Question about behavior with no array passed to dpt.broadcast_arrays #1718
Description
The below example works differently with numpy and numpy.array_api:
import numpy, dpctl, dpctl.tensor as dpt
import numpy.array_api as na
numpy.broadcast_arrays()
# Out: []
na.broadcast_arrays()
# Out: []
numpy.__version__
# Out: '1.26.4'
dpt.broadcast_arrays()
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
Cell In[9], line 1
----> 1 dpt.broadcast_arrays()
File ~/miniforge3/envs/dpnp_dev/lib/python3.11/site-packages/dpctl/tensor/_manipulation_functions.py:272, in broadcast_arrays(*args)
269 if not isinstance(X, dpt.usm_ndarray):
270 raise TypeError(f"Expected usm_ndarray type, got {type(X)}.")
--> 272 shape = _broadcast_shapes(*args)
274 if all(X.shape == shape for X in args):
275 return args
File ~/miniforge3/envs/dpnp_dev/lib/python3.11/site-packages/dpctl/tensor/_manipulation_functions.py:83, in _broadcast_shapes(*args)
78 """
79 Broadcast the input shapes into a single shape;
80 returns tuple broadcasted shape.
81 """
82 array_shapes = [array.shape for array in args]
---> 83 return _broadcast_shape_impl(array_shapes)
File ~/miniforge3/envs/dpnp_dev/lib/python3.11/site-packages/dpctl/tensor/_manipulation_functions.py:43, in _broadcast_shape_impl(shapes)
41 mutable_shapes = False
42 nds = [len(s) for s in shapes]
---> 43 biggest = max(nds)
44 sh_len = len(shapes)
45 for i in range(sh_len):
ValueError: max() arg is an empty sequence
dpctl.__version__
# Out: '0.18.0dev0+77.g26d34f565d'From the Python array API it doesn't look clear enough if empty list should be supported and the empty list to be returned.
In any case that corner case needs to be properly handled by dpctl, because the raised exception ValueError: max() arg is an empty sequence doesn't sound clear about the reason of failure.
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels