My C++ library depends on a 3rd party C++ library with its own bindings.
I bind a struct that uses def_readwrite to expose its members. One of its members is a type from the 3rd party library.
Basically I have:
struct MyStruct {
ClassFromThirdParty member{};
}
py::class_<MyStruct>(m, "MyStruct")
.def_readwrite("member", &MyStruct::member)
In Python I try:
from my_bindings import MyStruct
obj = MyStruct()
print(obj.member)
but this raises TypeError: Unable to convert function return value to a Python type!.
It's worth also noting that if I do:
import ThirdPartyLibrary
from my_bindings import MyStruct
obj = MyStruct()
print(obj.member)
no error is raised. But I don't like this solution as the Python user would have to import ThirdPartyLibrary even if they don't explicitly need it.
How do I write my binding such that the first Python snippet works?
PS: The third party binding in question can be found here. In the absence of a general answer, I'd also be happy to hear answers related to that library in particular.
It's the same in pybind11 as it is in Python -- you need to import classes before you can rely on them.
Add py::module::import("pydrake.math"); to your PYBIND11_MODULE before trying to define function bindings that operate on a pydrake.math.RigidTransform as a C++ argument type or C++ return value type.