Could you actually tell me why you want the list of public objects, and how you want to use it? I'm not sure what you're doing with it, and so not sure why on_env is too late. You want to render the list of public objects in a page, probably?

Sure! For two purposes.

1. When show_bases: true, to rewrite the bases from somelib._internal.foo.Foo into their public somelib.Foo.
2. In function parameter annotations and return annotations, to rewrite from just Foo to somelib.Foo.

Overall this means that an object is consistently referred to by their fully-qualified public name in all locations.

I believe mkdocstrings actually renders everything into HTML during the Markdown phase of the MkDocs flow, which all happens prior to on_env. I can see why this is the case -- mkdocstrings works primarily by installing a blockprocessor -- so I can see that this is just an unfortunate edge case that's hit by an otherwise pragmatic choice.

If you're curious then I've just made my set of tweaks public here: https://github.com/patrick-kidger/hippogriffe

Thanks!

1. Ideally, Griffe would provide the API to allow mkdocstrings-python to show the public locations of objects (somelib.Foo) instead of their canonical ones (somelib._internal.Foo). I'm hitting this myself 😄
2. There's the annotations_path that could help you, maybe, though it affects both types and values rendered (we have an issue in the backlog to make the two distinct).

Overall this means that an object is consistently referred to by their fully-qualified public name in all locations.

Yep we should work towards this goal.

If you're curious then I've just made my set of tweaks public here: https://github.com/patrick-kidger/hippogriffe

Cool name haha!

Looking at the images, I'm not sure how your tuple[str, str, str] or Union[int, str] annotations are rendered as tuple and Union. This should work natively. Feel free to send a bug report.

For annotation modernization (Union[int, str] -> int | str), you might be interested in the modernize_annotations option (Insiders only, that's maybe why you're not using it 😄).

1. Haha!
2. Ah nice, I hadn't seen this. FWIW I ended up rendering types using my wadler_lindig library, but possibly I could have avoided that.

I'm glad you like the name :D

As for the annotations, I'll write up a MWE and bug report.

For modernization I'm just running pyupgrade on my codebase instead :)