It seems that the combination of Jeff T's and my inheritence patches collude to crashing the cgid worker.
If we fail to exec(), we end up calling log_error_core, which causes that fork()ed cgid worker to SIGPIPE and die. Apparently we don't have a valid fd to the server's error log at that point. Any volunteers to help track down this regression before we roll out this release??? This is the last unfixed regression/showstopper that either Sander Temme or I have identified. Bill
