Slice attributes can be of any type.

[brandtbucher]

>>> slice([], object(), type)
slice([], <object object at 0x7f2e0e522120>, <class 'type'>)

[JelleZijlstra]

They theoretically can be, but is it actually useful to construct slices like the one in your comment? I'm worried that this will sacrifice some type safety for code that uses standard integer slices for little real benefit.

See python/typing#159 for prior discussion on slices.

[brandtbucher]

I agree that it's uncommon (even impossible) to use when slicing the built-in types. In data processing libraries, however, it is not unusual to slice on any key type. I encountered this false-positive when working on StaticFrame, which allows slicing by index values:

https://static-frame.readthedocs.io/en/latest/api_selection.html?highlight=slice#frame

The widely used Pandas library allows this as well:

https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#selection-by-label

Slices are rarely created using the slice constructor in typical code (attribute access is rare as well), but it's very common to explicitly build and manipulate them internally in libraries such as these.

(Thanks for the link. I did a quick search for this, but it didn't occur to me to check the typing repo!)

[brandtbucher]

So is this something that can be considered?

[JelleZijlstra]

I guess we should do this then to avoid false positive type errors.

[brandtbucher]

Bump. Is there anything keeping this from being merged?

[rushabh-v]

It gives error: Slice index must be an Integer of None for below kind of code.
pd.IndexSlice[:"b":-1]
pd.IndexSlice["m":"m":-1]
see pandas-dev/pandas#29188