TypeError with numpy.can_cast() in newer NumPy versions (NEP 50 compatibility issue)

[yahah100]

TypeError with numpy.can_cast() in newer NumPy versions (NEP 50 compatibility issue)

Description

The library fails with a TypeError when using newer versions of NumPy due to changes in numpy.can_cast() behavior as part of NEP 50 adoption.

Error Details

TypeError Traceback (most recent call last)
File ~/workspace/HWS-Algo-Dev/.venv/lib/python3.11/site-packages/pysheds/sview.py:85, in Raster.new(cls, input_array, viewfinder, metadata)
84 try:
---> 85 assert np.can_cast(viewfinder.nodata, obj.dtype, casting='safe')
86 except:
TypeError: can_cast() does not support Python ints, floats, and complex because the result used to depend on the value.
This change was part of adopting NEP 50, we may explicitly allow them again in the future.

Root Cause

The issue occurs in pysheds/sview.py:85 where np.can_cast() is called with Python scalar types (like int, float) rather than NumPy scalar types. NumPy's NEP 50 implementation removed support for Python scalars in can_cast() because the behavior was value-dependent and inconsistent.

Proposed Solution

Replace Python scalar default values with NumPy scalar types throughout the codebase. For example:

Fixed code:

# sgrid.py line 624 instead of nodata_out = 0
def _d8_flowdir(self, dem, nodata_cells, nodata_out=np.int64(0), flats=-1, pits=-2,
                    dirmap=(64, 128, 1, 2, 4, 8, 16, 32)):

# sgrid.py line 602 instead of nodata_out = 0
nodata_out = np.int64(0)

Files Affected

Based on the error trace and code inspection, the following areas need attention:

pysheds/sgrid.py - Several functions already use np.int64(0) correctly (lines 602, 624-625) but others may need updates

Environment

• NumPy version: Recent versions implementing NEP 50
• Python version: 3.11+
• pysheds version: pysheds==0.4

[st4ck3]

The issue described is real, and the proposed solution is both simple and effective. Why hasn't this change been implemented? Is the project still actively maintained? Should it be forked and patched?

[JonKing93]

Hi @st4ck3 and @yahah100, just letting you know that this project appears to be abandoned. That said, we're planning to set up a community fork later this month (read #272 for details), and this numpy issue will be the first priority. And if you'd like to contribute to the fork, we'd welcome your input!

[st4ck3]

I totally agree with the idea of reactivating this project!

[yahah100]

I agree, this is a good suggestion. I've implemented a few quick fixes in my fork here: https://github.com/yahah100/pysheds. I'm sure there's a more elegant solution, but this should resolve the immediate issue.