Distribution request: rocrate-py on conda forge

[Denubis]

(Requested duplicate to ResearchObject/ro-crate#217)

As an academic distributing software, I've just found myself making a conda environment (instead of a pip requirements.txt file). I'd like to be able to conda install -c conda-forge rocrate instead of:

dependencies:
  - python=3.10
  - pip
  - pip:
    - "rocrate"

A search https://anaconda.org/search?q=rocrate doesn't show rocrate as available inside the anaconda ecosystem.

[simleo]

Conda distributions tend to be more useful when the package has non-Python dependencies, e.g. C/C++, as is often the case for packages including extension modules. ro-crate-py, OTOH, is pure Python. That being said, Conda packages would still be nice to have, though the recipe requires work and then there's the burden of maintenance. I think the main hurdle would be the dependency from galaxy2cwl, which is not available from conda-forge.

Volunteers?

[jmfernandez]

Several months ago I made a contribution to bioconda, a conda channel more biased to life sciences. I contributed the Python crypt4gh conda recipe, inspired in the public indications from an Spanish colleague (sorry, instructions are in Spanish).

But, I guess the right conda channel should be bioconda instead of conda-forge. The main reason is that one of the rocrate-py dependencies is galaxy2cwl, which does not have conda package yet (as @simleo rightly pointed out). Exploring galaxy2cwl dependencies, it depends on gxformat2, which is already at bioconda channel.

[Denubis]

Bioconda instead of conda-forge is entirely acceptable for my needs, if that helps @jmfernandez.

[stain]

ro-crate is not a life-science library, so should really be on conda-forge.

We had Gromacs in bioconda (maintained by UNIMAN for a while) and then moved it to CondaForge as it was also used by Material Sciences.

Can the galaxy2cwl dependency be optional somehow in Conda?

[simleo]

I think we can make galaxy2cwl an optional dependency. Without it one would be able to do everything except generating abstract CWL from Galaxy workflows.

[stain]

Instead of galaxy2cwl could we not use gxformat2 directly? It can also convert to abstract CWL with gxwf-abstract-export command

https://github.com/galaxyproject/gxformat2

[corneliusroemer]

Just to add to the motivation: rocrate is a dependency of galaxy-data and so the the galaxy-data recipe in bioconda would like to require it at runtime - that's not possible while there isn't a conda-forge or bioconda package.

[ewels]

And to add one more voice: we'd like to use it in the @nf-core tooling and so a conda package would be great!

[mirpedrol]

Hello, since we would like to use rocrate in the nf-core tooling as @ewels mentioned, I tried adding the recipe and all the missing dependencies to conda-forge. I have some tests still failing.
Please let me know if that's ok and if you would like to be added as maintainers, thanks!

[mirpedrol]

I am tagging different implicated devs here to have the discussion centralized, please feel free to ping more people if needed.
@ewels @elichad @stain @pvanheus @nsoranzo @bedroesb

We have been trying to add the rocrate recipe to conda-forge and currently found some blockers:

• building the update of arcp v0.2.1 recipe
• using bioblend and gxformat2 packages, since they are already in bioconda but these can't be used in a conda-forge recipe

As @nsoranzo mentioned, there are two options:

• Make the galaxy2cwl dependency optional in rocrate, as suggested before
• Move bioblend and gxformat2 to conda-forge (dropping them from bioconda)

We would have to decide on the best way to deal with this. What is your opinion?
From nf-core we are happy to help with what we can to make this happen :)

[bedroesb]

I like the idea of adding the galaxy2cwl functionality into the ro-crate library directly, as @stain mentioned. This way we can indeed drop it as a dependency. Adding it to conda-forge is also fine by me, but in both cases I think we will have to retest the functionality with recent dependencies. The repo is untouched for a while now.

[simleo]

• Make the galaxy2cwl dependency optional in rocrate, as suggested before

Working on this, see #210. Without the ga2cwl option, dependencies are reduced to:

MarkupSafe
arcp
certifi
charset-normalizer
click
idna
jinja2
python-dateutil
requests
six
urllib3

[simleo]

I have just released rocrate 0.13.0, which does not require galaxy2cwl by default. bioblend and gxformat2 are thus not required anymore for the default installation.

[mirpedrol]

Thank you @simleo !

[elichad]

It's done! You can now conda install rocrate from conda-forge 🎉
Thanks @mirpedrol @ewels and colleagues for sorting out the recipe :)
https://anaconda.org/conda-forge/rocrate
https://github.com/conda-forge/rocrate-feedstock