search

sizmailov / pybind11-stubgen

Generate stubs for python modules
Other
218 stars 44 forks source link

Using Eigen::VectorXd results in numpy.ndarray[numpy.float64[m, 1]], Error: Can't find/import 'm' #213

Closed GenieTim closed 4 months ago

GenieTim commented 4 months ago

I have a Python library that makes heavy use of the Eigen library. Pybind11 supports that, thankfully, making it easy to bind the C++ code to Python.

Now I am trying to use this library (v2.4.2) for generating stubs.

In principle, this works fine, I am very thankful to you. An exception are the Eigen types: e.g., an Eigen::VectorXd, which represents a vector or doubles of indefinite length, is represented with a numpy.ndarray[numpy.float64[m, 1]], which results in an error (Can't find/import 'm'). As I understand, this m just symbolizes the dynamic nature, i.e., that these vectors (in Python: float arrays) may have any length.

I don't understand enough about where this stub generation takes its data from to know whether this is an issue with the stub generation or pybind, but I would be thankful for some guidance.

sizmailov commented 4 months ago

Stubgen extracts the signatures and annotations from .__doc__ properties. pybind11 not always produces perfectly valid annotations, numpy.ndarray[numpy.float64[m, 1]] is one of such examples. You can try one of following CLI options to "fix" the stubs: https://github.com/sizmailov/pybind11-stubgen/blob/7ce3b75a1185326933da90fe31ffb799c6f9d08e/README.md?plain=1#L32-L34