Repair the processing of cmd line options that mapped to MCA params.

[rhc54]

This was responsible for breaking things like map-by foo

Fixes #1623

[rhc54]

Spent a bunch of time tracking down this "error" from the Mellanox Jenkins, and I don't think the test is correct. The cmd line shown in fact works correctly:

$  mpirun -np 2 -mca mca_base_env_list 'XXX_A=1;XXX_B=2;XXX_C;XXX_D;XXX_E' env | grep XXX
OMPI_MCA_mca_base_env_list=XXX_A=1;XXX_B=2;XXX_C;XXX_D;XXX_E
XXX_E=5
XXX_D=4
XXX_C=3
OMPI_MCA_mca_base_env_list=XXX_A=1;XXX_B=2;XXX_C;XXX_D;XXX_E
XXX_E=5
XXX_D=4
XXX_C=3
$

So when you run that thru wc -l, you will get an answer of 8, not 6.

[rhc54]

Should note: the Jenkins test only exports valus for C, D, and E. Perhaps it was expecting that A and B would be set? If so, then I'm not sure where and how that was supposed to be done, and will have to look further at some point.

[ggouaillardet]

@rhc54 I previously noticed Jenkins script explicitly export A and B before invoking mpirun

[rhc54]

Sadly, there is no documentation anywhere on these tests, nor are they available to me - nor is the functionality particularly documented in the code. So I'm working my way thru them as best I can.

Progress is being made...

[ggouaillardet]

I looks like there is a
set -x
in the Jenkins script, this is how I found it when reading the logs
I previously suggested all Jenkins scripts should be public (read only github repo looks fine) so anyone can understand what is tested and how. shall I add this to the agenda of the next telcon ?

[rhc54]

sure - it would definitely help

[jsquyres]

@jladd-mlnx @Di0gen can you please comment on this conversation?

[mike-dubman]

@ggouaillardet - the script is public and here (also mindist.c test)

https://github.com/mellanox-hpc/jenkins_scripts

[ggouaillardet]

thanks @miked-mellanox
@rhc54 here is what is broken

export XXX_A=1
echo XXX_B=2 > xxx.conf
mpirun -np 1 -am xxx.conf env | grep ^XXX

XXX_B is not exported

git bisect indicates 58dd41f is the guilty commit
according to the logs 2839484 might have been an attempt to fix it, but it did not fix this specific test case.

[rhc54]

Yes, I know - I pointed that out in an issue I filed, and requested help to decipher the AMCA stuff as it isn't very well commented 😄