[scikit-image] Re: NumPyDoc validation

On 12 Sep 2019, at 5:50 am, Egor Panfilov wrote: I think the maintenance cost of such custom linter would be comparable or higher than the benefits of using it. One docstring example, a link to NumPy standard, and Sphinx translator in CI should be sufficient for everyone to write "good enough" docstrings.

The review workflow is “good enough” for sure, and we’ve been doing all right with that from the very beginning of the project (together with building the docs with Travis). However, one use case for *perfect* docstrings is that we can potentially extract type annotation information automatically if all of our docstrings conform to the correct standard. That is potentially extremely valuable.