Reference to internal are not registered for EigenSparseMatrix + errors with pickable

[cmastalli]

If I tried to bind a sparse matrix using .def_readwrite, then I obtained

TypeError: No Python class registered for C++ class Eigen::SparseMatrix<double, 0, int>

If my understanding is correct, it is because Eigenpy is not yet registering references to Eigen's sparse matrices.

Additionally, I observed that when loading a pickable, the number of rows in an Eigen sparse matrix is always lower than the original matrix. I believe the error is coming from Eigenpy as well. I can provide more details if needed.

Thanks in advance,
Carlos

[jcarpent]

We need more details of course. You can expose yourself the SparseMatrix types. All the helpers are there.

Yet, I don't understand why you are using int instead of Eigen::Index, for the third template arg.

[cmastalli]

I apologize for not responding earlier, I have been crazily busy these days. Anyway, let me try to better explain this issue. There are two points to discuss.

Python registration

If I try to bind a sparse matrix like this

          .def_readwrite("A", &MyClass::A, "my sparse matrix")

and write a Python code like this

m = MyClass()
m.A

Then it prints

TypeError: No Python class registered for C++ class Eigen::SparseMatrix<double, 0, int>

even if I use Eigen::Index. The int is because we are in Python.

However, I share below a workaround that allows me to retrieve a sparse matrix

          .add_property(
              "A",
              bp::make_function(
                  +[](const MyClass& self) -> Matrix { return self.A; },
                  bp::return_value_policy<bp::return_by_value>()),
              +[](MyClass& self, const Matrix& A) { self.A = A; },
              "my sparse matrix")

It would be nice not to do this, as I am unable to write internal values (although it is not recommended for sparse matrices).

Retrieving a pickable sparse matrix

This bug disappeared after updating eigenpy. I can't tell you anymore. It is unclear to me where the source of the error was. Likely to be in an old eigenpy version.

[jcarpent]

Please share a complete reproducible example
for this case. Best way for us to understand really your issue.