Hi all, BACKGROUND:
Over the past year or so we have been nudging researchers to conda environments where we determine their performance needs are not mission critical and their software needs are best handled by the conda-forge/bioconda ecosystem. We have been doing so by providing a bare-bones miniconda module from which they can bootstrap their environments. We like the side-effects of having a read-only base environment - we find that forcing people to think about and create different environments for different purposes cuts down on deeply broken root environments where conda install commands were run with reckless abandon. However, the conda authors have gotten more aggressive with messaging for the deprecation of the source activate method of activating environments, prompting folks to run conda init. This ends up tying them to a specific path and version of miniconda regardless of updates / changes we make, and can also result in strange behavior if a job environment inherits portions of an activated conda environment. QUESTION: For other sites that allow/encourage this behavior, how have you dealt with this issue? We have mused about trying to write a lightweight conda function that mimics the real one and that plays better with Lmod, but worry about then having to chase the actual distribution's functionality. Thanks, Ben Evans, PhD Senior Computational Research Support Analyst Y|CRC Site <https://research.computing.yale.edu/> / Docs <https://docs.ycrc.yale.edu/>
