Check whether a column is a Datetime without relying on default Datetime LogicalType

[tamargrey]

In checking whether a column is a valid time index, we check if ltype_class == ww.logical_types.Datetime, which means that if a user wanted to deregister the Datetime LogicalType and add in one of their own, they would not be able to make any columns that use that LogicalType a time index.

We should have another way of defining if a LogicalType is permissable for time indices.

• Maybe a datetime standard tag (not great, because it'll look weird with the Datetime Logical Type)

[gsheni]

@tamargrey not sure of the solution, but I agree that datetime is a weird semantic tag, and doesn't fit the "provides additional data about the meaning of the data".

Is there a situation where someone might want to change the Datetime Logical type with their own custom Datetime Logical Type? With Datetime I can imagine custom datetime formats, and timezones. But those would be handled with our datetime_format.

[tamargrey]

@gsheni I'm not sure why someone would want to change the dtype, but maybe there's a situation where someone needs to have the object dtype, so the normal solution would be to just create a new Datetime LogicalType.

Another idea is that when we define is_numeric and is_categorical we could do it in such a way that users can change that definition themselves, and we can have an is_datetime that currently checks for the Datetime LogicalType, but it can be changed to another Logical Type if desired.

But this might just be something that we can wait to handle until someone specifically asks for it.

[tyler3991]

Ice boxing on zenhub for now