py::vectorize with multiple outputs

[grzegorz-graczyk]

Hi, I was wondering what are the options of vectorizing function with multiple outputs. A good example, maybe missing in tests/examples could be just a divmod operation: [](int a, int b) -> T { return { a / b, a % b }; }. But it is probably much more common in practice when dealing with 2d or 3d coordinates.

I know it is possible to have a custom type and bind it to a structured array. It works great, but the resulting Python code is a bit unusual - because structured arrays are not typically used for that.

C = bindings.divmod(A, B)
A_result = B * C["div"] + C["mod"]

What I am wondering is if there already is a way to use py::vectorize to get API that is more consistent with how this is typically done:

div, mod = binidings.divmod(A, B)
A_result = B * div + mod

or

C = bindings.divmod(A, B)
A_result = B * C[:, 0] + C[:, 1]