>The wait() race I referred to before occurs when the child
>calls exit(). At that moment, all of the parent's threads
>are made runnable and the other thread in wait() may
>see the process before the posix_spawn() thread gets
>to dispose of it. Of course, this is what obtains when the
>posix_spawn() call succeeds, but when posix_spawn() fails
>(returns non-zero), the Posix spec says that no child process
>shall be created. The implementation actually creates a
>child process when posix_spawn() fails with an error from
>the child, but if the failing child can be made to disappear
>completely then it is "as if" the child had never been created.
There are a number of other cases where we really want to create a process
in a library where we:
- don't want a SIGCHLD
- don't want "wait on any pid" to reap it
(Think of system, popen, and others were the library really owns the PID)
Casper
