Drawing of certain paths breaks in `cartopy` 0.25.0

[andrewdnolan]

Hello! I’m a developer of mosaic, a python package for unstructured mesh visualization which is heavily reliant on cartopy.

We’ve noticed an issue drawing certain paths when switching from cartopy 0.24.0 to 0.25.0. See E3SM-Project/mosaic/#42 for the original issue where this was brought to our attention. Here’s a minimum working example:

import cartopy
import cartopy.crs as ccrs
import numpy as np
import matplotlib.pyplot as plt

from matplotlib.collections import PolyCollection

# https://wktmap.com/?fae129b6
patch = np.array([[
    [-180.00020734,  -63.5383884 ],
    [-179.93611911,  -63.61745971],
    [-179.77001049,  -63.62512738],
    [-179.67109735,  -63.55616224],
    [-179.74190981,  -63.49385818],
    [-179.91362571,  -63.48807984],
    [-180.00020734,  -63.5383884 ],
    [-180.00020734,  -63.5383884 ],
]])

fig, ax = plt.subplots(
    constrained_layout=True, subplot_kw={"projection": ccrs.PlateCarree()}
)

collection = PolyCollection(patch, array=[1.], ec='k', alpha=0.5)

collection.set_transform(ccrs.PlateCarree())

ax.add_collection(collection)
ax.autoscale_view()

ax.set_title(f"cartopy v{cartopy.__version__}")
ax.gridlines(draw_labels=True)

plt.show() 

Here's what the patch looks like in cartopy 0.24.0:

conda create -n cartopy_0.24.0 python=3.13 cartopy=0.24.0

And, here's what the patch looks like in cartopy 0.25.0:

conda create -n cartopy_0.25.0 python=3.13 cartopy=0.25.0

---

My hunch is that the issue arises because the first/last index lies outside of the projection boundary (i.e. -180 deg). But, I haven't had time to through and test that hunch.

For some context, in mosaic we repeat the first index as the last to ensure the polygons are closed. Because we have variable resolution meshes, the polygons can have variable number of sides. To support ragged array likes this, we further repeat the first/last index as many times as needed to fill in up to the max number of sides.

cc'ing @xylar @cbegeman

[rcomer]

Thanks for the report @andrewdnolan. This bisects to #2470, but I have no intuition for how that change caused this 😕

[rcomer]

Confirmed there is no problem if you enforce all points are >= -180. longitude.

[greglucas]

🤔 it looks like the interior_polys is an empty list with this specific geometry and then the empty multi_polygon somehow leads to this issue.

If we put the following lines into a block, only if there are interior_polys that fixes the issue for me locally. if interior_polys:

cartopy/lib/cartopy/crs.py

Lines 1257 to 1259 in fa64deb

	# Invert the polygons
	multi_poly = shapely.make_valid(sgeom.MultiPolygon(interior_polys))
	polygon = box.difference(multi_poly)

I haven't investigated why and if this is the right fix, but figured I would post here for others debugging.

[rcomer]

My favourite part of working with people in different time zones is that problems get solved while I sleep 😄

@greglucas this makes perfect sense to me

• Before FIX: create a single inverted polygon when no exteriors found #2470, we took each non-empty interior and made a new polygon as the difference between the whole map and that interior. So if all interiors were empty we just didn’t create anything.
• Since FIX: create a single inverted polygon when no exteriors found #2470, we gather all the non-empty interiors into the list and make the difference between the whole map and the combination of the list. If all interiors are empty then the list is empty and we should skip creating the difference polygon.

[rcomer]

For the current case, we start with one interior ring, which is invalid. Previous to #2470, we did shapely.geometry.Polygon(ring).buffer(0) to try to make it valid, and this made it empty. Since #2470, we use shapely.make_valid, which gives us a LineString, which we also cannot treat as an interior/hole, so we ignore.

[andrewdnolan]

Thanks for addressing this so quickly!

Did a developer install of cartopy and can confirm that the original issues is resolved after #2594.

Thanks!

[xylar]

@rcomer and @greglucas, thanks so much for your help with this.

[rcomer]

Apologies for breaking it to begin with!