(In reply to :Gijs (back Jan 2; he/him) from comment #1)
Sorry, I don't understand the question because there's a load of context I don't have.
Bug 1548941 landed in May. So I have questions, like, do we currently try to run marionette in non-e10s mode on infra or locally? If so, why? And why wasn't this picked up sooner? And, does this affect e.g. the non-e10s mochitest runs that we currently run on infra, (AIUI) on top of marionette? If not, why not?
I found the problem 3 days ago when I noticed that we force all consumers of Marionette to run in non-e10s mode by default due to setting
False. This should have been removed a long time ago. So I fixed that over on bug 1606325. (Note that Marionette still supports older Firefox releases not including your patch from bug 1548941)
While verifying that we can still run in both modes and that the
--disable-e10s mach option can turn e10s off, I just noticed the problem. Nothing in CI uses this option, so it just happens for locally run tests. And as such we never noticed that before.
(In reply to Henrik Skupin (:whimboo) [⌚️UTC+1] from comment #0)
This is because Marionette and geckodriver are not covered by
xpc::IsInAutomation because they are also running outside of our tree.
This is confusing - is this only the case outside of the tree, or is
isInAutomation also false when marionette/geckodriver run inside Firefox's automation? The exception from comment #0 seems to be about in-tree, so I'm confused why that wouldn't be covered by
Marionette tests don't run with
MOZ_DISABLE_NONLOCAL_CONNECTIONS set (bug 1272255) and as such
isInAutomation is always false due to https://searchfox.org/mozilla-central/rev/331f0c3b25089c9a16be65f4dc8c601aeaac8cc4/js/xpconnect/src/xpcpublic.h#683. So yes, we hit this situation also all the time in tree. Even if Marionette would support it the Firefox ui remote test jobs, which have to allow external addresses, would still be affected.
We had a similar situation with updates over on bug 1508726
Gijs, given that we got fixes for other test harnesses, I assume we should also fix Marionette?
I'm not sure that assumption holds. Why does Marionette/Geckodriver need to run with e10s turned off?
I think it's a fair question, and we could consider not allowing that anymore. Personally I would be all for that. But I would need the buy-in from the team. Andreas, what's your take here?
Also, is that error actually about the behaviour caused by bug 1548941 ? We still "allow" you to set the pref you cite, it will just not have any effect. What is marionette checking in terms of configuration?
It ensures that the expected e10s mode is set. It does it via