fix a bug introduced by PR 11854 #12242
Conversation
opal/mca/pmix/base/pmix_base_fns.c
Outdated
| mca_base_parse_paramfile(file, ¶ms); | ||
| free(file); | ||
| OPAL_LIST_FOREACH (fv, ¶ms, mca_base_var_file_value_t) { | ||
| pmix_overlap = (NULL == evar) ? 0 : check_pmix_overlap(fv->mbvfv_var, evar); |
There was a problem hiding this comment.
Please see my comment on your mpi4py PR - this is an incorrect fix. The evar variable is not set - you are using a dangling value from the prior code block.
There was a problem hiding this comment.
hu? the triad has check for evar being null
There was a problem hiding this comment.
okay i see what you're saying. we just shouldn't use evar in this section.
opal/mca/pmix/base/pmix_base_fns.c
Outdated
| mca_base_parse_paramfile(file, ¶ms); | ||
| free(file); | ||
| OPAL_LIST_FOREACH (fv, ¶ms, mca_base_var_file_value_t) { | ||
| if (check_pmix_param(fv->mbvfv_var)) { |
There was a problem hiding this comment.
This still isn't quite correct. You do need to check the params from the file for overlap - i.e., legacy ORTE params that need to be converted to PMIx. The problem was that we were passing the wrong things to check_pmix_param - the params in that function are (the variable name, the variable's value). So here you would pass check_pmix_overlap(fv->mbvfv_var, fv->mbvfv_value).
You still need check_pmix_param in case they put a PMIx var in the file - you just need to check overlap as well.
HTH
|
there is a typo in the commit message (PR ... introduced a big) |
hppritcha
force-pushed
the
fix_for_pr_11854
branch
from
c48044e to
f2ecd67
Compare
|
i think the nvidia failure is related to #12245 |
|
@qkoziol please review when you have a chance |
@hppritcha That's correct. The fix is here #12246 I will ping Quincey to take a look at the PR. |
wenduwan
changed the title
wenduwan
changed the title
|
Hmmm not sure how to trigger nvidia CI. The failure should have been fixed. |
PR open-mpi#11854 introduced a big that causes singleton runs to segfault at startup in some cases this bug was rooted out by the github action in PR open-mpi#12217 Signed-off-by: Howard Pritchard <[email protected]>
hppritcha
force-pushed
the
fix_for_pr_11854
branch
from
f2ecd67 to
86a05c1
Compare
PR #11854 introduced a big that causes singleton runs to segfault at startup in some cases
this bug was rooted out by the github action in
PR #12217