Apparently conda can drop still supported upstream versions of Python, see https://github.com/datalad/datalad-installer/pull/156/files and links there in. But that relates to newer versions of the installer, and previous ones seems to be just fine. So I think it would be valuable to add ability to install versioned miniconda environments. Then in CIs we can specify those guaranteed to exist versions until we, not forced by conda, decide to drop support for a specific Python release.
Apparently conda can drop still supported upstream versions of Python, see https://github.com/datalad/datalad-installer/pull/156/files and links there in. But that relates to newer versions of the installer, and previous ones seems to be just fine. So I think it would be valuable to add ability to install versioned miniconda environments. Then in CIs we can specify those guaranteed to exist versions until we, not forced by conda, decide to drop support for a specific Python release.
E.g. I have tried on https://repo.anaconda.com/miniconda/Miniconda3-py37_23.1.0-1-Linux-x86_64.sh and it installed miniconda with python3.7 just fine. Looking at https://repo.anaconda.com/miniconda/ , I think we should not try to be "smart" and just expect full version specification such as
py37_23.1.0-1and simply default tolatestfor the version. Then an alternative to https://github.com/datalad/datalad-installer/pull/156/files would be not to skip, but to use"miniconda" + '=py37_23.1.0-1' if sys.version_info[:2] == (3, 7) else ''or alike.