switch to gauss algo

Author: mcgeestocks

pandas/_libs/tslibs/ccalendar.pyx

@@ -2,7 +2,7 @@
 """
 Cython implementations of functions resembling the stdlib calendar module
 """
-
+from libc.stdlib cimport abs as c_abs
 cimport cython
 from numpy cimport (
     int32_t,
@@ -20,6 +20,7 @@ cdef int32_t* days_per_month_array = [
     31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31]
 
 cdef int* sakamoto_arr = [0, 3, 2, 5, 0, 3, 5, 1, 4, 6, 2, 4]
+cdef int* em = [0 ,0 ,31 ,59 ,90 ,120 ,151 ,181 ,212 ,243 ,273 ,304 ,334]
 
 # The first 13 entries give the month days elapsed as of the first of month N
 # (or the total number of days in the year for N=13) in non-leap years.
@@ -73,14 +74,23 @@ cpdef int32_t get_days_in_month(int year, Py_ssize_t month) noexcept nogil:
     """
     return days_per_month_array[12 * is_leapyear(year) + month - 1]
 
+@cython.wraparound(False)
+@cython.boundscheck(False)
+@cython.cdivision(True)
+cdef long quot(long a , long b) noexcept nogil:
+    cdef long x
+    x = a/b 
+    if ( a < 0 ):
+        x -= ( a % b != 0 )
+    return x
 
 @cython.wraparound(False)
 @cython.boundscheck(False)
-@cython.cdivision
+@cython.cdivision(True)
 cdef int dayofweek(int y, int m, int d) noexcept nogil:
     """
     Find the day of week for the date described by the Y/M/D triple y, m, d
-    using Sakamoto's method, from wikipedia.
+    using Gauss' method, from wikipedia.
 
     0 represents Monday.  See [1]_.
 
@@ -103,16 +113,26 @@ cdef int dayofweek(int y, int m, int d) noexcept nogil:
     [1] https://docs.python.org/3/library/calendar.html#calendar.weekday
 
     [2] https://en.wikipedia.org/wiki/\
-    Determination_of_the_day_of_the_week#Sakamoto.27s_methods
+    Determination_of_the_day_of_the_week#Gauss's_algorithm
     """
-    cdef:
-        int day
-
-    y -= m < 3
-    day = (y + y / 4 - y / 100 + y / 400 + sakamoto_arr[m - 1] + d) % 7
-    # convert to python day
-    return (day + 6) % 7
-
+      
+    cdef: 
+        long c 
+        int g
+        int f
+        int e
+    
+    if (m < 3):
+        y -= 1
+        
+    c = quot(y, 100)    
+    g = y - quot(y, 100) * 100    
+    f = 5 * (c - quot(c, 4) * 4)
+    e = em[m]    
+    
+    if (m > 2):
+        e -= 1;
+    return (-1 + d + e + f + g + g/4) % 7
 
 cdef bint is_leapyear(int64_t year) noexcept nogil:
     """

pandas/_libs/tslibs/timestamps.pyx

@@ -2463,7 +2463,7 @@ default 'raise'
         Return the day of the week represented by the date.
 
         Monday == 0 ... Sunday == 6.
-        """
+        """        
         # same as super().weekday(), but that breaks because of how
         #  we have overridden year, see note in create_timestamp_from_ts
         return ccalendar.dayofweek(self.year, self.month, self.day)

pandas/tests/scalar/timestamp/test_timestamp.py

@@ -223,6 +223,13 @@ def test_resolution(self):
         assert dt.as_unit("ms").resolution == Timedelta(milliseconds=1)
         assert dt.as_unit("s").resolution == Timedelta(seconds=1)
 
+    @pytest.mark.parametrize("date_string,expected", [("0000-2-29", 1), ("0000-3-1", 2), ("1582-10-14", 3), ("1582-10-15", 4)])
+    def test_dow_historic(self, date_string, expected):
+        # GH 53738
+        dt = Timestamp(date_string)
+        dow = dt.weekday()        
+        assert dow == expected
+        
 
 class TestTimestamp:
     def test_default_to_stdlib_utc(self):