when parallel debugger attached to all tasks by blocking on SIGCHLD
   before calling waitpid() (gnats:217)

variable and new sinfo.c module, accept multiple node states, underscore
in header removed, data reported in all lower-case, slight changes to
header values.

as a valid value.

"INFINITE".

in a partition (if it were UP on a Quadrics switch then all nodes would
need to have the same processor count).

This bug was sometimes setting a node's state to DRAINING instead of
DRAINED when there were no jobs on that node.

Perform some general code clean-up in those areas.

Convert xmalloc()/strcpy() to xstrdup().

shutdown only).

end_time to zero (previously was just setting the end_time so it looked
like the job ran for a huge time interval).

while fail/if too many then fatal/sleep and retry

 o Block SIGALRM by default
 o Only allow one thread to enter fwd_signal() at a time.

   blocked signals in threads.

   to "bar = xstrdup(foo);"

   to "bar = xstrdup(foo);":

  - use pthread_sigmask() instead of sigprocmask()
  - save blocked signals just before unblocking them in
    src/srun/allocate.c:allocate_nodes, then reinstate this mask before
    removing handlers.

Define new node configuration parameter for DOWN/DRAIN nodes: Reason=whatever.
Define node reason field for update and get info RPCs.
Slurmctld maintains the node reason, clears when node returned to service,
preserves value across restarts.

Define new node configuration parameter for DOWN/DRAIN nodes: Reason=whatever.
Define node reason field for update and get info RPCs.