Add conversion between cf and shapely for lines

[jsignell]

I started on this at the SciPy sprints with the idea of pushing forward the work described in this ticket: xarray-contrib/xvec#48. It took me a while to get the logic right 😅 so I wanted to make sure it was somewhere public in case I don't get a chance to come back to it in the near future.

[codecov]

Codecov Report

Attention: 8 lines in your changes are missing coverage. Please review.

Comparison is base (2648109) 83.10% compared to head (bea680c) 83.26%.

Files	Patch %	Lines
cf_xarray/geometry.py	89.74%	7 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #460      +/-   ##
==========================================
+ Coverage   83.10%   83.26%   +0.15%     
==========================================
  Files          14       14              
  Lines        2676     2749      +73     
  Branches      189      199      +10     
==========================================
+ Hits         2224     2289      +65     
- Misses        402      409       +7     
- Partials       50       51       +1     

Flag	Coverage Δ
mypy	42.37% <26.92%> (-0.50%)	⬇️
unittests	94.11% <100.00%> (+0.20%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[dcherian]

Thanks! This is great to see

cc @aulemahal

[jsignell]

Just to make sure this is visible from this side as well: while working on this PR I was pointed to https://github.com/twhiteaker/CFGeom. I have opened an issue on that repo to see if the contributors over there have opinions about how these efforts relate to each other.

[jsignell]

This part caught me up for a bit. As I understand it, the node_count is the number of nodes per line segment. So each line segment in a multiline results in a value in this array. This means that the dims is not index and is not in the shapely version at all. I gave it a kind of random name counter which should probably be changed.

[dcherian]

@jsignell Any interest in picking this back up?

[martinfleis]

This could be significantly simplified by using shapely.to_ragged_arrays and from_ragged_arrays. That is used to create a GeoArrow arrays of coordinates, which are nearly the same as CF representation. Just one uses offsets to indicate where a new geom starts and the others counts of coords per geom, which should be an easy conversion.

[jsignell]

The motivation for this work was to support IO in xvec, so I kind of stalled after @martinfleis came up with a better solution on that side. If there is still desire for this work then sure I can pick it back up.

[martinfleis]

@jsignell if you mean the WKB option, there is a bit of disagreement if that is a good idea. If we can get conversion to CF geometries via ragged array IO shapely has, we can use practically the same for GeoArrow option and Zarr (if we decide that is what we want). I still like the elegance of WKB but understand that some folks don't like saving binary objects into Zarr.

[jsignell]

Ah yeah I did mean the WKB option. In that case I can pick this back up. Or @dcherian feel free to take it over if you have some urgency around it.

[jsignell]

Thanks for the tip @martinfleis! I looked into from_ragged_array and I agree that it is pretty similar. Actually the logic that is in this PR for creating linestrings is nearly identical: https://github.com/shapely/shapely/blob/9540b77a939ba35d066d618183e67fc905ed9ec4/shapely/_ragged_array.py#L334-L337

I think the complexity in this PR comes from how cf intermingles MULTILINESTRING and LINESTRING. So once you have the line segments you still need to figure out which of them are going to be MULTILINESTRING objects. That is basically what the entirety this section is doing: https://github.com/xarray-contrib/cf-xarray/pull/460/files#diff-6484364cf7fa3216828551b2531094b30d77cf393313226a01d425c22580688aR353-R370

That is probably not the best possible way to do that conversion but I don't see anything similar in the shapely.from_ragged_array implementation

[jsignell]

Ok I added the shapely to CF direction for that one I found it pretty straightforward to use to_ragged_array so that one should read pretty cleanly! I'm going to try again now to see if I can simplify the other direction using from_ragged_array

[jsignell]

Ok this is clearly a case of Monday brain > Friday brain.

[jsignell]

I'm going to see what I can do to fix CI and improve coverage.

[dcherian]

I can't really review this so I'm happy to merge when you're ready.

[martinfleis]

I'll do the review, it is on my to-do list with an increasing priority :D.

[jsignell]

Yeah it's finally green and fairly well tested so it is actually ready for review for the first time 😅

[dcherian]

Thanks @jsignell . We can address @martinfleis comments as they come in :)

[jsignell]

YOLO!

[martinfleis]

Approved (even if too late)! Looks good @jsignell!

I suppose that Polygons will look similar but we'll need to deal with holes as one layer on top of this.

[jsignell]

I suppose that Polygons will look similar but we'll need to deal with holes as one layer on top of this.

Yeah I will probably start on that one now with all this fresh in my head.

Also I don't think I really considered z, so that might be something to look out for.

[dcherian]

Thanks for taking a look @martinfleis 🙏