PERF: PeriodDtype hash and eq (pandas-dev#52336)

* PERF: PeriodDtype __hash__/__eq__

* de-duplicate fastpath

* Revert accidental

* mypy fixup

* troubleshoot min-versions build

Author: jbrockmendel

pandas/_libs/tslibs/dtypes.pxd

@@ -102,5 +102,6 @@ cdef enum PeriodDtypeCode:
 cdef class PeriodDtypeBase:
     cdef readonly:
         PeriodDtypeCode _dtype_code
+        int64_t _n
 
     cpdef int _get_to_timestamp_base(self)

pandas/_libs/tslibs/dtypes.pyi

@@ -14,16 +14,18 @@ def abbrev_to_npy_unit(abbrev: str) -> int: ...
 
 class PeriodDtypeBase:
     _dtype_code: int  # PeriodDtypeCode
+    _n: int
 
     # actually __cinit__
-    def __new__(cls, code: int): ...
+    def __new__(cls, code: int, n: int): ...
     @property
     def _freq_group_code(self) -> int: ...
     @property
     def _resolution_obj(self) -> Resolution: ...
     def _get_to_timestamp_base(self) -> int: ...
     @property
     def _freqstr(self) -> str: ...
+    def __hash__(self) -> int: ...
 
 class FreqGroup(Enum):
     FR_ANN: int

pandas/_libs/tslibs/dtypes.pyx

@@ -18,17 +18,22 @@ cdef class PeriodDtypeBase:
     """
     # cdef readonly:
     #    PeriodDtypeCode _dtype_code
+    #    int64_t _n
 
-    def __cinit__(self, PeriodDtypeCode code):
+    def __cinit__(self, PeriodDtypeCode code, int64_t n):
         self._dtype_code = code
+        self._n = n
 
     def __eq__(self, other):
         if not isinstance(other, PeriodDtypeBase):
             return False
         if not isinstance(self, PeriodDtypeBase):
             # cython semantics, this is a reversed op
             return False
-        return self._dtype_code == other._dtype_code
+        return self._dtype_code == other._dtype_code and self._n == other._n
+
+    def __hash__(self) -> int:
+        return hash((self._n, self._dtype_code))
 
     @property
     def _freq_group_code(self) -> int:

pandas/_libs/tslibs/period.pyx

@@ -1671,7 +1671,7 @@ cdef class _Period(PeriodMixin):
         # Note: this is more performant than PeriodDtype.from_date_offset(freq)
         #  because from_date_offset cannot be made a cdef method (until cython
         #  supported cdef classmethods)
-        self._dtype = PeriodDtypeBase(freq._period_dtype_code)
+        self._dtype = PeriodDtypeBase(freq._period_dtype_code, freq.n)
 
     @classmethod
     def _maybe_convert_freq(cls, object freq) -> BaseOffset:
@@ -1686,7 +1686,7 @@ cdef class _Period(PeriodMixin):
         """
         if isinstance(freq, int):
             # We already have a dtype code
-            dtype = PeriodDtypeBase(freq)
+            dtype = PeriodDtypeBase(freq, 1)
             freq = dtype._freqstr
 
         freq = to_offset(freq)

pandas/core/dtypes/dtypes.py

@@ -862,6 +862,7 @@ class PeriodDtype(PeriodDtypeBase, PandasExtensionDtype):
     _metadata = ("freq",)
     _match = re.compile(r"(P|p)eriod\[(?P<freq>.+)\]")
     _cache_dtypes: dict[str_type, PandasExtensionDtype] = {}
+    __hash__ = PeriodDtypeBase.__hash__
 
     def __new__(cls, freq):
         """
@@ -879,7 +880,7 @@ def __new__(cls, freq):
             return cls._cache_dtypes[freq.freqstr]
         except KeyError:
             dtype_code = freq._period_dtype_code
-            u = PeriodDtypeBase.__new__(cls, dtype_code)
+            u = PeriodDtypeBase.__new__(cls, dtype_code, freq.n)
             u._freq = freq
             cls._cache_dtypes[freq.freqstr] = u
             return u
@@ -945,22 +946,11 @@ def name(self) -> str_type:
     def na_value(self) -> NaTType:
         return NaT
 
-    def __hash__(self) -> int:
-        # make myself hashable
-        return hash(str(self))
-
     def __eq__(self, other: Any) -> bool:
         if isinstance(other, str):
             return other in [self.name, self.name.title()]
 
-        elif isinstance(other, PeriodDtype):
-            # For freqs that can be held by a PeriodDtype, this check is
-            # equivalent to (and much faster than) self.freq == other.freq
-            sfreq = self._freq
-            ofreq = other._freq
-            return sfreq.n == ofreq.n and self._dtype_code == other._dtype_code
-
-        return False
+        return super().__eq__(other)
 
     def __ne__(self, other: Any) -> bool:
         return not self.__eq__(other)

pandas/core/indexes/period.py

@@ -25,7 +25,6 @@
 )
 
 from pandas.core.dtypes.common import is_integer
-from pandas.core.dtypes.dtypes import PeriodDtype
 from pandas.core.dtypes.generic import ABCSeries
 from pandas.core.dtypes.missing import is_valid_na_for_dtype
 
@@ -52,6 +51,9 @@
         Self,
         npt,
     )
+
+    from pandas.core.dtypes.dtypes import PeriodDtype
+
 _index_doc_kwargs = dict(ibase._index_doc_kwargs)
 _index_doc_kwargs.update({"target_klass": "PeriodIndex or list of Periods"})
 _shared_doc_kwargs = {
@@ -314,20 +316,7 @@ def _is_comparable_dtype(self, dtype: DtypeObj) -> bool:
         """
         Can we compare values of the given dtype to our own?
         """
-        if not isinstance(dtype, PeriodDtype):
-            return False
-        # For the subset of DateOffsets that can be a dtype.freq, it
-        #  suffices (and is much faster) to compare the dtype_code rather than
-        #  the freq itself.
-        # See also: PeriodDtype.__eq__
-        freq = dtype.freq
-        own_freq = self.freq
-        return (
-            freq._period_dtype_code
-            # error: "BaseOffset" has no attribute "_period_dtype_code"
-            == own_freq._period_dtype_code  # type: ignore[attr-defined]
-            and freq.n == own_freq.n
-        )
+        return self.dtype == dtype
 
     # ------------------------------------------------------------------------
     # Index Methods