Remove some dead code in core/arrays/interval.py

[Dr-Irv]

The typing work both @twoertwein and I did in #46080 and #44922 found some code that I conjecture is dead code as it doesn't make sense from a typing perspective.

So this PR removes the dead code, which will then allow either of those PR's to get rid of some #type : ignore lines.

Not clear to me how to add tests for dead code removal. In theory, all tests should pass when dead code is removed.

[jbrockmendel]

i think this will break if self._left is a DTA/TDA?

[twoertwein]

I'm honestly not sure what type _left and _right has (in #44922 I assumed np.ndarray but added a note that this is probably not correct - needs some mypy ignore comments). In your attempt at interval.pyi, you marked it as on of np.ndarray, "DatetimeArray", "TimedeltaArray".
https://github.com/pandas-dev/pandas/pull/41059/files#diff-fb052ef16d456d32858ec68d7e9449d4749166ea97c9a79b22d0ccf233dbb3a4R205

[Dr-Irv]

i think this will break if self._left is a DTA/TDA?

Well, that's why I did the test. I don't think that self._left can be a DTA or TDA. I guess there is other logic that causes that never to be the case, and the tests seem to confirm that.

[jbrockmendel]

dti = pd.date_range("2016-01-01", periods=3)
ii = pd.IntervalIndex.from_breaks(dti)
ser = pd.Series(ii)

ser.unique()  # <- gets here with DTA for self._left

[Dr-Irv]

Yes, but the dead code doesn't go there. The old code is:

        if needs_i8_conversion(dtype):
            new_left = type(self._left)._from_sequence(nc[:, 0], dtype=dtype)
            new_right = type(self._right)._from_sequence(nc[:, 1], dtype=dtype)
        else:
            new_left = nc[:, 0].view(dtype)
            new_right = nc[:, 1].view(dtype)

The new code is just:

        new_left = nc[:, 0].view(dtype)
        new_right = nc[:, 1].view(dtype)

In your example, the test needs_i8_conversion(dtype) is False

Here is what happens with 1.4.0 and this PR:

>>> import pandas as pd
>>> pd.__version__
'1.4.0'
>>> dti = pd.date_range("2016-01-01", periods=3)
>>> ii = pd.IntervalIndex.from_breaks(dti)
>>> ser = pd.Series(ii)
>>> ser.unique()
<IntervalArray>
[(2016-01-01, 2016-01-02], (2016-01-02, 2016-01-03]]
Length: 2, dtype: interval[datetime64[ns], right]

With the PR:

>>> import pandas as pd
>>> pd.__version__
'1.5.0.dev0+401.g84b119faea'
>>> dti = pd.date_range("2016-01-01", periods=3)
>>> ii = pd.IntervalIndex.from_breaks(dti)
>>> ser = pd.Series(ii)
>>> ser.unique()
<IntervalArray>
[(2016-01-01, 2016-01-02], (2016-01-02, 2016-01-03]]
Length: 2, dtype: interval[datetime64[ns], right]

That's why we can remove the if test.

[jbrockmendel]

In your example, the test needs_i8_conversion(dtype) is False

This doesn't make sense to me. dtype should be np.dtype("M8[ns]"), for which needs_i8_conversion(dtype) is True.

For a starker example, if you pass a tz to the date_range, the ser.unique raises in this branch.

[Dr-Irv]

OK, the tz example shows that what I thought was dead code can't be removed. Will close this PR then.

[Dr-Irv]

closing per comment above.

[jbrockmendel]

Could use a test