I spent some time trying to smooth out the python version checking / notifications in addition to merely fixing the blind invocation of python. It is worth nothing that it would simplify the script to give up on sanity checking a path python and simply always install miniconda.

Could one of (Tim Jenness, John Swinbank) review the GH PR?

I tested the changes on this ticket by hand but confess that newinstall.sh is, and long has been, in need of a testing framework. I've identified several bash testing frameworks but some experimentation is needed to make a selection.

Looks basically fine. I think there's an error in the argument handling in n8l::pyverok which is being hidden because it's falling back to defaults. Couple of other trivial comments on the PR.

After addressing all the GH PR comments, I concluded that basic unit testing was badly needed and did a quick review of bash testing frameworks. The only framework I was able to identify that supported "mocking" was rspec-bash. I've add rspec coverage for the new and modified functions + converted the "manual" unit tests for miniconda::install's handling of mktemp.