naturally, removing the sort affects the tests, I'll take care of it

without the sort, the tests run differently in the CI and locally, I suppose because the order of glob is not well defined. So it would seem we do need to sort somewhere, but making sure higher versions come first when using --review-pr-max (?)

I suggest just making the globbed values stable:

potential_paths = sorted(sum(potential_paths, []), reverse=True) # flatten

by sorting them what most likely is the best -> word order (unless corner cases where a software goes from 2.9 to 2.10, but, the version matching logic mostly takes care of that.

I think this leaves find_related_easyconfigs in a pretty good state, and it can be improved in the future if an even better matching procedure can be constructed independently of this code for limiting the output.
I really don't think we need, or should, do anything special whether these new flags are in use.

(I was planning on also using this method for the boegelbot check, but with some modifications;

1. Just the first match
2. Just show a normal 2 file diff without the other stuff
3. Not the other text output
4. Probably add a link to the best matching existing easyconfig on github)

@boegel I tested test.framework.suite github locally with python 2 and 3, but it would probably be best if you do too :)

The many tests configurations done by the CI should suffice here no? I think it's good to be merged for 4.5.0

@Micket for PRs, the CI skips the tests that require a github token such as test_github_review_pr. I did run all the github tests locally from this branch, and you can too if you install a token for the easybuild_test user

I'd still want to hear from @boegel though. This seems both useful and harmless to me, but maybe we're missing something.