Is there a way to parse docstrings in Cython (.pyx) files ?

I have a Cython module that I can import this way : import cython_module. Its source code is in a cython_module.pyx file, at the same level of other .py files that already work well with mkdocstrings:

.
├── cython_module.pyx
├── predict.py
├── setup.py
└── etc

This cython module contains a class with a docstring:

cdef class Model:
    """
    This is the lower-level class used for inference, it stores the model in an optimized way and features a predict method.

    Attributes:
        trees : The array of trees composing the model.
        nb_trees : The number of trees in the model.
    """

    cdef Tree* trees
    cdef uint16_t nb_trees

When specifying a ::: cython_module in one of the .md files, I get this warning: WARNING - mkdocstrings.handlers.python: Couldn't read source for 'cython_module': source code not available, and in the generated doc, the only thing showed after “cython_module” is the class name (“Model”), but no docstring, nor attributes.

Is there a way to generate documentation from docstrings in Cython files/modules ?

Thanks!

[WillDaSilva]

Looks good to me. Works on the Cython projects I can’t share, as well as Numpy, Pandas, and more. Great work!

[pawamoy]

This is indeed what I want to do: compile the code into an AST to re-use the utilities I’ve written for visiting ASTs. But for this, I actually need the string representation of these actual Python objects 😄 Anyway, I think I’ll simply go with annotation.__name__ for builtin types, and repr(annotation) for the rest, at least for now. Thanks for sharing ideas 🙂

And yes, I just read all the discussions around PEP 563 and PEP 649. If PEP 563 is deprecated, we’ll have to deal with actual Python objects for annotations anyway 🙂 But it won’t impact plain-text modules since I can parse them and play with their AST. It only impacts compiled modules which can only be inspected, so we should be totally fine.