Pandas BUG/API: preserve dtype in Index append()

If you combine two Index objects using append, if they are both object dtype to start with, it seems we still infer the dtype for the result instead of preserving object dtype:

idx1 = pd.Index(["b", "a"], dtype=object)
idx2 = pd.Index(["c", "d"], dtype=object)

>>> idx1.append(idx2)
Index(['b', 'a', 'c', 'd'], dtype='object')

>>> pd.options.future.infer_string = True
>>> idx1.append(idx2)
Index(['b', 'a', 'c', 'd'], dtype='str')    # <-- upcast to string

I would expect that to preserve the dtype of the calling / passed Index (or at least its "common" dtype, which in this case clearly is object dtype)

Comment From: rhshadrach

Hopefully a rare case, but what about the behavior of:

idx1 = pd.Index([1, 2], dtype=object)
idx2 = pd.Index([3, 4])
print(idx1.append(idx2))

I would think this should also be object, as otherwise we get values-dependent behavior.

Comment From: Swati-Sneha

take

Comment From: jorisvandenbossche

I would think this should also be object, as otherwise we get values-dependent behavior.

Indeed, I would also expect object dtype in that case

Comment From: jbrockmendel

+1 for getting rid of dtype inference. want to Just Do It for 3.0?