578893 - DEBUG_CC is really crashy

Status: RESOLVED WONTFIX

(Core :: General, defect, P1)

Description

[Robert Sayre]

This means no one is running it. I bet if they did, they'd find some problems with our cycle collection.

Comment 1

[Andreas Gal :gal]

DEBUG_CC massively changes the way CC works and the way we build the model. I am not sure its the right thing to do for anything but leak finding. But anyway, can you post a few stacks?

Comment 2

[Robert Sayre]

(In reply to comment #1)
>
> not sure its the right thing to do for anything but leak finding. 

Yuh. we got some leaks. For instance, it looks like mrbkap forgot to add nsGlobalWindow.mInnerWindow to CC traversal.

Comment 3

[Robert Sayre]

Attachment: a couple stacks

Comment 4

[Andreas Gal :gal]

Don't get me wrong. I am all for leak detection. I am just saying the current DEBUG_CC output is nuts and not very useful for that.

Comment 5

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Yes, it does mean no one is running it.

However, a bunch of it does exist for a reason, and fixing it is probably a lot less work than throwing it out and rewriting it.  Debugging complex leaks in cycle collection is a nontrivial task, and it makes a big difference in a bunch of cases.

Comment 6

[Peter Van der Beken [:peterv]]

I do run with DEBUG_CC and it didn't use to crash for me. The first stack looks like it's from the new compartments code?

Comment 7

[Paul O'Shannessy [:zpao] (not bugmail, email directly)]

Attachment: segfault at shutdown stack

I'm segfaulting at shutdown, which isn't so helpful.

Comment 8

[Peter Van der Beken [:peterv]]

How is that related to DEBUG_CC? It looks like sCollector is corrupt, I don't think we do anything different there if DEBUG_CC is defined.

Comment 9

[Robert Sayre]

now I'm getting 


###!!! ASSERTION: Didn't call FinishTraverse?: '!mCycleCollectionContext', file /home/sayrer/dev/mozilla-central/js/src/xpconnect/src/nsXPConnect.cpp, line 462
Assertion failure: CURRENT_THREAD_IS_ME(t), at /home/sayrer/dev/mozilla-central/js/src/jsapi.cpp:847


we actually need this to work to debug leaks, right?

Comment 10

[Peter Van der Beken [:peterv]]

What's the stack? I'm not seeing that on a trunk DEBUG_CC build.

I don't need this to debug leaks, I use the CC logs which work in any build. But I guess others might prefer the DEBUG_CC output.

Comment 11

[Robert Sayre]

Attachment: calling XPCCallContext::~XPCCallContext on thread 19

Comment 12

[Andreas Gal :gal]

We should do the explain live garbage stuff only from the main thread.

Comment 13

[Robert Sayre]

(In reply to comment #10)
> What's the stack? I'm not seeing that on a trunk DEBUG_CC build.
> 
> I don't need this to debug leaks, I use the CC logs which work in any build.
> But I guess others might prefer the DEBUG_CC output.

How can I turn on those logs?

> (In reply to comment #12)
> We should do the explain live garbage stuff only from the main thread.

I don't see that happening in my stack. Am I mistaken?

Comment 14

[Andrew McCreight [:mccr8]]

RIP DEBUG_CC (bug 845441).  Plus I think I tore out the crashy stuff a year or so ago.