FWIW I don't like this. partial should remain within a plugin. It is not special enough to make significant changes in core mypy logic.

TBH, I tend to agree with @ilevkivskyi - there's a balance between "we want to support this case correctly" and "we don't want to embed too many library-specific details into the typechecker". As far as I see, there's no other way to track such partial applications to ParamSpec components in any other way, but maybe the right thing is to just say "we don't want to support this".

@ilevkivskyi tagging you since you gave some feedback on the first revision. I managed to remove the scary part - error messages are less specific now, but only special_sig = "partial" escapes the plugin boundary. Are you comfortable with this amount of special-casing? (filtering error messages for "too few arguments" is way too fragile to consider here, IMO - those diagnostics can be issued by multiple checks, and we're only interested in one specific source of those)

LG, I don't really understand the bound start args logic, but it will be easy to tweak if it will cause problems.

I just invented a counterexample to that test, it's indeed not sufficient to check by arg kinds. Last commit should address that problem and make the logic more clear as well.

Diff from mypy_primer, showing the effect of this PR on open source code:

starlette (https://github.com/encode/starlette)
- starlette/concurrency.py:38: error: Too few arguments  [call-arg]