REF: tighter typing in parsing.pyx (pandas-dev#50851)

Author: jbrockmendel

pandas/_libs/tslib.pyx

@@ -273,7 +273,7 @@ def array_with_unit_to_datetime(
         bint is_coerce = errors=="coerce"
         bint is_raise = errors=="raise"
         ndarray[int64_t] iresult
-        object tz = None
+        tzinfo tz = None
         float fval
 
     assert is_ignore or is_coerce or is_raise
@@ -346,7 +346,7 @@ cdef _array_with_unit_to_datetime_object_fallback(ndarray[object] values, str un
     cdef:
         Py_ssize_t i, n = len(values)
         ndarray[object] oresult
-        object tz = None
+        tzinfo tz = None
 
     # TODO: fix subtle differences between this and no-unit code
     oresult = cnp.PyArray_EMPTY(values.ndim, values.shape, cnp.NPY_OBJECT, 0)

pandas/_libs/tslibs/dtypes.pxd

@@ -11,6 +11,7 @@ cpdef int64_t periods_per_second(NPY_DATETIMEUNIT reso) except? -1
 cpdef NPY_DATETIMEUNIT get_supported_reso(NPY_DATETIMEUNIT reso)
 
 cdef dict attrname_to_abbrevs
+cdef dict npy_unit_to_attrname
 
 cdef enum c_FreqGroup:
     # Mirrors FreqGroup in the .pyx file

pandas/_libs/tslibs/dtypes.pyx

@@ -423,3 +423,15 @@ cdef dict _reso_str_map = {
 }
 
 cdef dict _str_reso_map = {v: k for k, v in _reso_str_map.items()}
+
+cdef dict npy_unit_to_attrname = {
+    NPY_DATETIMEUNIT.NPY_FR_Y: "year",
+    NPY_DATETIMEUNIT.NPY_FR_M: "month",
+    NPY_DATETIMEUNIT.NPY_FR_D: "day",
+    NPY_DATETIMEUNIT.NPY_FR_h: "hour",
+    NPY_DATETIMEUNIT.NPY_FR_m: "minute",
+    NPY_DATETIMEUNIT.NPY_FR_s: "second",
+    NPY_DATETIMEUNIT.NPY_FR_ms: "millisecond",
+    NPY_DATETIMEUNIT.NPY_FR_us: "microsecond",
+    NPY_DATETIMEUNIT.NPY_FR_ns: "nanosecond",
+}

pandas/_libs/tslibs/parsing.pyx

@@ -12,7 +12,10 @@ from cpython.datetime cimport (
     datetime,
     datetime_new,
     import_datetime,
+    timedelta,
+    tzinfo,
 )
+from datetime import timezone
 from cpython.object cimport PyObject_Str
 from cython cimport Py_ssize_t
 from libc.string cimport strchr
@@ -49,6 +52,7 @@ from dateutil.tz import (
 from pandas._config import get_option
 
 from pandas._libs.tslibs.ccalendar cimport c_MONTH_NUMBERS
+from pandas._libs.tslibs.dtypes cimport npy_unit_to_attrname
 from pandas._libs.tslibs.nattype cimport (
     c_NaT as NaT,
     c_nat_strings as nat_strings,
@@ -120,7 +124,9 @@ cdef int _parse_4digit(const char* s):
     return result
 
 
-cdef object _parse_delimited_date(str date_string, bint dayfirst):
+cdef datetime _parse_delimited_date(
+    str date_string, bint dayfirst, NPY_DATETIMEUNIT* creso
+):
     """
     Parse special cases of dates: MM/DD/YYYY, DD/MM/YYYY, MM/YYYY.
 
@@ -138,12 +144,12 @@ cdef object _parse_delimited_date(str date_string, bint dayfirst):
     ----------
     date_string : str
     dayfirst : bool
+    creso : NPY_DATETIMEUNIT*
+        For specifying identified resolution.
 
     Returns:
     --------
     datetime or None
-    str or None
-        Describing resolution of the parsed string.
     """
     cdef:
         const char* buf
@@ -157,53 +163,53 @@ cdef object _parse_delimited_date(str date_string, bint dayfirst):
         month = _parse_2digit(buf)
         day = _parse_2digit(buf + 3)
         year = _parse_4digit(buf + 6)
-        reso = "day"
+        creso[0] = NPY_DATETIMEUNIT.NPY_FR_D
         can_swap = 1
     elif length == 9 and _is_delimiter(buf[1]) and _is_delimiter(buf[4]):
         # parsing M?DD?YYYY and D?MM?YYYY dates
         month = _parse_1digit(buf)
         day = _parse_2digit(buf + 2)
         year = _parse_4digit(buf + 5)
-        reso = "day"
+        creso[0] = NPY_DATETIMEUNIT.NPY_FR_D
         can_swap = 1
     elif length == 9 and _is_delimiter(buf[2]) and _is_delimiter(buf[4]):
         # parsing MM?D?YYYY and DD?M?YYYY dates
         month = _parse_2digit(buf)
         day = _parse_1digit(buf + 3)
         year = _parse_4digit(buf + 5)
-        reso = "day"
+        creso[0] = NPY_DATETIMEUNIT.NPY_FR_D
         can_swap = 1
     elif length == 8 and _is_delimiter(buf[1]) and _is_delimiter(buf[3]):
         # parsing M?D?YYYY and D?M?YYYY dates
         month = _parse_1digit(buf)
         day = _parse_1digit(buf + 2)
         year = _parse_4digit(buf + 4)
-        reso = "day"
+        creso[0] = NPY_DATETIMEUNIT.NPY_FR_D
         can_swap = 1
     elif length == 7 and _is_delimiter(buf[2]):
         # parsing MM?YYYY dates
         if buf[2] == b".":
             # we cannot reliably tell whether e.g. 10.2010 is a float
             # or a date, thus we refuse to parse it here
-            return None, None
+            return None
         month = _parse_2digit(buf)
         year = _parse_4digit(buf + 3)
-        reso = "month"
+        creso[0] = NPY_DATETIMEUNIT.NPY_FR_M
     else:
-        return None, None
+        return None
 
     if month < 0 or day < 0 or year < 1000:
         # some part is not an integer, so
         # date_string can't be converted to date, above format
-        return None, None
+        return None
 
     if 1 <= month <= MAX_DAYS_IN_MONTH and 1 <= day <= MAX_DAYS_IN_MONTH \
             and (month <= MAX_MONTH or day <= MAX_MONTH):
         if (month > MAX_MONTH or (day <= MAX_MONTH and dayfirst)) and can_swap:
             day, month = month, day
         # In Python <= 3.6.0 there is no range checking for invalid dates
         # in C api, thus we call faster C version for 3.6.1 or newer
-        return datetime_new(year, month, day, 0, 0, 0, 0, None), reso
+        return datetime_new(year, month, day, 0, 0, 0, 0, None)
 
     raise DateParseError(f"Invalid date specified ({month}/{day})")
 
@@ -264,6 +270,7 @@ def parse_datetime_string(
 
     cdef:
         datetime dt
+        NPY_DATETIMEUNIT creso
 
     if not _does_string_look_like_datetime(date_string):
         raise ValueError(f'Given date string "{date_string}" not likely a datetime')
@@ -274,7 +281,7 @@ def parse_datetime_string(
                       yearfirst=yearfirst)
         return dt
 
-    dt, _ = _parse_delimited_date(date_string, dayfirst)
+    dt = _parse_delimited_date(date_string, dayfirst, &creso)
     if dt is not None:
         return dt
 
@@ -351,18 +358,19 @@ def parse_datetime_string_with_reso(
         bint string_to_dts_failed
         npy_datetimestruct dts
         NPY_DATETIMEUNIT out_bestunit
-        int out_local
+        int out_local = 0
         int out_tzoffset
+        tzinfo tz
 
     if not _does_string_look_like_datetime(date_string):
         raise ValueError(f'Given date string "{date_string}" not likely a datetime')
 
-    parsed, reso = _parse_delimited_date(date_string, dayfirst)
+    parsed = _parse_delimited_date(date_string, dayfirst, &out_bestunit)
     if parsed is not None:
+        reso = npy_unit_to_attrname[out_bestunit]
         return parsed, reso
 
     # Try iso8601 first, as it handles nanoseconds
-    # TODO: does this render some/all of parse_delimited_date redundant?
     string_to_dts_failed = string_to_dts(
         date_string, &dts, &out_bestunit, &out_local,
         &out_tzoffset, False
@@ -372,31 +380,25 @@ def parse_datetime_string_with_reso(
                            NPY_DATETIMEUNIT.NPY_FR_ps,
                            NPY_DATETIMEUNIT.NPY_FR_fs,
                            NPY_DATETIMEUNIT.NPY_FR_as}
-        if out_bestunit in timestamp_units or out_local:
-            # TODO: the not-out_local case we could do without Timestamp;
-            #  avoid circular import
+        if out_bestunit in timestamp_units:
+            # TODO: avoid circular import
             from pandas import Timestamp
             parsed = Timestamp(date_string)
         else:
-            parsed = datetime(
-                dts.year, dts.month, dts.day, dts.hour, dts.min, dts.sec, dts.us
+            if out_local:
+                tz = timezone(timedelta(minutes=out_tzoffset))
+            else:
+                tz = None
+            parsed = datetime_new(
+                dts.year, dts.month, dts.day, dts.hour, dts.min, dts.sec, dts.us, tz
             )
         # Match Timestamp and drop picoseconds, femtoseconds, attoseconds
         # The new resolution will just be nano
         # GH 50417
         if out_bestunit in timestamp_units:
             out_bestunit = NPY_DATETIMEUNIT.NPY_FR_ns
-        reso = {
-            NPY_DATETIMEUNIT.NPY_FR_Y: "year",
-            NPY_DATETIMEUNIT.NPY_FR_M: "month",
-            NPY_DATETIMEUNIT.NPY_FR_D: "day",
-            NPY_DATETIMEUNIT.NPY_FR_h: "hour",
-            NPY_DATETIMEUNIT.NPY_FR_m: "minute",
-            NPY_DATETIMEUNIT.NPY_FR_s: "second",
-            NPY_DATETIMEUNIT.NPY_FR_ms: "millisecond",
-            NPY_DATETIMEUNIT.NPY_FR_us: "microsecond",
-            NPY_DATETIMEUNIT.NPY_FR_ns: "nanosecond",
-            }[out_bestunit]
+
+        reso = npy_unit_to_attrname[out_bestunit]
         return parsed, reso
 
     try: