DanCardin
forcing less strict type checking on projects that use cappa
The biggest barrier preventing me from using cappa right now is that it forces me to use less strict type-checking in my project, because of all of the untyped parts of the cappa API.
I see that it would be a lot of work to introduce strict typing to the whole cappa code base.
A way to help make less of this untyped API bleed over into the public interface could be to do a stricter type check on just the examples directory (since that uses a lot of the public interface).
This would be much less work than introducing strict type checking into the whole cappa code base.
I'd be interested in any type checking problems, and would consider them bugs. i see you addressed some of them in your PR. I dont currently encounter with the settings i'm using personally, which is why they're not currently fixed :P
If you want to see what I'm seeing, you can add this to the pyproject.toml:
[tool.pyright]
typeCheckingMode = "strict"
I use something close to that setting (a little bit of customization around it), and I can't use cappa without it reporting lots of issues all over the cappa API.
That reports thousands of issues if you check the whole cappa code base. That's why I say it would be a lot of work to address all of that.
But it's significantly less work to just address what it reports on the examples, and that covers a lot of the API that users will see.
from that, i would believe that the untyped params on the inputs of @command would be causing what at least I am seeing on my own CLIs, but i wouldnt think (from reading their docs) that dict vs dict[Any, Any] would be meaningfully different.
I certainly have been typing like that when i can't be meaningfully more specific than the concrete part of the type. kind of annoying if that's not "strict" compliant, but i can definitely look more into it.
you only addressed some of the issues yes? like i said on the PR, i'm definitely seeing conflicting issues with mypy given your PR, and I believe i'm seeing similar issues on the invoke api.