398462 - Leak nsBaseAppShell and nsRunnable with <iframe src="aim:foo"> and goQuitApplication

Status: RESOLVED WORKSFORME

(Toolkit :: Startup and Profile System, defect)

Description

[Jesse Ruderman]

Steps to reproduce:
1. Save the testcase as "a3.xhtml".
2. Edit the testcase to point to a URL that calls goQuitApplication (and is allowed to do so).
3. Exit Firefox.
4. export XPCOM_MEM_LEAK_LOG=1
5. Run a debug build of Firefox with "a3.xhtml" on the command line.

The testcase will automatically instruct Firefox to quit about a second after the browser window appears.

Result: 1 nsBaseAppShell and 1 nsRunnable leak.

Note that if bug 167475 is fixed, the testcase will stop working.

I have Adium installed but I have not instructed Firefox to allow clicks on aim: links.  This results in a protocol-handling dialog appearing, but somehow *behind* the browser window.  CCing sdwilsh on suspicion that this leak involves the new protocol-handling dialog (since he added it in bug 385065).

Comment 1

[Jesse Ruderman]

Attachment: testcase

Comment 2

[Jesse Ruderman]

When I log nsBaseAppShell or nsRunnable, the leak goes away!  (Logging a non-existent class doesn't make it go away.)  Here's what I'm trying to use:

  export XPCOM_MEM_LEAK_LOG=2
  export XPCOM_MEM_REFCNT_LOG=log-file.dat
  export XPCOM_MEM_COMPTR_LOG=log-comptr.dat
  export XPCOM_MEM_LOG_CLASSES=nsBaseAppShell

Comment 3

[Boris Zbarsky [:bzbarsky]]

Does it matter which objects in that class you log?  Sounds like a possible timing bug, with the logging affecting the timing....

Comment 4

[Jesse Ruderman]

There's only 1 nsBaseAppShell allocated according to the "Total" column, so I can't use XPCOM_MEM_LOG_OBJECTS  with a specific serial number to avoid timing issues.

Comment 5

[Jesse Ruderman]

Attachment: balance tree for leaked nsRunnable

I got a balance tree for the leaked nsRunnable (mDummyEvent) using XPCOM_MEM_LOG_OBJECTS=262.

It does seem to be a timing issue, but adjusting the timeout in the testcase doesn't help :(

Comment 6

[Jesse Ruderman]

Attachment: testcase 2

I managed to reproduce with logging on with this nastier testcase, after increasing THREAD_EVENT_STARVATION_LIMIT from 20ms to 200ms.

Comment 7

[Jesse Ruderman]

Attachment: balance tree for leaked nsBaseAppShell

Comment 8

[Jesse Ruderman]

WFM on mozilla-central.  Had to tweak the testcase a bunch to get it to work with my new setup involving Quitter.

Comment 9

[Jesse Ruderman]

Attachment: testcase that can be used with Quitter (no repro)