Create new PyObject from std::type_info and pointer

[noahbkim]

Hi all, I'm in an odd situation where I have the std::type_info of a primitive/STL object (e.g. int, float, std::string) and an untyped pointer to it (char*). I would like to copy/cast this data into a PyObject to be returned by a bound function.

I imagine this is possible with existing machinery, but I'm having trouble figuring out how! I'd greatly appreciate any help :D

I've gotten as far as the following:

nanobind::object cast_value(
    const std::type_info* type_info,
    void* address,
    nanobind::rv_policy policy = nanobind::rv_policy::copy
)
{
    nanobind::handle handle = nanobind::detail::nb_type_put(type_info, address, policy, nullptr);
    if (!handle.is_valid())
    {
        nanobind::detail::raise_cast_error();
    }
    return nanobind::steal(handle);
}

When provided with the std::type_info from unsigned int and a valid pointer to one, handle.is_valid() is false--maybe I'm missing something about how nb_type_put should be used.

Followup: it seems like at the time of casting, std::type_index(typeid(unsigned int)) is not in internals.type_c2p. I'll try instantiating a static caster for each type I'd expect to show up.

Followup: oh, I guess internals.type_c2p is for registered classes--primitives are handled by a custom caster (which doesn't seem to be accessible via std::type_info. So I guess the best route is to switch on the type info...

[wjakob]

If it's primitive types, your best bet will be to iterate through all of the possibilities, static_cast in C++, and then call nb::cast. The nanobind data structures would only be helpful for types exposed using bindings (i.e. via nb::class_<T>), which isn't even possible for int etc.

[noahbkim]

Thanks for the quick response!