SIGCHLD race condition?

[Paul Menage]

>
>Can you test against 2.9p2 or the current snapshots.. There has been some
>SIGCHLD changes since 2.5.2pX series.
>


The signal handling strategy has changed, the race condition in
wait_until_can_do_something(), between checking child_terminated and
calling select(), is still there.

I can reproduce exactly the same lockup with RedHat/RawHide 2.9p2. 

Would putting a maximum timeout for select() break anything? If not,
then it would at least prevent the system from deadlocking permanently, 
even if it's not a very elegant fix.

Paul