So we've been handling dependencies per rule in snakemake, making a new environment each time. This has mostly worked for us, but maybe it's worth reconsidering due to the following things:
requires an internet connection to work (e.g. I cannot run pling as is on the Pasteur cluster, as compute nodes don't connect to the internet)
we weren't able to make pling work as a pip package as a result of weird complicated nesting of virtual environments that came from the unholy union of snakemake, conda and pip (see #47 and #48)
since we use DingII now, which is written in python3, we actually shouldn't have any dependency conflicts between rules
additional dependency on both mamba and conda
I've given a go aggregating all the necessary dependencies in one environment, and running pling from there without the --use-conda flag. It's worked without any issue! So basically should be feasible, and could open up the possibility of bringing back pip installation (important since we've had complaints about installation looking off-putting).
My suggestion would be to handle dependencies for integerisation from alignment directly in pling, and maybe leave the dependencies for annotation as optional, and to be installed manually by the user if they want to run it (dissuading anybody from using annotation appraoch unless they really, really want to). As an initial thing we could remove the snakemake handling of conda environments and dependencies, and add a configuration file for dependencies that can be installed via conda, eventually hopefully upgrading to something like installation via pip and nothing else.
So we've been handling dependencies per rule in snakemake, making a new environment each time. This has mostly worked for us, but maybe it's worth reconsidering due to the following things:
I've given a go aggregating all the necessary dependencies in one environment, and running pling from there without the --use-conda flag. It's worked without any issue! So basically should be feasible, and could open up the possibility of bringing back pip installation (important since we've had complaints about installation looking off-putting).
My suggestion would be to handle dependencies for integerisation from alignment directly in pling, and maybe leave the dependencies for annotation as optional, and to be installed manually by the user if they want to run it (dissuading anybody from using annotation appraoch unless they really, really want to). As an initial thing we could remove the snakemake handling of conda environments and dependencies, and add a configuration file for dependencies that can be installed via conda, eventually hopefully upgrading to something like installation via pip and nothing else.