With Python libraries we do not use multiple outputs.

Caffe is not a Python library. How about we stick to the buildEnv behaviour, that is,

extraOutputsToInstall ? []

What is needed long-term is a "higher-level" buildEnv that is aware of how to building Python envs, and Haskell envs, and Perls envs, and ...., all together in one env.

@FRidh Caffe is actually a Python library (it has Python bindings). The problem is that because they are installed in out and the default output is bin buildEnv gets confused now.

I just saw you added the bindings to python-packages.nix.

In any case, I still think we should not deviate from buildEnv by installing more outputs.

Yeah, I see why this is a good approach. Should we then add extraOutputsToInstall?

I still don't like that people will ask questions like "I made a Python environment with Theano and it doesn't work, what's the matter?" but I don't see any better way to do it (apart from making Theano not use multiple outputs but I don't like it...).

Yeah, I see why this is a good approach. Should we then add extraOutputsToInstall?

Yes!

I still don't like that people will ask questions like "I made a Python environment with Theano and it doesn't work, what's the matter?" but I don't see any better way to do it (apart from making Theano not use multiple outputs but I don't like it...).

Don't you mean another package? Theano does not have multiple outputs.

@FRidh Argh, sorry, I'm a little tired with messing with all ML libraries at once lately ^_^" I meant caffe.

Closing in favor of a commit that adds extraOutputsToInstall ? []