Category

• Report an error in the documentation.
• Request for something to be documented.
• Suggestion to improve the documentation.
• Other: Warning / error message

Description

The documentation recommends constructing warp arrays from torch (I guess the same holds for numpy) using from_torch. However, it is also possible to do the following:

a = torch.arange(10)
wp1 = wp.array(a, dtype=wp.int32)
wp2 = wp.array(a, dtype=wp.int64)

While this does not seem to be recommended anywhere, I suggest raising a warning or even an error if this is done and it triggers a data type conversion.

The above example will result in:

wp1 -> [0 0 1 0 2 0 3 0 4 0]
wp2 -> [0 1 2 3 4 5 6 7 8 9]

without any warning, which can lead to hard-to-detect bugs.