SIGCHLD race condition?

[Paul Menage]

>I see. Well, it seems that the fundamental problem is that select()
>can't be interrupted when the child exits, no? You can't have SIGCHLD
>unblock when select() is called...

The SIGCHLD has already arrived by the time select() is called - it
arrives in the gap between checking child_terminated and calling
select(). It's a very small race, but one that we've seen occur
sometimes. (Or if this isn't the race that's biting us, there's another
race that can produce exactly the same symptoms.)

Paul