xrealloc does, the previous code was resulting in all sorts of bad memory
references/frees.  This should fix it.  I checked the rest of the code and
these were the only ones that made the mistake.

One uses the job_id while the other uses job_id_str.
Changes to comments only. No changes in logic

Conflicts:
	doc/html/faq.shtml

Describe how jobs could be lost on slurmctld crash
Backport MPI performance FAQ
Fix bad href tag

allocation code. The first would trigger an assertion when trying to
reserve a port one beyond the bitstring range, and the second would
prevent releasing the last port in the range. This patch fixes those
issues.

When "scontrol update job" is executed with a JobName=, allow
filtering by user with the option UserID=

error has happened.

This modifies the reservation test to not fail if node names are of
the form "rack[0-3]_blade[0-63]"

 Added ability for "scontrol update" to references jobs by JobName.
bug 725

Remove redundant non-zero checks on xfree (e.g. "if (x) xfree(x);")
Remove some duplicate code for switch unpack logic for different
  RPC versions.

Conflicts:
	src/plugins/switch/cray/switch_cray.c

Mostly cosmetic changes. The only real problem was adding un/pack
logic support for different versions of Slurm.

The alpscomm prologue and epilogue don't need to be called for batch
steps, so this patch skips it.

This was missed in the original Cray GA patch set. It is some html
documentation for the new switch plugin functions job_step_pre_resume,
job_step_post_resume, etc.

failures. This patch fixes the build.

but confusing users when it appeared in job output. This patch changes
it to a debug message.

down (like during a sync) new events wouldn't be added to the event list.
This patch fixes the bug.

which should just be set to the same value as the memory limit. This
patch sets that parameter.

* Don't propagate virtual memory by default since DMAPP and others require
  unlimited
* Use default performance governor to match current Cray systems
* Only allocate 2 craynetwork resources to KNC nodes since they have half
  the network resources of other nodes.