Returning NumPy Array of Wrapped C++ Objects

[keltonhalbert]

I'm interested in being able to return a numpy array of C++ objects that should be wrapped and accessible through Python. I've successfully wrapped the class, and I can create and return numpy arrays of primitive types, but can't seem to find any examples or documentation on how to return an array of custom object types. I assume std::unique_ptr will be needed for ownership...

Can anyone point me to an example, or is this unsupported functionality?

[wjakob]

This is not supported and will never be. Please see the limitations section of the NDArray documentation.

[wjakob]

(Consider using pybind11, which does have this feature due to its narrower-scoped NumPy integration.)

[keltonhalbert]

I understand an appreciate the insight, this was very helpful in preventing me from beating my head against a wall for something the library doesn't support :). I had hoped maybe there was a way to pull this off, since the light-weight and fast bindings provided by Nanobind are really awesome, but such functionality probably comes at the cost to that speed and light-ness.

For those who come across this in the future, I did figure out that you can return a nb::list of custom objects, and later wrap them with numpy. However, this would not solve the issue of receiving an array of dtype=object.