pmi2: check failures and warn. Singloton check must be elsewhere.

Depending on slurm versions, PMI2 can fail differently when there is
no server. But it can fail to contact its server for different reasons:
- no slurm flag `--mpi=pmi2` given. This needs a warning
- singleton : no server when no srun. There should be no warning here

Here both are handle with a warning, so disqualifying singleton must be
made before.

Author: Piotr Lesnicki

opal/mca/common/pmi/common_pmi.c

@@ -46,7 +46,6 @@ bool mca_common_pmi_init (void) {
         rank = -1;
         appnum = -1;
 
-
         /* if we can't startup PMI, we can't be used */
         if (PMI2_Initialized ()) {
             return true;
@@ -57,17 +56,13 @@ bool mca_common_pmi_init (void) {
             mca_common_pmi_init_count--;
             return false;
         }
-        if (size < 0 || rank < 0 ) {
+        /* depending on slurm versions, we may get bad rank/size or bad jobid */
+        if (size < 0 || rank < 0 || PMI2_SUCCESS != PMI2_Job_GetId(buf, PMI2_MAX_VALLEN)) {
             opal_show_help("help-common-pmi.txt", "pmi2-init-returned-bad-values", true);
             mca_common_pmi_init_count--;
             return false;
         }
 
-        if (PMI2_SUCCESS != PMI2_Job_GetId(buf, PMI2_MAX_VALLEN)) {
-            /* PMI2 can't be used if no job in singloton mode */
-            mca_common_pmi_init_count--;
-            return false;
-        }
         mca_common_pmi_init_size = size;
         mca_common_pmi_init_rank = rank;
         mca_common_pmi_init_count--;

opal/mca/common/pmi/help-common-pmi.txt

@@ -13,7 +13,7 @@ We cannot use PMI2 at this time, and your job will
 likely abort.
 #
 [pmi2-init-returned-bad-values]
-PMI2 initialized but returned bad values for size and rank.
+PMI2 initialized but returned bad values for size/rank/jobid.
 This is symptomatic of either a failure to use the
 "--mpi=pmi2" flag in SLURM, or a borked PMI2 installation.
 If running under SLURM, try adding "-mpi=pmi2" to your