Logic used to identify cores which are usable with each GRES weas incorrect.
bug 905

QOS had a GrpNodes set.

Fix for core-based advanced reservations where the distribution of cores
across nodes is not even. Failing test case:
system has 10 nodes, 1 of which is fully occupied
create reservation with 9 nodes and 10 cores
always would fail with "busy nodes" error

The hostname was being set as an HDF5 value before the field was
set in slurmstepd, resulting in SEGV. This change sets hostname
before the HDF5 call and also tests for NULL before trying to set
the value. Backtrace of failure:
(gdb) bt
0  strlen () at ../sysdeps/x86_64/strlen.S:106
1  0x00007f4af5bb756d in put_string_attribute (parent=33554432,
   name=0x7f4af5bb8591 "Node Name", value=0x0)
   at src/plugins/acct_gather_profile/hdf5/hdf5_api.c:1711
2  0x00007f4af5bad224 in acct_gather_profile_p_node_step_start (job=0x194cc80)
   at src/plugins/acct_gather_profile/hdf5/acct_gather_profile_hdf5.c:372
3  0x000000000052ca86 in acct_gather_profile_g_conf_set (tbl=0x194cc80)
   at src/common/slurm_acct_gather_profile.c:490
4  0x000000000042e028 in batch_stepd_step_rec_create (msg=0x194d280)
   at src/slurmd/slurmstepd/slurmstepd_job.c:496
5  0x0000000000426ae5 in mgr_launch_batch_job_setup (msg=0x194d280, cli=0x194bec0)
   at src/slurmd/slurmstepd/mgr.c:422
6  0x00000000004263da in _step_setup (cli=0x194bec0, self=0x0, msg=0x194bd90)
   at slurmd/slurmstepd/slurmstepd.c:516
7  0x0000000000424302 in main (argc=1, argv=0x7fff1f7c6c98)
    at src/slurmd/slurmstepd/slurmstepd.c:127

errno before each call to the API.

Correct Shared field in job state information seen by scontrol, sview, etc.

Only for pending jobs
bug 899

allocation and CR_PACK_NODES is set layout tasks appropriately.

This is related to bug 890

the sinfo command is parallelized for performance reasons and it
really can not be completely parallelized for some use cases.
see bug 883

Correct logic to support Power7 processor with 1 or 2 threads per core
(CPU IDs are not consecutive).
bug 891

There was some logic copied from bitstring.c and an unused variable
problem reported by the compiler.

This is due to a bug introduced in commit 83d626caa
SOme configurations could result in NULL names in the node list table
(e.g. hidden partitions).

SLowness introduced in commit 83d626ca

This reverts commit 0d6a9965
That patch would permit a job with shared resources to run on
the same node as a job without shared resources, unfortunately
it let those jobs share CPUs. Finer grained sharing might be
possible with extensive code changes, but not something to
work on now.

Without this change, the job's --shared option when used with a partition
configuration of Shared=YES was not being honored by the select/cons_res
or select/serial plugin.

Original code was implicitly setting a job's shared field to 1
for select/cons_res.

This reverts commit c773b750

display larger sharing values.

cores specialization test 17.34 failed with bad assignment logic

If FastSchedule=0 is configured and some nodes have not registered
for service (so we do not know their actual resource counts), then
leave the job pending rather than rejecting it without knowing if
it can run later (when the node registers and we know its specs).
bug 872

Rather than the numeric value