Documentation generated with Sphinx

Let's say that I have a C++ class myclass with a method double loglik(const Eigen::MatrixXd& u). To expose this method in python, I have something like .def("loglik", &myclass::loglik, "computes the log-likelihood.", py::arg("u")), which works fine. In other words, when doing help(mymodule.myclass), I see something like loglik(self: mymodule.myclass, u: numpy.ndarray[float64[m, n]]). However, when generating the documentation with Sphinx, the generated signature is loglik(self, u, n]]).

Similarly, I have a method in C++ with some argument const Eigen::MatrixXd& u = Eigen::MatrixXd() (i.e., the only difference is the "default"), which I expose in python using py::arg("u") = Eigen::MatrixXd(), and again this works fine, but the generated signature for the documentation is myfunction(self, u, n]] = array[, dtype])...

And it does not stop here. The same class has a constructor taking an argument const Eigen::MatrixXd& u, which I expose in python using .def(py::init<const Eigen::MatrixXd&>(), "my constructor", py::arg("u")). In this case, the signature generated by Sphinx is __init__(self: mymodule.myclass, u: numpy.ndarray[float64[m, n]]), so this one works as expected...

What am I missing here?

Aucun fichier source ni aucun test n’est indiqué. Commencez par comparer les liaisons de méthodes pybind11 aux signatures produites par Sphinx, en utilisant le constructeur fonctionnel et les méthodes défaillantes comme points d’entrée contrastés. Le travail est terminé lorsque Sphinx conserve la signature complète de numpy.ndarray et la valeur par défaut de ces méthodes.