Add netCDF4 as a dependency and install netcdf4

[seisman]

Description of proposed changes

xarray needs the netCDF4 package for reading (perhaps also writing) netCDF files. However, netCDF4 isn't installed by default since it's an optional dependency of xarray. Although pygmt doesn't import netCDF4 package explicitely, some features won't work without the package.

See #239 for past discussions.

Reminders

• Run make format and make check to make sure the code follows the style guide.
• Add tests for new features or tests that would have caught the bug that you're fixing.
• Add new public functions/methods/classes to doc/api/index.rst.
• Write detailed docstrings for all functions/methods.
• If adding new functionality, add an example to docstrings or tutorials.

[weiji14]

Thanks @seisman, was surprised that netcdf4 wasn't listed! Just one minor comment on whether we should streamline the conda install instructions once #430 is merged in.

I do wonder whether we should list netcdf as a strict dependency, or have it as a (strongly) recommended one. For sure that conda users should install it and get it. But I wonder if there are advanced users that might not want to install it because of GPL issues (see intake/intake-xarray#41), though that would need to fixed on our setup.py.

[weiji14]

Once #430 is merged, would it be better to just instruct users to conda install pygmt since it'll pull in all the side dependencies as well?

[seisman]

Yes, perhaps we should merge conda-forge/pygmt-feedstock#2 first, then instruct users to install pygmt with:

conda create --name pygmt pygmt

[weiji14]

But I wonder if there are advanced users that might not want to install it because of GPL issues (see intake/intake-xarray#41), though that would need to fixed on our setup.py.

Then again, GMT is LGPL licensed, but PyGMT is BSD3 licensed, so I'm not sure if it matters. I'm not a lawyer though.

[weiji14]

Yeah. It falls back to a scipy netcdf3 implementation but the GMT files don't conform to that style so no luck.

_Originally posted by @leouieda in #239 (comment)

Just a note that xarray.open_dataset supports other 'engines' besides netcdf4 and scipy. See http://xarray.pydata.org/en/v0.15.1/generated/xarray.open_dataset.html?highlight=engine. For example, I've often used the h5netcdf engine (BSD3 licensed), and have been playing with the Zarr engine too (MIT licensed).

In the future, we might want to look into supporting other PyData 'array' type inputs (via some setting, e.g. as a kwargs) as an alternative to non-netcdf4 users, just not now in this PR 😄

[seisman]

@weiji14 I just moved netcdf4 to an optional dependency, and remove it from the conda install command.

[weiji14]

If we're going to put this as optional, then maybe remove it from setup.py too?

[seisman]

If we're going to put this as optional, then maybe remove it from setup.py too?

I personally prefer to have all dependencies installed.

AFAIK, pygmt.datasets.load_earth_relief() needs netCDF4 for working. Since this function is used in the tests and the tutorial, it's better to have it installed, especially for new Python users.

[weiji14]

Yeah, better to have it as a non-optional dependency then. We could have a pip install pygmt[minimal] setup in the future if anyone really doesn't want netcdf.

[seisman]

Yeah, better to have it as a non-optional dependency then. We could have a pip install pygmt[minimal] setup in the future if anyone really doesn't want netcdf.

Done.

[seisman]

Merged into PR #430. Closing.