Compiler in base-spack

[CodeGat]

Pushing the compiler down into base-spack as it is not updated as frequently in the pipeline and can therefore help compilation times. This PR changes the workflows somewhat, in which the matrix is created relatively early in the pipeline, and is fed into the new workflow (dependency-image-pipeline.yml)

Should close #50 !

[aidanheerdegen]

I have some comments!

I can't say I have checked it exhaustively, but I can do another pass if required.

[Quartzer2]

All good. Though just a cursory run through before I have to head off.

Only point in relation to something Aidan mentioned (and this my opinion only) is to use traditional GNU stuff (sed/awk/etc). Jq is great and I've used it often - just that it is 3rd party. Again just my opinion, I'm an old timer (pre-JSON standard) so probably a bit conservative in what I use for parsing :) Otherwise happy to use new stuff - just depends if it is critical or not. Depending on definition of citical!

[CodeGat]

@Quartzer2 thats a fair opinion too! Looking through the build-ci code again now and there seems to be precedent for jq but I'll wait to change anything for the moment. Interestingly, the runner comes with jq installed(?!)

[aidanheerdegen]

@Quartzer2 thats a fair opinion too!

Indeed it is. But as you say @CodeGat we're already using it and it is very nice and fast and actually parses the JSON, so has more chance of picking up problems, and working with a wider range of possible inputs.

I like jq because it is obvious what it is doing. We could arguably make it even more obvious using the full command line option --compact-output.

I admit I couldn't think of a way the current method breaks, but I only spent 5 minutes thinking, and I may not be very imaginative. I did think of a way: DOS line-endings, but that seems a bit far-fetched, so I'm not counting it.

[aidanheerdegen]

Looks great!

Thanks for the good work.