Making a conda-forge DGL package

[knc6]

🚀 Feature

Hi,

I would like to make a conda-forge package for ALIGNN network which uses DGL. As DGL is only available on its own conda channel and not on conda-forge, I submitted a PR for making a conda-forge DGL implementation here.

I would like to make sure if DGL team is ok with this strategy, or please suggest an alternative.

Currently, I am getting an error No such file or directory

Could someone from DGL team please take a look? My conda-recipe is here

Thanks.

[BarclayII]

Hi,

Thank you for your contribution! I'm not quite familiar with conda-forge so I was wondering what will be the difference between using a conda-forge repo and a third-party (i.e. dglteam's) repo?

[github-actions]

This issue has been automatically marked as stale due to lack of activity. It will be closed if no further activity occurs. Thank you

[hadim]

I already made an attempt making a conda forge package for dgl. You can find more details at #1855 and conda-forge/staged-recipes#12552

[hadim]

Thank you for your contribution! I'm not quite familiar with conda-forge so I was wondering what will be the difference between using a conda-forge repo and a third-party (i.e. dglteam's) repo?

In prod, it's a good practice to reduce the number of conda channels to depend on to reduce dep graph solving complexity. Also depending on multiple channels increase the surface of potential bugs in the packages since not all conda channels are maintained the same.

Conda forge is a community effort that guarantee package building will follow certain guidelines and good practices, so builds are reproducible.

Also currently, dgl provides different package names for different CUDA version (or CPU) which makes the maintenance of conda env file quite challenging (ideally a single env file should be valid and the same for multiple machine: CPU-only, CUDA-10, CUDA-11, etc). Conda has features to be able to automatically select a particular build (cpu, cuda, etc) according to the context.

[mikemhenry]

1.1.2 is now on conda-forge, so this issue can be closed.

[Rhett-Ying]

1.1.3 is now on conda-forge.