Investigate what other environment variables are needed for postprocessing

[ilectra]

... in addition to the fields in the perflogs - things like compilers and their versions, MPI implementations, the benchmark app version, etc.

• Could any of them be fished out of the perflog fields?
• Should we just ask users to add whatever they need logged as an environment variable?

[giordano]

I think compiler version and MPI implementation should be fished out from the Spack environment (#110), and probably use Spack programmatic API to get specific information about compiler, MPI, etc....

[giordano]

For example

% spack -e . spec sombrero
Input spec
--------------------------------
sombrero

Concretized
--------------------------------
sombrero@2021-08-16%[email protected] build_system=makefile arch=linux-arch-haswell
    ^[email protected]%[email protected]~atomics~cuda~cxx~cxx_exceptions~gpfs~internal-hwloc~java~legacylaunchers~lustre~memchecker~orterunprefix+romio+rsh~singularity+static+vt+wrapper-rpath build_system=autotools fabrics=none schedulers=none arch=linux-arch-haswell
% spack -e . python
Spack version 0.20.0.dev0
Python 3.10.9, Linux x86_64
>>> from spack.spec import Spec
>>> spec = Spec("sombrero").concretized()
>>> spec.compiler
[email protected]
>>> spec["mpi"]
[email protected]%[email protected]~atomics~cuda~cxx~cxx_exceptions~gpfs~internal-hwloc~java~legacylaunchers~lustre~memchecker~orterunprefix+romio+rsh~singularity+static+vt+wrapper-rpath build_system=autotools fabrics=none schedulers=none arch=linux-arch-haswell

[ilectra]

So maybe add functionality to the high-level script (#104) to search for some standard things like those inside the spack speck file, if it cannot find them as fields in the perflog? The input to that script is looking more and more like it should be a config file instead of command-line options....

[ilectra]

This is partly being taken care of by #262