What appears to have happened is we changed only the non-internal setup.  So
technically this isn't needed, but does help with handling confusion later.

Coverity CID 44977

_alloc_mask() can return NULL, which was not handled.
Coverity CID 44859

Kept separate for clarity.

Bug 4014.

Note: This check is needed since the maximum file path length really
  is file system dependent and not something local to the Linux distro
Coverity CID 45200

Coverity CID 45343

Really should not happen, but this adds checks
Coverity CID 171449

Coverity CID 45167

Coverity CID 44823

Coverity CID 45160

Setting -Werror tricks the test into failing on the

"error: In the GNU C Library, "major" is defined by <sys/sysmacros.h>.
 For historical compatibility, it is  currently defined by <sys/types.h>
 as well, but we plan to remove this soon. To use "major", include
 <sys/sysmacros.h> directly. If you did not intend to use a system-defined
 macro "major", you should undefine it after including <sys/types.h>."

error. Since the normal Slurm build uses -Werror, this warning on including
both headers will then cause the build itself to fail.

Bug 3982.

deadline_time_limit is ensuring this is always set, but the optimizer
appears to not be able to work out the flow here.

Move test above the accounting limit check. Change the test
to verify first_job_ptr instead of pack_job_id to avoid warning
about a possible dereference of an uninitialized variable lower down.
(Note that both pack_job_id and first_job_ptr are set at the same time,
so this test is functionally equivalent but able to convince the compiler
that things will be okay.)

This one's a bit confusing, and should be a false positive, but
initializing the saveptr argument to strtok_r quiets things down.

Treat default case as an error.
Visible with GCC 6.3 and -01.

Pending job with administrator extended TimeLimit beyond partition's
MaxTime remains pending with reason PartitionTimeLimit
bug 4262

The contoller will return node records with a NULL name for nodes that
are hidden. This is so that you can map a partition_info's nodes --
using it's node_inx[] -- to a node record in the returned node_array
from slurm_load_node().

Previously the perl api would leave an undefined object in the the
node_array if the hidden nodes were found before a real node in the
node_array and any hidden nodes at the end of the array from the
controller wouldn't be counted for in the perl node_array. This patch
adds empty hashes for hidden nodes and preserves the record_count and
node_array from the slurmctld.

Bug 4250

The perl api leaves undefined objects in the node_array returned by
load_nodes() for any node that is hidden.

But 4250

With our current packaging approach this appears to be
the preferrable format. We only include a release version
number in the tarball when != 1. It's a rare occurance, but
the spec file needs to be flexible enough to handle it without
modifications.