mpiwg-sessions / sessions-issues

3 stars 1 forks source link

MPI_INFO_ENV and sessions - may need clarification/testing in implementation #9

Closed hppritcha closed 2 years ago

hppritcha commented 5 years ago

We probably need to add verbiage about use of MPI_INFO_ENV pre-defined info object in the peer to peer model. I'm in favor of saying this info object is available for use by the application after the first invocation of MPI_Session_init

dholmes-epcc-ed-ac-uk commented 5 years ago

The current text states (MPI-3.1, page 356, line 20): "After MPI is initialized"

If MPI_INFO_ENV is available after the first session init then it would also be available before MPI_Init, which is not currently (expressly) permitted.

MPI_Session_init();
/* query MPI_INFO_ENV */
MPI_Init();

Most of the key/value pairs seem innocuous except: if the value MPI uses for the argv key/value comes from the parameter passed in to MPI_Init, then the user may get a different answer from queries before and after MPI_Init.

hppritcha commented 5 years ago

Need to see if its possible to do this with the proto-type.

dholmes-epcc-ed-ac-uk commented 5 years ago

Item from a forum feedback issue:

Discussion of MPI_INFO_ENV - should this be usable if no active session - for now only allow in world model. Dan asks: can MPI_INFO_ENV be populated without access to argv? See issue:mpi-forum/mpi-issues#115 (PR mpi-forum/mpi-standard#110), which also punts on this change for the same reason.

hppritcha commented 2 years ago

closing this issue as the MPI_Info_create_env route was taken as a means for providing support for an info object similar to MPI_INFO_ENV when only using the sessions process model.