Deprecate is_extension_type

[TomAugspurger]

In favor of is_extension_array_type. This is blocked till sparse, period, and datetimetz are all EAs.

---

This should probably be True

In [3]: pd.api.types.is_extension_type(pd.interval_range(0, 10))
Out[3]: False

Likewise for Period?

In [3]: pd.api.types.is_extension_type(pd.period_range(2017, periods=2))
Out[3]: False

Currently we hardcode checks for categorical, sparse, and datetime w/ tz. A general solution would be to have all our extension types inherit from some base class or satisfy the interface required for an extension type.

xref #18767

[toobaz]

If i understand correctly, @TomAugspurger you are (in today's terms) just suggesting that is_extension_type should return True if is_extension_array_dtype does, right? (Notice this is already true for Categorical).

Then I think the only two hardcoded types would be datetimetz and sparse.

By the way: is there any plan to make those extension types too? The fact that

In [2]: s = pd.date_range(start='2010-08-08', end='2010-08-09', tz='utc')

In [3]: s._values is s
Out[3]: True

is quite annoying in parts of the code like this:

pandas/pandas/core/base.py

Line 942 in ffae158

values = self._values

which assume that the _values of an Index are not an Index themselves.

[TomAugspurger]

I think eventually we'll just remove is_extension_type, in favor of is_extension_array_dtype.

For context, changing everything in one go was too difficult, so we've kept is_extension_type for backwards compat. Once everything is an EA, we could remove it (after deprecating).

By the way: is there any plan to make those extension types too? The fact that

(just replied on the mailing list). I just opened a PR for sparse.

[jbrockmendel]

closed by #29457