search

makepath / xarray-spatial

Raster-based Spatial Analytics for Python
https://xarray-spatial.readthedocs.io/
MIT License
844 stars 85 forks source link

Numpy 2.0.0 doesn't allow out of range integers when casting to np.uint64 #803

Open ehiggins0 opened 5 months ago

ehiggins0 commented 5 months ago

As of June 16, numpy 2.0.0 was released with breaking changes regarding the handling of out of bounds values when casting. This breaks the below line since python's native hash function can output out of bound values:

https://github.com/makepath/xarray-spatial/blob/555706cca8b1fdcd03837c10c3ed77f2b586cd6b/xrspatial/gpu_rtx/mesh_utils.py#L7

2024-06-18T22:51:25.961Z    File "/usr/local/lib/python3.10/dist-packages/xrspatial/viewshed.py", line 1663, in viewshed
2024-06-18T22:51:25.961Z    return viewshed_gpu(raster, x, y, observer_elev, target_elev)
2024-06-18T22:51:25.961Z    File "/usr/local/lib/python3.10/dist-packages/xrspatial/gpu_rtx/viewshed.py", line 280, in viewshed_gpu
2024-06-18T22:51:25.961Z    scale = create_triangulation(raster, optix)
2024-06-18T22:51:25.961Z    File "/usr/local/lib/python3.10/dist-packages/xrspatial/gpu_rtx/mesh_utils.py", line 7, in create_triangulation
2024-06-18T22:51:25.961Z    datahash = np.uint64(hash(str(raster.data.get())))
2024-06-18T22:51:25.961Z    OverflowError: Python integer -2413162623232959393 out of bounds for uint64

The easy workaround for now is to pin the local numpy version to 1.26.4, but it might be a good idea to a) add this to readme/dependencies for now, and/or b) move to a better hash function that does not result in negative numbers.

brendancol commented 4 months ago

@ehiggins0 hey thanks so much for the info and we'll look to upgrade the datahash above