1028511 - Assertion failure: "if paused, aTime must be at least mPauseStart"

Status: RESOLVED FIXED

(Core :: CSS Parsing and Computation, defect)

Description

[Jesse Ruderman]

Attachment: testcase

Assertion failure: !IsPaused() || aTime >= mPauseStart (if paused, aTime must be at least mPauseStart), at layout/style/AnimationCommon.h:335

Comment 1

[Jesse Ruderman]

Attachment: stack

Comment 2

[Brian Birtles (:birtles)]

I spent some time tracking this down and it appears the problem arises because times from the refresh driver can go backwards putting the time recorded for pause start in BuildAnimations *later* than the time used in FlushAnimations.

This seems to occur when we create the window because inside the refresh driver when we switch from the throttled-rate timer back to the regular-rate timer. I think we should probably make the refresh driver guarantee that time never goes backwards by clamping times if necessary. This still won't be true when the refresh driver is under test control but long-term I hope to fix that too and make sure in that case too times never go backwards.

Comment 3

[Brian Birtles (:birtles)]

Attachment: Don't let refresh driver time go backwards when changing timers

I've spent a few hours trying but I can't seem to turn the attached test case
in a crashtest that fails correctly. For now, though, this patch fixes the
problem for me.

Basically, as described in comment 2, the popping up a window and closing it
again, causes us to switch timers which leads to the refresh driver time going
backwards just briefly but in the interim time we sample the paused animation at
a time before its pause start.

This patch simply clamps the refresh driver time so that when we change timers
we don't go backwards. This should be sufficient because the only other times we
update the refresh driver time are when we are under test control (which needs
to be fixed so that it also never goes backwards but that's a whole lot of
separate work) or when we get a tick in which case we use the current timestamp
which should always be >= the previous refresh time.

Comment 4

[Brian Birtles (:birtles)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/ccdb8d43db9f

Comment 5

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/ccdb8d43db9f