Researched and drafted by Claude (Anthropic's AI model, in Claude Code), which also ran the measurements below; reviewed by @bjlittle before posting.

Type checking geovista against types-rasterio 1.5.2.20261005 turned up three places where the rasterio stubs are narrower than rasterio 1.5.2 at runtime.

1. rasterio.transform.xy types rows and cols as int | Sequence[int], and its result as tuple[float, float] | tuple[list[float], list[float]]. rasterio takes numpy arrays, and gives arrays back for them:

   import numpy as np
   from rasterio.transform import from_origin, xy
   
   transform = from_origin(10.0, 20.0, 1.0, 1.0)
   rows, cols = np.meshgrid(np.arange(2), np.arange(3), indexing="ij")
   xs, ys = xy(transform, rows.flatten(), cols.flatten())
   type(xs)  # <class 'numpy.ndarray'>

   An ndarray is not a Sequence, so a caller passing one needs a cast.

2. DatasetBase.crs is typed CRS, but a GeoTIFF without a CRS gives None.

3. DatasetReaderBase.read is typed to return NDArray[Any] whatever masked is, but with masked=True it returns a numpy.ma.MaskedArray, so using its mask doesn't type check.

If any of these would be welcome as a pull request, Claude can prepare one the same way, for @bjlittle to review before it is opened.

🤖 Generated with Claude Code