Hi, the default of 'catch_system_errors = yes' when using the boost unit testing framework is a bit problematic. Apparently, it installs a signal handler for SIGCHILD. I didn't notice this under Linux. But under Solaris I've got an interrupted waitpid system call in one test case[1]. Well, and the trace looks really funny, if waitpid is interrupted by ... SIGCHILD! (the testcase basically does a fork/exec/waitpid) Thus, the default of this parameter is at least a bit counter-intuitive. Perhaps, you[2] consider to change this default in the future. Best regards Georg Sauthoff [1] I assume, that utf does this on Linux in such a way, that waitpid is automatically resumed [2] I mean the utf maintainer, using gmane, I am not able to CC ...