Bug: ak.isfinite fails on non-float dtypes (int/uint/bool)

[ajpotts]

Bug: ak.isfinite fails on non-float dtypes (int/uint/bool)

Summary

ak.isfinite raises a runtime error when called on integer, unsigned integer, or boolean arrays. This diverges from NumPy semantics, where isfinite is defined for all numeric dtypes and always returns True for non-floating inputs.

Expected Behavior (NumPy)

np.isfinite(np.array([1, 2, 3], dtype=int))      # array([True, True, True])
np.isfinite(np.array([1, 2, 3], dtype=uint64))   # array([True, True, True])
np.isfinite(np.array([True, False]))             # array([True, True])

Actual Behavior (Arkouda)

ak.isfinite(ak.array([1, 2, 3]))

Raises:

RuntimeError: Error executing command isfinite<1>: cannot cast class to type - runtime types not compatible

This occurs for:

• int64
• uint64
• bool

Root Cause

The Arkouda backend dispatches isfinite<1> for non-float inputs, but the server kernel expects floating-point types and fails during runtime casting.

Impact

• NumPy alignment tests fail
• Users cannot safely call ak.isfinite on non-float arrays
• Inconsistent behavior relative to NumPy and pandas expectations

Proposed Fix

Short-circuit ak.isfinite for non-floating dtypes in the Python wrapper:

• If dtype is integer, unsigned, or boolean:
  • Return ak.full(x.size, True, ak.bool)
• Only dispatch the server-side isfinite kernel for floating-point inputs

This exactly matches NumPy semantics and avoids unnecessary server calls.

Temporary Workaround (Tests)

Mark isfinite for non-float kinds as xfail in NumPy-alignment tests until the backend behavior is fixed.

Related Test Failure

test_unary_alignment[False-isfinite-isfinite-isfinite-kinds]
RuntimeError: cannot cast class to type