Jesse Ruderman Reporter
	Description • 11 years ago

With these prefs:
  user_pref("layers.acceleration.disabled", true);
  user_pref("layers.force-active", true);

the testcase makes Firefox infinitely recurse due to a cyclic layer tree.

	Corey Ford [:coyotebush]
	Comment 2 • 11 years ago

Caught that cycle being created. This presumably has something to do with how we process all descendants of a fixed position layer specially, and a bad interaction of that with sticky position layers.

	Jesse Ruderman Reporter
	Comment 3 • 11 years ago

Should that assertion be added to mozilla-central?

	Kyle Wooten
	Comment 4 • 11 years ago

Thanks!

	Corey Ford [:coyotebush]
	Comment 5 • 11 years ago

(In reply to Jesse Ruderman from comment #3)
> Should that assertion be added to mozilla-central?

Sure. I had been thinking to add it in the same patch that actually fixes this problem, but then I never made any headway on figuring out the cause of the bug. I think this will need input from someone who knows the layer system better than I do.

	Ben Kelly [:bkelly, not reviewing]
	Comment 6 • 11 years ago

It seems like this should block bug 916315, no?

	Corey Ford [:coyotebush]
	Comment 7 • 11 years ago

I don't reproduce either this or bug 931450 in the latest Nightly. Discussion on bug 919156 suggests that they could have been fixed in bug 919144?

	Daniel Holbert [:dholbert]
	Comment 8 • 11 years ago

(In reply to Corey Ford [:corey] from comment #7)
> I don't reproduce either this or bug 931450 in the latest Nightly.
> Discussion on bug 919156 suggests that they could have been fixed in bug
> 919144?

Nice, yup.

I verified that bug 919144 did indeed fix this, using targeted builds. (Its parent cset, 21df28ade757, reproduces this bug for me -- whereas the cset for the last patch in bug 919144 does not reproduce this bug for me.)

 --> Resolving as FIXED by bug 919144.

	Daniel Holbert [:dholbert]
	Comment 9 • 11 years ago

(Also: I'm using 64-bit Linux --> broadening Platform from Mac to All)

	Paul Silaghi, QA [:pauly]
	Comment 10 • 11 years ago

Reproduced in Nightly 2013-09-25.
Verified fixed 28.0a1 (2013-12-05), Win 7 x64

	Mats Palmgren (inactive)
	Comment 11 • 10 years ago

Landed the crashtest:
https://hg.mozilla.org/integration/mozilla-inbound/rev/a1c8f8c10088

	Ryan VanderMeulen [:RyanVM]
	Comment 12 • 10 years ago

https://hg.mozilla.org/mozilla-central/rev/a1c8f8c10088

	Milan Sreckovic [:milan] (needinfo for best results)
	Comment 13 • 10 years ago

Pretty sure this test never actually tested the correct configuration.  The layers.acceleration.disabled preference is queried once and the value reused, so setting it for a particular test is just ignored.

	Milan Sreckovic [:milan] (needinfo for best results)
	Comment 14 • 10 years ago

I can't build this far back, but certainly in today's builds, this does not run the test with layers.acceleration.disabled actually set, as the pref() line in crashtests.list sets it too late to get picked up.  And if it was getting picked up, it would mean that the value of that preference changes mid execution, which is at best scary and certainly not supported.