shapely as a dependency (descartes)

[espg]

I'm working on implementing geojson data structure storage for the ui, and as I'm reading up on the specification, it does look and seem like shapely is very standard in the python ecosystem for dealing with this type of data. We finally got the docker builds working (thanks @fperez !), in part by removing geopandas and shapely, which were causing build problems since descartes wasn't able to build. I think that geopandas is a bit of a dependency mess, and I'm happy to leave it out as a requirement, but shapely does have some obvious use, and a tight build structure. This is shapely's environment list:

dependencies:

• python==3.6.*
• geos==3.8.*
• matplotlib
• descartes

...all of which are reasonable--except for descartes, which besides breaking our build is marked as broken in several build systems, and is pinned to a static legacy version in their requirements-dev.txt (descartes==1.0.1). Initially, I thought this was referencing this library from descartes-labs, but it appears to be an older matplotlib adjacent library that implements mostly just PolygonPatch ... which as far as I can tell is used as an external import when using shapely.

@whyjz do you know if I'm missing something and this is used elsewhere in shapely? I'm thinking it may be worth a pull request to shapely to remove the dependency from shapely after the EarthCube poster is wrapped up, so we can pull shapely in at some point....

[whyjz]

I am going to give a stab at it now, but I actually talked about the descartes dependency issue with the icepyx team (see icesat2py/icepyx#171). So the geopandas team is aware of this issue, and since geopandas is built on shapely, I guess the shapely team is tracking this too. We can still try to open an issue or PR to shapely to see what they actually think!

[espg]

@whyjz Honestly, the descartes code is really, really tiny-- 55 lines in one file with a single line __init__, some tests, and a setup.py. Looking at earlier it occurred to me that the correct resolution might be a PR to merge it into matplotlib. All the other shapely and geopandas have it as requirement because they have examples and documentation that uses descartes for plots... but if it was in matplotlib, they could remove the dependency and not change any code (just import calls for examples and documentation)