Remove the dependency of netCDF4

[seisman]

Description of proposed changes

Inspired by #3639 (comment), now it's possible to remove the netCDF4 package from PyGMT's dependency list.

TODO:

• **BREAKING** pygmt.grdcut: Refactor to store output in virtualfiles for grids #3115
• Implement gmt xarray BackendEntrypoint #3919
• Deprecate pygmt.io.load_dataarray, use xarray.load_dataarray instead #3922

[weiji14]

You know what? Maybe we could think of refactoring pygmt.io.load_dataarray to use this with block? It would be a breaking change, but that could mean we can use GMT's I/O directly.

[seisman]

Yes, actually, I'm thinking if we should provide a wrapper for the special read module, which can read a table, grid, or image, and return pd.DataFrame or xarray.DataArray. The function looks like:

def read(file, kind=None, **kwargs):
if kind == None:
# Try our best to decide the data kind. Or make 'kind' a required parameter
pass code = {"dataset": "d", "grid": "g", "image": "i"}[kind]
func = with Session() as lib:
with lib.virtualfile_out(kind=kind) as voutfile:
lib.call_module(
module="read", args=[file, voutfile, f"-T{code}"]
)
if kind == "dataset":
return lib.virtualfile_to_dataset(vfname=voutfile, **kwargs)
elif kind in {"grid", "image"}
return lib.virtualfile_to_raster(vfname=voutgrd, kind=kind)

For comparison, GMT.jl provides the gmtread/gmtwrite functions, but As I understand it, the functions return the GMT_DATASET/GMT_GRID/GMT_IMAGE containers, rather than data structures in the Julia world.

[weiji14]

Not sure if removing a required dependency counts as a breaking change (i.e. if we should should we put **Breaking** in the title), the main non-test-related user-facing change was to _load_earth_relief_holes in pygmt/dataset/samples.py in #3922 which is quite minor. I suppose we could just label this under 'deprecation' instead of 'maintenance' to make it a bit more prominent in the changelog?

[weiji14]

There's a test failing on ubuntu latest - Python 3.12. Need to add a @pytest.mark.skipif(condition=not _HAS_NETCDF4, reason="netCDF4 is not installed") marker for this line:

pygmt/pygmt/tests/test_xarray_backend.py

Line 27 in ae82fd6

grid.to_netcdf(tmpfile.name)

because .to_netcdf() still relies on the netCDF4 engine.

[weiji14]

Nice, glad to have one less dependency to handle! Remember to remove netCDF4 from the conda-forge feedstock for the next release 🔥