420150 - Cycle collector fault when launching runtests.pl

Status: RESOLVED WORKSFORME

(Core :: XPCOM, defect)

Description

[Ben Turner (not reading bugmail, use the needinfo flag!)]

sdwilsh said he has seen this fault while running runtests.pl:

multiple registrations of language runtime (ptr: 4a11cd0)

Worth looking into.

Comment 1

[Jesse Ruderman]

This happens for me, too, every time I run "perl runtests.pl".  This is frustrating because I'm trying to debug another problem that might be specific to runtests.py.

Comment 2

[Jeff Walden [:Waldo]]

I've seen this running make check before; a make build will fix the matter.  I don't recall seeing it in Mochitests, but I might just not be remembering.

Comment 3

[Mark Banner (:standard8)]

(In reply to comment #0)
> sdwilsh said he has seen this fault while running runtests.pl:
> 
> multiple registrations of language runtime (ptr: 4a11cd0)

I'm seeing this frequently on my local mac builds of Thunderbird (OS X 10.5). Though I've not always been running tests.

The sort of situations it seems to occur (and I can't quite say the definitive possibilities yet):

* Rebuilding parts of Thunderbird just doing mailnews and mail, though most of the time this seems ok.
* Messing around with profiles, such that a default doesn't exist, or trying to load a profile that doesn't exist.
* After running xpcshell tests (via make check or check-interactive).

Comment 4

[Ben Turner (not reading bugmail, use the needinfo flag!)]

I know what causes this.

Comment 5

[Ben Turner (not reading bugmail, use the needinfo flag!)]

Attachment: Patch, v1

The problem was that we were getting a relative path in the module loader hash which was causing the xpconnect module (among others!) to initialize twice.

Comment 6

[Benjamin Smedberg]

Comment on attachment 339138 [details] [diff] [review]
Patch, v1

You really don't need the extra COMPtr, you could just do (*aDirectory)->Normalize() but this is ok.

Comment 7

[Ben Turner (not reading bugmail, use the needinfo flag!)]

But then your static checker would yell at me for setting an outparam and then maybe returning failure, right?