Dealing with optional cartopy dependencies

[ngoldbaum]

Right now it's a little bit awkward from a packaging perspective to depend on cartopy in an external package and also to use optional functionality in cartopy. For example, let's say I want my package to directly depend on cartopy and I also want to use cartopy's integration with matplotlib or the image transforms, which depend on scipy.

Now I can't tell my users that they should just install cartopy, since that won't also install all the direct dependencies of cartopy. I also can't list just cartopy in my pyproject.toml or install_requires.

One way to fix this would be to define a set of "extras" for optional functionality, then I could tell my users to install e.g. cartopy[plotting] from pip.

Another way to fix this would be to add e.g. matplotlib, scipy, and other python libraries with a relatively "solved" packaging story on most major platforms as direct dependencies of cartopy.

[QuLogic]

Extras are defined here; do they not work?

[dopplershift]

To be even more clear, pip install cartopy[plotting] should work today already. Have you tried it?

[ngoldbaum]

Ah cool I didn't appreciate this already works. I guess it just needs to be documented? For example, in the installation docs:

https://scitools.org.uk/cartopy/docs/latest/installing.html#installing

[pelson]

Agreed. We can improve the docs here.

As for conda packages, it is pretty much the same argument as conda-forge/xarray-feedstock#5 - should conda install cartopy install the full package, or the minimum viable product. Conda variants are perhaps the closest thing we have to being able to deliver that, but I'm not 100% on the status of using them (there is quite some irony here).

[greglucas]

Some documentation on installing the extras is now included in the release.