Issue420/get nb key@v5

[abouteiller]

Import changes from v4.2 to v5.1

#491
closes #420

• waiting on Change type of key parameter to PMIx_Get_nb to match PMIx_get #488 merged in v4
• review openPMIx for conformity with proposed change: NOT CONFORMANT
• review v4.2 and potential need to rollback Change type of key parameter to PMIx_Get_nb to match PMIx_get #488 ? NEED TO ROLLBACK

[raffenet]

OK, after I clicked approve I went and double checked the OpenPMIx and see that neither API there uses pmix_key_t. I forget if this was discussed when we proposed the change?

PMIX_EXPORT pmix_status_t PMIx_Get(const pmix_proc_t *proc, const char key[],
                                   const pmix_info_t info[], size_t ninfo,
                                   pmix_value_t **val);
PMIX_EXPORT pmix_status_t PMIx_Get_nb(const pmix_proc_t *proc, const char key[],
                                      const pmix_info_t info[], size_t ninfo,
                                      pmix_value_cbfunc_t cbfunc, void *cbdata);

[rhc54]

Yeah, you can't make it a pmix_key_t as the key parameter can be NULL - and IIRC, the compiler will complain if it is declared as a pmix_key_t in that instance.

[abouteiller]

We need to double check what happened with #488 as this is when we voted on it. The way that was understood (see #420) the implementation was already using the pmix_key_t and one of the procedures took it as such.

[rhc54]

implementation was already using the pmix_key_t and one of the procedures took it as such.

Correct - and as I stated in #420, we corrected it to both APIs using const char key[]. Otherwise, you cannot pass a NULL argument, which would break some of the rules for PMIx_Get

[abouteiller]

I did exactly the wrong fix... Good that we caught that before it went out the door. I'll update #420 and we will have to revote on it, I'll check if that'll delay v4.2 based on our rules but I think it will.

[abouteiller]

This will now need to mirror #510