Fix ExtensionCrashRecoveryTest.TwoExtensionsIgnoreFirst.

Fix ExtensionCrashRecoveryTest.TwoExtensionsIgnoreFirst.

Avoid a race condition in reordering of notification bubbles when canceling them.

BUG=89078
TEST=none

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=112835

[Yoyo Zhou, 2011-12-02 03:17:56]

Actually, this is still failing on Mac for some reason; I need to investigate
tomorrow.

[Yoyo Zhou, 2011-12-02 19:47:27]

On 2011/12/02 03:17:56, Yoyo Zhou wrote:
> Actually, this is still failing on Mac for some reason; I need to investigate
> tomorrow.

Ok, it turns out there is a race with canceling the notification bubble 0 that
behaves differently on different platforms.

Mac: Bubble 0 goes away immediately. Bubble 1 becomes bubble 0. Accepting it
reloads the second extension (as expected).
Win: Bubble 0 doesn't go away immediately. But accepting it still works,
reloading the first extension (not expected, but sneaks through).
Linux: Bubble 0 doesn't go away immediately. Accepting it crashes.

I've changed the ordering to accept 1 before canceling 0, which (fingers
crossed) should not be so unreliable.

[Yoyo Zhou, 2011-12-02 21:20:56]

On 2011/12/02 19:47:27, Yoyo Zhou wrote:
> On 2011/12/02 03:17:56, Yoyo Zhou wrote:
> > Actually, this is still failing on Mac for some reason; I need to
investigate
> > tomorrow.
> 
> Ok, it turns out there is a race with canceling the notification bubble 0 that
> behaves differently on different platforms.
> 
> Mac: Bubble 0 goes away immediately. Bubble 1 becomes bubble 0. Accepting it
> reloads the second extension (as expected).
> Win: Bubble 0 doesn't go away immediately. But accepting it still works,
> reloading the first extension (not expected, but sneaks through).
> Linux: Bubble 0 doesn't go away immediately. Accepting it crashes.
> 
> I've changed the ordering to accept 1 before canceling 0, which (fingers
> crossed) should not be so unreliable.

It passes on win, mac, linux trybots.

[sadrul, 2011-12-02 23:42:11]

On 2011/12/02 21:20:56, Yoyo Zhou wrote:
> On 2011/12/02 19:47:27, Yoyo Zhou wrote:
> > On 2011/12/02 03:17:56, Yoyo Zhou wrote:
> > > Actually, this is still failing on Mac for some reason; I need to
> investigate
> > > tomorrow.
> > 
> > Ok, it turns out there is a race with canceling the notification bubble 0
that
> > behaves differently on different platforms.
> > 
> > Mac: Bubble 0 goes away immediately. Bubble 1 becomes bubble 0. Accepting it
> > reloads the second extension (as expected).
> > Win: Bubble 0 doesn't go away immediately. But accepting it still works,
> > reloading the first extension (not expected, but sneaks through).
> > Linux: Bubble 0 doesn't go away immediately. Accepting it crashes.
> > 
> > I've changed the ordering to accept 1 before canceling 0, which (fingers
> > crossed) should not be so unreliable.
> 
> It passes on win, mac, linux trybots.

I think someone from extensions/OWNERS should have a look :)

[Yoyo Zhou, 2011-12-02 23:46:56]

Okay. Aaron, can you please review?

[Aaron Boodman, 2011-12-03 00:16:51]

lgtm