Arguments and return type hints are fully qualified although they should not be

[peske]

First, thanks for the great work!

Let me describe the problem on the real-world example. My function is defined in the following way:

from collections.abc import Sequence

from nptyping import Float64, NDArray, Shape


def create_operator_matrices(  # noqa: C901
    relative_radii: Sequence[float],
    r_min: float | None = None,
) -> tuple[NDArray[Shape["N, N"], Float64], NDArray[Shape["N, N"], Float64]]:
    pass

The extension configuration:

# typehints_fully_qualified = False
# always_document_param_types = False
# typehints_document_rtype = True
# typehints_use_rtype = True
typehints_defaults = "braces"
# simplify_optional_unions = True
# typehints_formatter = None
typehints_use_signature = True
typehints_use_signature_return = True

The resulting function signature in the generated docs:

grc.create_operator_matrices(relative_radii: collections.abc.Sequence[float], r_min: float | None = None) → tuple[nptyping.ndarray.NDArray[nptyping.base_meta_classes.Shape[N, N], numpy.float64], nptyping.ndarray.NDArray[nptyping.base_meta_classes.Shape[N, N], numpy.float64]]

Parameters in the resulting docs are as follows:

relative_radii: collections.abc.Sequence[float]

The problem is obvious: instead of having things like nptyping.ndarray.NDArray and nptyping.base_meta_classes.Shape[N, N], I expect simply NDArray and Shape[N, N].

Strangely enough, the "Return type" section of the generated documentation looks more or less ok:

tuple[NDArray[Shape[N, N], float64], NDArray[Shape[N, N], float64]]

The only problem with the "Return type" is that it uses float64 instead of Float64 (see the actual function signature).

Any help will be appreciated!

[peske]

Probably would be nice to have a configuration option that says something like "do not do anything smart - just use the existing hints from the original signature".

[gaborbernat]

I don't think that's possible because python does not show/give us access to the as typed names, but only to the runtime version of it. PR welcome though if you can make it work.

[peske]

Hmmm, the first part of the problem does not look to hard to me at the first glance. If typehints_fully_qualified = False, We can simply regex and remove period-concatenated words, right? Just identify things like nptyping.ndarray.NDArray in the output, and strip it to NDArray.

@gaborbernat what do you think, is this approach too naive?