938889 - IdleService tries to set a pref from a content process when running tests on B2G emulator

Status: RESOLVED FIXED

(Core :: Widget, defect)

Description

[Reuben Morais [:reuben]]

See https://tbpl.mozilla.org/php/getParsedLog.php?id=30577161&tree=Pine&full=1#error1

FWIW, dom/permission/tests/test_idle.html runs earlier in the same chunk.

Comment 1

[Gregor Wagner [:gwagner]]

I think this is the cause for the random orange on M4.
We should try to disable this test:
/tests/dom/permission/tests/test_idle.html

Comment 2

[Jonathan Griffin (:jgriffin)]

I've just disabled this on cedar.

Comment 3

[Andrew Halberstadt [:ahal]]

I haven't seen this error since jgriffin disabled it. Was this debug only though? As I don't think it got disabled for opt yet.

Comment 4

[Jonathan Griffin (:jgriffin)]

It looks like disabling that test did clear up M4, aside from the shutdown crash which bug 934009 addresses.

Comment 5

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I think this service, as it is, should only run in the master process. Child processes should simply forward their requests to the master.

Comment 6

[Robert O'Callahan (:roc) (email my personal email if necessary)]

At the very least, the daily stuff should just be disabled in child processes.

Comment 7

[Reuben Morais [:reuben]]

Attachment: Remote IdleService, disable idle-daily in content processes

Something like this?

Keeping track of an id->pointer map in parent and child is a PITA, does anyonek now if there's an easier way to do that? Is sending |static_cast<intptr_t>(ptr)| too hacky?

Comment 8

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Comment on attachment 8346571 [details] [diff] [review]
Remote IdleService, disable idle-daily in content processes

Review of attachment 8346571 [details] [diff] [review]:
-----------------------------------------------------------------

This is good but I think we can simplify it a bit. Instead of allocating IDs in ContentChild and keeping an ID-to-observer map there, we can just have ContentChild::AddIdleObserver/RemoveIdleObserver cast the observer pointer to a uint64_t and forward directly to the parent. When notified by the parent, we just cast the uint64_t back to a pointer and notify on it. We have to trust the parent process anyway. ContentChild still needs to AddRef and Release the observer to maintain current lifetime semantics.

Comment 9

[Reuben Morais [:reuben]]

Attachment: Remote IdleService, disable idle-daily in content processes, v2

Nice, that makes things significantly simpler! I assumed sending pointers across processes would be frowned upon, but this makes a lot of sense :)

Comment 10

[Reuben Morais [:reuben]]

https://hg.mozilla.org/integration/b2g-inbound/rev/70675d0a0533

Comment 11

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/70675d0a0533