Refactor DatetimeArray._generate_range

[TomAugspurger]

Currently, DatetimeArray._generate_range calls DatetimeArray._simple_new 4 times, 3 times with i8 values and ones with M8[ns] values.

diff --git a/pandas/core/arrays/datetimes.py b/pandas/core/arrays/datetimes.py
index 8b0565a36..a7f8e303a 100644
--- a/pandas/core/arrays/datetimes.py
+++ b/pandas/core/arrays/datetimes.py
@@ -213,12 +213,16 @@ class DatetimeArrayMixin(dtl.DatetimeLikeArrayMixin,
     _dtype = None  # type: Union[np.dtype, DatetimeTZDtype]
     _freq = None
 
+    i = 0
+
     @classmethod
     def _simple_new(cls, values, freq=None, tz=None):
         """
         we require the we have a dtype compat for the values
         if we are passed a non-dtype compat, then coerce using the constructor
         """
+        cls.i += 1
+        print(f"DTA._simple_new: {cls.i}")
         assert isinstance(values, np.ndarray), type(values)
         if values.dtype == 'i8':
             # for compat with datetime/timedelta/period shared methods,
diff --git a/pandas/core/indexes/datetimes.py b/pandas/core/indexes/datetimes.py
index 690a3db28..ab08bbf6f 100644
--- a/pandas/core/indexes/datetimes.py
+++ b/pandas/core/indexes/datetimes.py
@@ -273,6 +273,7 @@ class DatetimeIndex(DatetimeIndexOpsMixin, Int64Index, DatetimeDelegateMixin):
                 dayfirst=False, yearfirst=False, dtype=None,
                 copy=False, name=None, verify_integrity=None):
 
+        print("hi")
         if verify_integrity is not None:
             warnings.warn("The 'verify_integrity' argument is deprecated, "
                           "will be removed in a future version.",

In [2]: idx = pd.date_range('2014-01-02', '2014-04-30', freq='M', tz='UTC')
DTA._simple_new: 1
DTA._simple_new: 2
DTA._simple_new: 3
DTA._simple_new: 4

---

I'm not familiar with this code, but I would naively hope for a function that

1. Extracts there correct freq / tz from all the arguments (start, end, etc.)
2. Generates the correct i8 values for start, end, tz
3. Wraps those i8 values in a DatetimeArray._simple_new at the end.

I'm investigating if this can be done.

I'm not sure if this applies to timedelta as well.

[jbrockmendel]

Is this actually a perf issue?

[TomAugspurger]

I'd say so, especially if we implement frequency validation.

[jbrockmendel]

I'd say so, especially if we implement frequency validation.

The OP refers to _simple_new. I know in the new status quo simple_new just calls __init__, but if it behaved like the old simple_new it wouldn't need to do validation.

None of which affects the main point that range generation can be improved by a refactor.