COMPAT: can we reduce pathological cases in Index comparisons

[jreback]

from SO

In [1]: i1 = pd.DatetimeIndex(['1971-10-01'], dtype='datetime64[ns]', freq=None)

In [2]: i2 = pd.Index([u'100171'], dtype='object')

In [3]: i1
Out[3]: DatetimeIndex(['1971-10-01'], dtype='datetime64[ns]', freq=None)

In [4]: pd.DatetimeIndex(i2)
Out[4]: DatetimeIndex(['1971-10-01'], dtype='datetime64[ns]', freq=None)

We might be able to elminate the coercion in .equals() in some cases.

[sinhrks]

Met the same issue when fixing #13107. How about stop coercing normal Index if its inferred_dtype is string like?

[jreback]

I agree I think coercion needs to be explicit
actually I don't think we should wrap in DTI at all
of its not already the same class that it's not equal