Bruno Martínez a écrit : 
On Thu, 22 Jun 2006 06:15:58 -0300, Eric Colleu  
<ecolleu@besancon.parkeon.com> wrote:

  
Hi,

I'm using boost 1.33.1 on multi-processor plateform (2 Pentium4 3.2 GHz)
and Windows XP SP2 (compiled with MinGW).
I wrote a MessageQueue class which use boost::mutex and boost::condition
to wait for message and notify message reception (see code below)
I made a little test that create 2 boost::thread sending and receiving
messages in 2 queues.
    

I think you should poet the test.

Bruno

_______________________________________________
Boost-users mailing list
Boost-users@lists.boost.org
http://lists.boost.org/mailman/listinfo.cgi/boost-users
OK i found it finally.
In GCC (under Linux and Windows), there is an access to the GLIBCPP_FORCE_NEW or GLIBCXX_FORCE_NEW environment variable (use a memory cache by default instead of calling malloc / free directly). On a multiprocessor system, this "optimisation" seems to be unstable. I set this variable and the problem disapear.
So the problem is not in boost, sorry for my mistake.

Eric