[Boost-testing] Re: Re: [thread] test_read_write_mutex breaksregressiontesting

From: Gennadiy Rozental (gennadiy.rozental_at_[hidden])
Date: 2005-04-26 14:40:26

> It found a number of problems with test_mutex and
> test_read_write_mutex, which seem to be triggered by Boost.Test making
> use of a string after it has been freed.

> Attached is the Purify output from test_mutex and test_read_write_mutex.

What I see in these logs doesn't make too much sense ... for single threaded 
application. Multithreaded situation is a different story. But Boost.Test 
*is not thread-safe* (and never stated to be). There is no single mutex/ or 
any other kind of guard in Boost.Test code anywhere. Boost.Test code is 
reentrable, which is according to what I remember should've been enough for 
Thread library unit tests to work properly (according to Thread library 
author).
   I do plan to remedy this situation. But there are some decisions to make. 
Specifically what kind of primitive to use. Could Boost.Test afford 
dependency on Boost.Thread e.t.c. Any help in this direction is greatly 
appreciated.

Regards,

Gennadiy