BUG: ser.to_frame().all() inconsistent with ser.all() with CategoryDtype

[jbrockmendel]

Based on tests.frame.test_analytics.test_any_all_np_func

ser = pd.Series([0, 1], dtype="category", name="A")
df = ser.to_frame()

>>> ser.all().
TypeError: Categorical cannot perform the operation all

>>> df.all().item()
False

Whats happening here is the DataFrame case is calling self.values which casts to int64 (avoidable with 2D EAs...).

If instead we refactor (xref #35881) to operate blockwise, this block is ignored, so we end up getting the result we'd get on an empty DataFrame, i.e. True. (xref #28900 ignoring failures is a footgun).

Expected Behavior: as long as have ignore_failures, df.all() should be True

[jorisvandenbossche]

Same happens for datetime/timedelta -> #34479 (which came out of the same PR experiments)

[jbrockmendel]

Any preferences on how to handle the inconsistency?

[jbrockmendel]

From the most recent call, the consensus seems to be that we should deprecate the current behavior in favor of matching the Series behavior

[jorisvandenbossche]

Yes, I would be in favor of that as well. The user can always convert to boolean to be explicit.