conda-forge recipes for biocPy

[UriNeri]

Hi @jkanche @LTLA ,
I'm preparing a software for publication and I used genomicranges and Iranges for interval operations.
However, trying to create bioconda recipe using grayskull suggests the BiocPy libraries are not available from conda (bioconda or conda-forge):

Could not solve for environment specs
The following packages are incompatible
├─ genomicranges >=0.6.3,<0.7 does not exist (perhaps a typo or a missing channel);
├─ iranges >=0.4.2,<0.5 does not exist (perhaps a typo or a missing channel);

This can be resolved by having biocutils, biocframe, genomicranges and iranges on conda-forge - which I am sure could help others too.
As these packages have little dependency overhead, and are all available from pypi, recipes for them could be generated using grayskull or pixi:

git clone [email protected]:conda-forge/staged-recipes.git
cd staged-recipes
git remote add upstream https://github.com/BiocPy/staged-recipes.git
git checkout -b add-biocpy

pixi run pypi genomicranges --recursive
pixi run lint

The linting step is suggested soem edits (adding home url for ncls, and adding {{ stdlib('c') }} for ncls and iranges), after adding these:

git add recipes/
git commit -m "Add recipe for genomicranges, iranges, biocutils, biocframe and ncls"
git push origin add-biocpy

I've ran the above (but with my github username instead of BiocPy) - https://github.com/UriNeri/staged-recipes/tree/add-biocpy - and I can a pull request to conda-forge staged-recipes if it is ok with you. I'd prefer to have BiocPy as the maintainer listed in the meta.yaml files, instead of my username as the default (or we could add both).
Let me know what you think.

[jkanche]

Hi @UriNeri,

This is awesome and thank you for making these packages more accessible on the conda infrastructure.

We don't commit as "BiocPy" & given its an org, listing that as a maintainer may not be that helpful unless thats the norm in conda (Sorry conda packaging is new to me). If you don't mind listing me as a maintainer, I am happy to resolve any issues in the future. Also helps me understand the process to publish rest of the packages to conda-forge.

[jkanche]

TIL: https://docs.github.com/en/pull-requests/committing-changes-to-your-project/creating-and-editing-commits/creating-a-commit-on-behalf-of-an-organization

[UriNeri]

Hi @jkanche - sounds good! I've added you as a maintainer - do you want to do the pull request?
We might have some troubleshooting do to if it fails the CI build tests (to be honest this is my first time uploading to conda-forge too).

[jkanche]

@UriNeri, seems like ncls is already part of bioconda, do you need that again on conda-forge?

[jkanche]

@UriNeri does it make sense to transfer over this repository to the BiocPy organization? May be thats what you meant here, but i misunderstood

I've ran the above (but with my github username instead of BiocPy)

[UriNeri]

I'm not sure about transferring it, I thought it was already part of biocpy but I think it's maintained/owned by the developers of pyranges/ncls. I think conda-forge doesn't accept bioconda as an additional channel for recipes. I think the best might be to contact them or submit to conda-forge on their behalf.

[jkanche]

@UriNeri i think right now the blocker is the ncls library. seems like one can provide bioconda deps. I'm happy to put back the configuration you added for ncls but that doesn't seem the right way to do this.

[UriNeri]

Are you sure conda-forge supports deps from bioconda?
If so than i agree with you, and dropping the ncls recipe is the way to go.
The commands suggested by conda-forge documentation for generations recipes for pypi packages include:
grayskull pypi --strict-conda-forge
But maybe I'm misinterpreting it.
I used the pixi task instead of grayskull directly as it has an option to recursively create recipes for ant missing dependencies.

I assume ncls a heavy dependency right? Not something that could be replaced/reimplemented?
Another thing I just noticed is joblib is an optional dep, I'm not sure how that is resolved.

An alternative is to instead aim to have the biocpy recipes in bioconda, which might be more fitting, though I think forge is much better...

[jkanche]

yup i've seen other recipes use bioconda.

[jkanche]

Closing this as the PR is ready for conda-forge to merge. This was helpful to do and at some point I would add the rest of the packages. Thanks again for bringing this up!