RE: [reSIProcate] Weird problem with Mutex

["Matthias Moetje - TERASENS GmbH" <moetje@xxxxxxxxxxxx>]

Alexander and Matt,

 

 

thanks very much for checking this. I have now re-downloaded

the complete stack, auto-upgraded the VC7.1 files and with this

setup the error went away.. :-)

 

Since the code files are identical, the problem must be due to

the compilation settings. Still strange how compilation settings

can make this kind of error happen, but at least I have a better

idea now about where the problem lies...

 

Thanks again and best regards,

Matthias Moetje

TERASENS GmbH
Augustenstraße 24
80333 Munich
GERMANY

Phone: Fax: e-mail: Web:

+49.89.143370-0 +49.89.143370-22 info@xxxxxxxxxxxx www.terasens.com

 

---

From: resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxx [mailto:resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Alexander Altshuler
Sent: Wednesday, June 21, 2006 8:20 AM
To: resiprocate-devel@xxxxxxxxxxxxxxxxxxx
Subject: RE: [reSIProcate] Weird problem with Mutex

It works. VS 7.1

 

-----Original Message-----
From: resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxx [mailto:resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Matthias Moetje - TERASENS GmbH
Sent: Wednesday, June 21, 2006 4:51 AM
To: resiprocate-devel@xxxxxxxxxxxxxxxxxxx
Subject: RE: [reSIProcate] Weird problem with Mutex

 

Could perhaps anyone using Visual Studio 8.0 (or maybe 7.1) be so kind

to paste these few lines of code into basicRegister to see if this problem

can be reproduced if compiled in release config?

Thanks,

Matthias

 

---

From: resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxx [mailto:resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Matthias Moetje - TERASENS GmbH
Sent: Friday, June 16, 2006 11:22 PM
To: resiprocate-devel@xxxxxxxxxxxxxxxxxxx
Subject: [reSIProcate] Weird problem with Mutex

Hi,

 

I was finally able to find a minimal piece of code to reproduce

an issue I am fighting against during the last days. To reproduce

you just need to paste the following code to the start of the

main() function in basicRegister for example:

 

---

#include "resip/dum/UserAuthInfo.hxx"

---

resip::SipStack* mStack;
resip::DialogUsageManager* mDum;

mStack = new resip::SipStack();
mDum = new resip::DialogUsageManager(*mStack);

mDum->addTransport(UDP,5060);
mDum->addTransport(TCP,5060);

UserAuthInfo* msg = new UserAuthInfo("asd","asd","asd","asd");
mDum->post(msg);

---

Now, the following happens during mDum->post:

in TransactionUser::post(Message* msg), mFifo.add is called.

then in TimeLimitFifo<Msg>::add it constructs the lock object:

Lock::Lock(Lockable & lockable, LockType lockType)

in this constructor, I get an access violation when it tries to
call myLockable.lock();

When I examine the myLockable I can see that the virtual function
pointers seem corrupted:

Usually (that means, when post is called from the dum or stack 
thread in my application) in the vfptr[] array I can see valid pointers,
but in this case they are wrong.

The mID (Windows Critical Section structure) member of the
Mutex (which inherits from Lockable) seems to remain OK,
just the vfptr to the base methods  is wrong.

I don't really have an idea why this is happening and I feel like
banging my head gainst the wall. In my application everything works
fine. If I put a breakpoint in one of these functions during normal
operation (inbound, outbound calls, client registration etc.), the
lockable is always OK, but as soon as I call mDum->post
manually, the lockable seems to be damaged and I get an access
violation exception.

If I compile the application in debug mode, the problem does not exist!

I am using VS 2005. Here are the relevant configuration settings I
am using for basicRegister:

Optimization: Disabled
Inline Function Expansion: Default
Favor Size or Speed: Neither
Omit Frame Pointers: No
Enable String Pooling: No
Runtime Library: Multi-threaded DLL
Buffer security check: Yes

Linker:

Generate Debug Info: Yes
References: Default
Enable COMDAT Folding: Default
Link time Code Generation: Use Link Time CG

I would be very thankful for any comments as I am running out of
ideas on this subject (apart from implementing a simple critical
section directly without the lockable and mutex objects).

Thanks and best regards,

Matthias Moetje

TERASENS GmbH
Augustenstraße 24
80333 Munich
GERMANY

 

Phone: Fax: e-mail: Web:

+49.89.143370-0 +49.89.143370-22 info@xxxxxxxxxxxx www.terasens.com