Inquiry about future support for custom NumPy dtypes in Pythran

[Ye980226]

I am an avid user of Pythran for accelerating numerical computations in Python and have greatly benefited from its ability to optimize NumPy operations. I have a question regarding the support for custom NumPy dtypes in Pythran.

I often use custom dtypes in NumPy for more complex data structures and was wondering if there are any plans to support custom NumPy dtypes in Pythran. This feature would be particularly useful for [describe your use case].

Here's an example of how I use custom dtypes in NumPy:

import numpy as np

# Example of a custom dtype
custom_dtype = np.dtype([('field1', np.float64), ('field2', np.int32)])

# Creating an array with the custom dtype
array = np.array([(1.0, 2), (3.0, 4)], dtype=custom_dtype)

[serge-sans-paille]

Hey @Ye980226 , from a typing point of view, it seems possible to build a custom dtype using pythran typing system. I am slightly less confident in the capability to convert that back to python mode (if you were to return that array), or to build generic field accessor (because you have a runtime value that would be part of the typing system).

I don't plan to support this any soon, it falls in the same category as user-defined classes, with an extra level of complexity and I like time (if not skill) to implement that in a proper way.