I'd suggest we make the typevar DTypeT, and then introduce a type alias DType, which would resolve #189

That sounds reasonable to me. Wouldn't you want to introduce the alias in this same PR, so that the signatures and how they're rendered don't change?

Hmm, that didn't completely work, maybe because of the if TYPE_CHECKING. In the preview for this PR we have for Column.get_rows (as an example):

get_rows(indices: Column[Any]) → Column[DtypeT]

so compared to in main, Dtype changed to DtypeT: Column.get_rows

Isn't that one correct? DTypeT is the typevar, so that's what we need to use in Column[

So, we use DType when annotating dtype, e.g. in column_from_sequence, and DTypeT when narrowing down the type of a generic Column (Column[DTypeT])

Oh wait, I was looking at a partial diff, you actually did change the get_rows signature in this PR still. I thought the point of our discussion on type aliases was to revert that, and only use DType. A "typevar" does not belong in any html rendering, it's an irrelevant type-checking detail to the reader.

happy to agree to disagree on this one, closing then