FIX: Add type check for shape elements in DeviceNDArrayBase constructor

[gnzng]

This PR addresses issue #351 by adding a type check to ensure all elements of the shape argument are integers during DeviceNDArrayBase construction.

Description

numba_cuda.device_array would accept a shape argument containing non-integer types, such as booleans or floats. The values were implicitly cast to floats, which could lead to unexpected behavior or downstream errors.

Behavior Before This PR

The following code would execute without raising an error, creating an array with a shape composed of floats:

import numpy as np
from numba_cuda import device_array

# No error was raised
arr = device_array(shape=np.array([True, False, 10, 10.12]))

# Resulting shape would be (1.0, 0.0, 10.0, 10.12)
print(arr.shape)

Behavior After This PR

With this change, providing non-integer elements in the shape argument now raises an immediate TypeError, ensuring that only valid shapes are used for array creation.

import numpy as np
from numba_cuda import device_array

# Now raises a TypeError
arr = device_array(shape=np.array([True, False, 10, 10.12]))

The resulting error is:

TypeError: all elements of shape must be ints

This change improves input validation and prevents the creation of improperly configured device arrays.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[gmarkall]

/ok to test

[copy-pr-bot]

/ok to test

@gmarkall, there was an error processing your request: E1

See the following link for more information: https://docs.gha-runners.nvidia.com/cpr/e/1/

[gmarkall]

/ok to test 85a4211

[gmarkall]

Thanks for the PR!

When a scalar float is passed for the shape, we get an error before the new check sees it (and I think the new check would also fail as a float is not iterable):

In [3]: cuda.device_array(1.2)

File ~/numbadev/numba-cuda/numba_cuda/numba/cuda/api.py:158, in device_array(shape, dtype, strides, order, stream)
    152 @require_context
    153 def device_array(shape, dtype=np.float64, strides=None, order="C", stream=0):
    154     """device_array(shape, dtype=np.float64, strides=None, order='C', stream=0)
    155 
    156     Allocate an empty device ndarray. Similar to :meth:`numpy.empty`.
    157     """
--> 158     shape, strides, dtype = prepare_shape_strides_dtype(
    159         shape, strides, dtype, order
    160     )
    161     return devicearray.DeviceNDArray(
    162         shape=shape, strides=strides, dtype=dtype, stream=stream
    163     )

File ~/numbadev/numba-cuda/numba_cuda/numba/cuda/api_util.py:11, in prepare_shape_strides_dtype(shape, strides, dtype, order)
      9     strides = (strides,)
     10 else:
---> 11     strides = strides or _fill_stride_by_order(shape, dtype, order)
     12 return shape, strides, dtype

File ~/numbadev/numba-cuda/numba_cuda/numba/cuda/api_util.py:16, in _fill_stride_by_order(shape, dtype, order)
     15 def _fill_stride_by_order(shape, dtype, order):
---> 16     nd = len(shape)
     17     if nd == 0:
     18         return ()


TypeError: object of type 'float' has no len()

What do you think about addressing this variant of the error too?

Would you also like to add a test of the fix(es)?

[gnzng]

Thanks a lot!

I added the float check further upstream in prepare_shape_strides_dtype() ae2305b

Added tests for testing various inputs and input dtypes:

• 36ed9ef (test for right errors)
• 6274af6 (make sure list and np.arrays are still working with valid ints)

[gmarkall]

/ok to test 6274af6

[gnzng]

Fix style via ruff format api_util.py since style CI ci/check_style.sh was failing a06a697

[gmarkall]

/ok to test a06a697

[gmarkall]

Many thanks!