TYP: NDFrame.pipe, GroupBy.pipe etc.

[topper-123]

Type up pipe function and methods. This allows type checkers to understand the return value type from pipes, which is nice.

[jreback]

cc @simonjayhawkins

[simonjayhawkins]

Thanks @topper-123 generally lgtm

[simonjayhawkins]

imports from _typing are already guarded.

[simonjayhawkins]

same

[simonjayhawkins]

same

[topper-123]

Updated.

[topper-123]

I think this is ok to merge?

[simonjayhawkins]

I think so. We may want to have an alias for this, but as the type is not yet complete I think we should leave that to later.

I say incomplete since we will replace the ... of Callable[..., T] at some point, and that will require a second typevar to bind to self/obj.

but in the spirit of gradual typing, this gets the return type sorted.

[topper-123]

.pipe takes very generic callables, so I don’t think it’s possible to type more. Can you give an example what you’re thinking?

In python 3.10 there’s something coming that may improve typing for .pipe, but that’s far in the future for Pandas...

[simonjayhawkins]

if func is a tuple, it is indeed not possible to bind to a keyword argument, but if func is not a tuple and args is not specified, then the function should take just 1 positional argument which is the object?

so one of the overloads would be

def pipe(
    obj: S, func: Union[Callable[[S], T], Tuple[Callable[..., T], str]], **kwargs
) -> T:

where S is a TypeVar and args not specified (args may need to be None, I can't remember off the top of my head)

Indeed, this is out of scope for now and I have approved the changes as is.

[topper-123]

Ok. I think it’s ok to wait with that...

[simonjayhawkins]

yeah, since we maybe pulling in some type stubs, #28142 (comment), I suspect that we can't yet inline those stubs where the types are not compatible with Python 3.7 (without typing_extensions) and I suspect the overloads will be in stub file instead of inline ( and may be able to adopt PEP 612 sooner.)

[simonjayhawkins]

Thanks @topper-123