Re: cleanup handlers and sem_wait ....

[webmyster at laposte dot net]

webmyster@laposte.net wrote:

> Jakub Jelinek wrote:
>
> > On Tue, Mar 22, 2005 at 03:03:14PM +0100, webmyster@laposte.net wrote:
> >  
> >
> > > I have a problem with cleanup handlers and sem_wait with the small 
> > > program attached under linux.
> > > On a first computer (glibc 2.3.4, gcc 3.3.3, kernel 2.6.8.1), when I 
> > > compile it with g++, although the thread seems to close, the cleanup 
> > > handler is not call. When I compile the programm with gcc, no 
> > > problem occurs : the cleanup handler is call and the thread stop.
> > >   
> >
> >
> > Works just fine here, glibc 2.3.4, gcc 3.4.3, kernel 2.6.10,
> > with all of
> > gcc
> > g++
> > gcc -I/usr/include/nptl -L/usr/lib64/nptl
> > g++ -I/usr/include/nptl -L/usr/lib64/nptl
> >
> > testCancel is called in all cases.
> >  
> I updated my system : gcc 3.4.3, glib 2.3.4 (compiled with gcc 3.3.3), 
> linux kernel 2.6.11.4 (also compiled with gcc 3.3.3).
> The same problem occurs !
>
> Actually, it seems that the problem comes from the definition of 
> pthread_cleanup_push and pthread_cleanup_pop.
> I modified my pthread.h file. Now, it don't use the first definitions 
> that use the __cleanup__ C attribut (through classes or explicit 
> definition). It is using the __pthread_register_cancel and 
> __pthread_unregister_cancel functions.
> With this modification, the program is working just fine : the 
> testCancle is called.

I've made other investigations : it seems that the __attribute__ 
((cleanup(function))) is not respected !
In other words, whenever I do pthread_cancel, the stack is released, but 
the "cleanup functions" associated with static datas are not called.
I think that the destructor of __pthread_cleanup_class is not called, so 
my cleanup function is not called ...

Damien Touraine