add Column.dtype

[MarcoGorelli]

closes #165

[rgommers]

Strangely enough CI is not happy with this:

docstring of dataframe_api.Column.dtype:1: WARNING: py:class reference target not found: dataframe_api._types.dtype

[jorisvandenbossche]

Should this say something about the returned object? Or do just see it as an opaque object without any guarantees? (except that you can pass it to functions with an argument that expects a dtype?)

Or do we want to specify that certain attributes or methods are available? (eg is __eq__ expected to work)

[rgommers]

Should this say something about the returned object? Or do just see it as an opaque object without any guarantees?

I'd say there are guarantees for dtypes that are returned, but that's gh-155 rather than this PR. We already use dtype in some places, and this PR is just one more - so this PR doesn't have to be the one to work out the hard part.

[MarcoGorelli]

so this PR doesn't have to be the one to work out the hard part.

besides, the hard part of this PR is figuring out how to make it pass CI ;)

EDIT: done 🎉 it was just that def dtype shadows the dtype type. Solutions could either be import dtype as dtype_t, or rename def dtype to a function like def get_dtype. I've gone with the latter

[rgommers]

it was just that def dtype shadows the dtype type. Solutions could either be import dtype as dtype_t, or rename def dtype to a function like def get_dtype. I've gone with the latter

I feel like the name chosen for type annotation earlier is a detail, and should not prevent choosing the most obvious name for the property (or method, but I think a .dtype property as you had it is best).

dtype_t works, although perhaps DType is better - I think that's the most often used solution, capitalize type annotation names.

[MarcoGorelli]

perhaps DType is better

agreed, thanks, have gone with that

[rgommers]

This is green and seems as uncontroversial as it can get for an API addition, so let's get this in.