For Kodiak we needed to find html comment positions inside a string of markdown. All the existing Python markdown parsers didn’t expose the necessary AST node positions, but the Rust package, pulldown-cmark did.

With a little more Rust code and pyo3, we get a working Python module as a .whl.

The only downside is that pyo3 doesn’t generate type stubs (.pyi files) for the package and since the module is a binary, using the module can’t be type checked.

So what can we do?

  1. Write the type stub by hand

    For markdown-html-finder it looks like this:


    from typing import List, Tuple
    def find_html_positions(markdown: str) -> List[Tuple[int, int]]: ...
  2. Add the type stub to the generated .whl

    Wheel provides a CLI for unpacking and packing wheels.

    ./.venv/bin/wheel unpack markdown_html_finder-0.1.0.whl
    # need the py.typed file so type checkers will use types
    # see:
    touch markdown_html_finder/py.typed
    cp -R ./markdown_html_finder markdown_html_finder-0.1.0
    ./.venv/bin/wheel pack markdown_html_finder-0.1.0
  3. Install the updated .whl and using the module will now be type checked. 🎉

For a more extensive example, checkout the build script in the markdown-html-finder repo.