Some people have commented on the "heavy" dependencies for SpacePy; in some constrained environments scipy can be relatively big. We only really hard-require numpy and document pretty heavily what functionality is lost without a given dependency. However, pip install gets almost all the dependencies, by popular demand. (We don't install AstroPy, since we only use it to put data in/out of AstroPy...if you need it, you've got it.)

People can use --no-deps and manage manually, of course, but this is hard to do in a requirements.txt. So it would be nice if we could pip install spacepy[minimal] or something to only install numpy. It appears this is not possible--pip "extras" are truly extra.

We could do something like default to numpy only and then use pip install spacepy[full], but I think the default should be to get a reasonable set of dependencies.

I was thinking of starting to treat scipy, matplotlib, h5py as full-required but based on what I'm hearing, some people do like having the lighter dependency set. We just need an easy way of getting it to them.