312769 - Webshell leak when closing any page with meta refresh

Status: RESOLVED FIXED

Description

[Blake Kaplan (:mrbkap) (inactive, use needinfo)]

I noticed today that visiting http://tinderbox.mozilla.org/SeaMonkey caused us
to leak webshells (the last line I get while shutting down is --WEBSHELL
0x86c1268 == 3).

STEPS TO REPRODUCE:
1. Start browser with the URL http://tinderbox.mozilla.org/SeaMonkey/
2. Close browser.
3. Notice the leaks.

I confirmed that I see this both on trunk and the 1.8 branch.

Comment 1

[Blake Kaplan (:mrbkap) (inactive, use needinfo)]

Created attachment 199858 [details]
testcase

More digging shows that simply loading this testcase and closing the browser
causes a leak.

Comment 2

[Blake Kaplan (:mrbkap) (inactive, use needinfo)]

bz debugged this and pointed his finger at bug 301358. This leak affects any
pages that use <meta http-equiv='refresh'> and bz says it's a big leak, so I'm
nominating this to block 1.8rc1.

Comment 3

[Boris Zbarsky [:bzbarsky, bz on IRC]]

Created attachment 199861 [details] [diff] [review]
Proposed fix

This just makes sure to cancel all the timers in Destroy().

Comment 4

[Darin Fisher]

Comment on attachment 199861 [details] [diff] [review]
Proposed fix

>+    // Cancel any timers that were set for this docshell; this is need
>+    // to break the cycle between us and the timers.
>+    CancelRefreshURITimers();

"this is need[ed] to break..."


>+DoCancelRefreshURITimers(nsISupportsArray* aTimerList)
> {
...
>     }
> 
>+}

nit: kill this blank line


r=darin

Comment 5

[Boris Zbarsky [:bzbarsky, bz on IRC]]

Created attachment 199877 [details] [diff] [review]
Updated to comments

Comment 6

[Boris Zbarsky [:bzbarsky, bz on IRC]]

Comment on attachment 199877 [details] [diff] [review]
Updated to comments

Requesting approval.  This fixes a leak that happens any time the user closes a
window/tab that contains a meta refresh page or navigates away from a page
which has a subframe that has a meta refresh.  We're leaking the docshell and a
bunch of stuff attached to it.

All the patch does is make sure to break the cycle with the refresh timer when
the docshell is being destroyed.  It should be quite safe.

Comment 7

[Boris Zbarsky [:bzbarsky, bz on IRC]]

Fixed on trunk.

Comment 8

[Boris Zbarsky [:bzbarsky, bz on IRC]]

Fixed on branch.

Comment 9

[Scott MacGregor]

making the blocking flag match reality