failed to call method: org.chromium.SessionManagerInterface.RetrievePolicyForUser: object_path= /org/chromium/SessionManager: org.chromium.SessionManagerInterface.SessionDoesNotExist: Cannot retrieve user policy before session is started.
[10660:10660:0614/094933.510199:ERROR:user_cloud_policy_store_chromeos.cc(121)] Session manager claims that session doesn't exist; signing out

I worked around the problem by prefixing the domain name "consumer." (i.e. stub-user@consumer.example.com"), which disables user policy code.

https://codereview.chromium.org/2938893002/ shows what is needed to get stub user work on device. This is not a well supported code path though. Think stub-user was intended to run on dev box and for tests.

Awesome, thanks, this worked for me!
I will update our guide, and I guess we can close this issue, since you are saying this is not an unsupported code path.

This did work at one point, but I've only used it once and I don't think it's a workflow we want to support on-device. I agree with WontFix.