python · markshannon · Apr 19, 2022 · Jan 23, 2022 · Jan 25, 2022 · Jan 25, 2022
diff --git a/Include/internal/pycore_floatobject.h b/Include/internal/pycore_floatobject.h
@@ -38,6 +38,9 @@ struct _Py_float_state {
 #endif
 };
 
+PyAPI_FUNC(void) _PyFloat_ExactDealloc(PyFloatObject *op);
+
+
 /* _PyFloat_{Pack,Unpack}{4,8}
  *
  * The struct and pickle (at least) modules need an efficient platform-

diff --git a/Include/internal/pycore_long.h b/Include/internal/pycore_long.h
@@ -67,6 +67,8 @@ PyAPI_FUNC(char*) _PyLong_FormatBytesWriter(
     int base,
     int alternate);
 
+PyAPI_FUNC(void) _PyLong_ExactDealloc(PyLongObject *op);
+
 #ifdef __cplusplus
 }
 #endif

diff --git a/Include/internal/pycore_unicodeobject.h b/Include/internal/pycore_unicodeobject.h
@@ -10,6 +10,7 @@ extern "C" {
 
 #include "pycore_fileutils.h"     // _Py_error_handler
 
+void _PyUnicode_ExactDealloc(PyUnicodeObject *op);
 
 /* runtime lifecycle */
 

diff --git a/Include/object.h b/Include/object.h
@@ -524,6 +524,37 @@ static inline void _Py_DECREF(
 #  define Py_DECREF(op) _Py_DECREF(_PyObject_CAST(op))
 #endif
 
+static inline void
+_Py_DECREF_SPECIALIZED(PyObject *op, const destructor destruct)
+{
+#ifdef Py_REF_DEBUG
+    _Py_RefTotal--;
+#endif
+    if (--op->ob_refcnt != 0) {
+        assert(op->ob_refcnt > 0);
+    }
+    else {
+#ifdef Py_TRACE_REFS
+        _Py_ForgetReference(op);
+#endif
+        destruct(op);
+    }
+}
+
+static inline void
+_Py_DECREF_IMMORTAL(PyObject *op)
+{
+#ifdef Py_REF_DEBUG
+    _Py_RefTotal--;
+#endif
+    op->ob_refcnt--;
+#ifdef Py_DEBUG
+    if (op->ob_refcnt <= 0) {
+        // Calls _Py_FatalRefcountError for None, True, and False
+        _Py_Dealloc(op);
+    }
+#endif
+}
 
 /* Safely decref `op` and set `op` to NULL, especially useful in tp_clear
  * and tp_dealloc implementations.
@@ -622,6 +653,7 @@ Don't forget to apply Py_INCREF() when returning this value!!!
 PyAPI_DATA(PyObject) _Py_NoneStruct; /* Don't use this directly */
 #define Py_None (&_Py_NoneStruct)
 
+
 // Test if an object is the None singleton, the same as "x is None" in Python.
 PyAPI_FUNC(int) Py_IsNone(PyObject *x);
 #define Py_IsNone(x) Py_Is((x), Py_None)

diff --git a/Misc/NEWS.d/next/Core and Builtins/2022-01-25-05-39-38.bpo-46509.ljrqrc.rst b/Misc/NEWS.d/next/Core and Builtins/2022-01-25-05-39-38.bpo-46509.ljrqrc.rst
@@ -0,0 +1 @@
+Add type-specialized versions of the ``Py_DECREF()``, and use them for ``float``, ``int``, ``str``, ``bool``, and ``None`` to avoid pointer-chasing at runtime where types are known at C compile time.
diff --git a/Modules/mathmodule.c b/Modules/mathmodule.c
@@ -60,7 +60,8 @@ raised for division by zero and mod by zero.
 #include "pycore_bitutils.h"      // _Py_bit_length()
 #include "pycore_call.h"          // _PyObject_CallNoArgs()
 #include "pycore_dtoa.h"          // _Py_dg_infinity()
-#include "pycore_long.h"          // _PyLong_GetZero()
+#include "pycore_long.h"          // _PyLong_GetZero(), _PyLong_ExactDealloc
+#include "pycore_floatobject.h"   // _PyFloat_ExactDealloc
 /* For DBL_EPSILON in _math.h */
 #include <float.h>
 /* For _Py_log1p with workarounds for buggy handling of zeros. */
@@ -3139,7 +3140,7 @@ math_prod_impl(PyObject *module, PyObject *iterable, PyObject *start)
         long i_result = PyLong_AsLongAndOverflow(result, &overflow);
         /* If this already overflowed, don't even enter the loop. */
         if (overflow == 0) {
-            Py_DECREF(result);
+            _Py_DECREF_SPECIALIZED(result, (destructor)_PyLong_ExactDealloc);
             result = NULL;
         }
         /* Loop over all the items in the iterable until we finish, we overflow
@@ -3158,7 +3159,8 @@ math_prod_impl(PyObject *module, PyObject *iterable, PyObject *start)
                 if (overflow == 0 && !_check_long_mult_overflow(i_result, b)) {
                     long x = i_result * b;
                     i_result = x;
-                    Py_DECREF(item);
+                    _Py_DECREF_SPECIALIZED(item,
+                                           (destructor)_PyLong_ExactDealloc);
                     continue;
                 }
             }
@@ -3187,7 +3189,7 @@ math_prod_impl(PyObject *module, PyObject *iterable, PyObject *start)
     */
     if (PyFloat_CheckExact(result)) {
         double f_result = PyFloat_AS_DOUBLE(result);
-        Py_DECREF(result);
+        _Py_DECREF_SPECIALIZED(result, (destructor)_PyFloat_ExactDealloc);
         result = NULL;
         while(result == NULL) {
             item = PyIter_Next(iter);
@@ -3200,7 +3202,8 @@ math_prod_impl(PyObject *module, PyObject *iterable, PyObject *start)
             }
             if (PyFloat_CheckExact(item)) {
                 f_result *= PyFloat_AS_DOUBLE(item);
-                Py_DECREF(item);
+                _Py_DECREF_SPECIALIZED(item,
+                                       (destructor)_PyFloat_ExactDealloc);
                 continue;
             }
             if (PyLong_CheckExact(item)) {

diff --git a/Objects/floatobject.c b/Objects/floatobject.c
@@ -238,23 +238,34 @@ PyFloat_FromString(PyObject *v)
     return result;
 }
 
+void
+_PyFloat_ExactDealloc(PyFloatObject *op)
+{
+    assert(PyFloat_CheckExact(op));
+#if PyFloat_MAXFREELIST > 0
+    struct _Py_float_state *state = get_float_state();
+#ifdef Py_DEBUG
+    // float_dealloc() must not be called after _PyFloat_Fini()
+    assert(state->numfree != -1);
+#endif
+    if (state->numfree >= PyFloat_MAXFREELIST)  {
+        PyObject_Free(op);
+        return;
+    }
+    state->numfree++;
+    Py_SET_TYPE(op, (PyTypeObject *)state->free_list);
+    state->free_list = op;
+#else
+    PyObject_Free(op);
+#endif
+}
+
 static void
 float_dealloc(PyFloatObject *op)
 {
 #if PyFloat_MAXFREELIST > 0
     if (PyFloat_CheckExact(op)) {
-        struct _Py_float_state *state = get_float_state();
-#ifdef Py_DEBUG
-        // float_dealloc() must not be called after _PyFloat_Fini()
-        assert(state->numfree != -1);
-#endif
-        if (state->numfree >= PyFloat_MAXFREELIST)  {
-            PyObject_Free(op);
-            return;
-        }
-        state->numfree++;
-        Py_SET_TYPE(op, (PyTypeObject *)state->free_list);
-        state->free_list = op;
+        _PyFloat_ExactDealloc(op);
     }
     else
 #endif

diff --git a/Objects/longobject.c b/Objects/longobject.c
@@ -39,7 +39,16 @@ medium_value(PyLongObject *x)
 #define IS_SMALL_INT(ival) (-_PY_NSMALLNEGINTS <= (ival) && (ival) < _PY_NSMALLPOSINTS)
 #define IS_SMALL_UINT(ival) ((ival) < _PY_NSMALLPOSINTS)
 
-static inline int is_medium_int(stwodigits x)
+static inline void
+_Py_DECREF_INT(PyLongObject *op)
+{
+    assert(PyLong_CheckExact(op));
+    _Py_DECREF_SPECIALIZED((PyObject *)op,
+                           (destructor)_PyLong_ExactDealloc);
+}
+
+static inline int
+is_medium_int(stwodigits x)
 {
     /* Take care that we are comparing unsigned values. */
     twodigits x_plus_mask = ((twodigits)x) + PyLong_MASK;
@@ -61,7 +70,7 @@ maybe_small_long(PyLongObject *v)
     if (v && IS_MEDIUM_VALUE(v)) {
         stwodigits ival = medium_value(v);
         if (IS_SMALL_INT(ival)) {
-            Py_DECREF(v);
+            _Py_DECREF_INT(v);
             return (PyLongObject *)get_small_int((sdigit)ival);
         }
     }
@@ -1824,7 +1833,7 @@ long_to_decimal_string_internal(PyObject *aa,
 #undef WRITE_DIGITS
 #undef WRITE_UNICODE_DIGITS
 
-    Py_DECREF(scratch);
+    _Py_DECREF_INT(scratch);
     if (writer) {
         writer->pos += strlen;
     }
@@ -3479,15 +3488,15 @@ k_mul(PyLongObject *a, PyLongObject *b)
      */
     i = Py_SIZE(ret) - shift;  /* # digits after shift */
     (void)v_isub(ret->ob_digit + shift, i, t2->ob_digit, Py_SIZE(t2));
-    Py_DECREF(t2);
+    _Py_DECREF_INT(t2);
 
     (void)v_isub(ret->ob_digit + shift, i, t1->ob_digit, Py_SIZE(t1));
-    Py_DECREF(t1);
+    _Py_DECREF_INT(t1);
 
     /* 6. t3 <- (ah+al)(bh+bl), and add into result. */
     if ((t1 = x_add(ah, al)) == NULL) goto fail;
-    Py_DECREF(ah);
-    Py_DECREF(al);
+    _Py_DECREF_INT(ah);
+    _Py_DECREF_INT(al);
     ah = al = NULL;
 
     if (a == b) {
@@ -3498,21 +3507,21 @@ k_mul(PyLongObject *a, PyLongObject *b)
         Py_DECREF(t1);
         goto fail;
     }
-    Py_DECREF(bh);
-    Py_DECREF(bl);
+    _Py_DECREF_INT(bh);
+    _Py_DECREF_INT(bl);
     bh = bl = NULL;
 
     t3 = k_mul(t1, t2);
-    Py_DECREF(t1);
-    Py_DECREF(t2);
+    _Py_DECREF_INT(t1);
+    _Py_DECREF_INT(t2);
     if (t3 == NULL) goto fail;
     assert(Py_SIZE(t3) >= 0);
 
     /* Add t3.  It's not obvious why we can't run out of room here.
      * See the (*) comment after this function.
      */
     (void)v_iadd(ret->ob_digit + shift, i, t3->ob_digit, Py_SIZE(t3));
-    Py_DECREF(t3);
+    _Py_DECREF_INT(t3);
 
     return long_normalize(ret);
 
@@ -3617,13 +3626,13 @@ k_lopsided_mul(PyLongObject *a, PyLongObject *b)
         /* Add into result. */
         (void)v_iadd(ret->ob_digit + nbdone, Py_SIZE(ret) - nbdone,
                      product->ob_digit, Py_SIZE(product));
-        Py_DECREF(product);
+        _Py_DECREF_INT(product);
 
         bsize -= nbtouse;
         nbdone += nbtouse;
     }
 
-    Py_DECREF(bslice);
+    _Py_DECREF_INT(bslice);
     return long_normalize(ret);
 
   fail:
@@ -5743,6 +5752,13 @@ int_from_bytes_impl(PyTypeObject *type, PyObject *bytes_obj,
     return long_obj;
 }
 
+void
+_PyLong_ExactDealloc(PyLongObject *op)
+{
+    assert(PyLong_CheckExact(op));
+    PyObject_Del(op);
+}
+
 static PyObject *
 long_long_meth(PyObject *self, PyObject *Py_UNUSED(ignored))
 {

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
@@ -15463,6 +15463,13 @@ unicode_subtype_new(PyTypeObject *type, PyObject *unicode)
     return NULL;
 }
 
+void
+_PyUnicode_ExactDealloc(PyUnicodeObject *op)
+{
+    assert(PyUnicode_CheckExact(op));
+    unicode_dealloc((PyObject *)op);
+}
+
 PyDoc_STRVAR(unicode_doc,
 "str(object='') -> str\n\
 str(bytes_or_buffer[, encoding[, errors]]) -> str\n\

@@ -2520,7 +2520,7 @@ builtin_sum_impl(PyObject *module, PyObject *iterable, PyObject *start)
             }
             if (PyFloat_CheckExact(item)) {
                 f_result += PyFloat_AS_DOUBLE(item);
-                Py_DECREF(item);
+                _Py_DECREF_SPECIALIZED(item, (destructor)_PyFloat_ExactDealloc);
                 continue;
             }
             if (PyLong_Check(item)) {
-Original file line number
+Diff line change
@@ Expand Up / @@ -38,6 +38,9 @@ struct _Py_float_state { @@
     #endif
     };
+    PyAPI_FUNC(void) _PyFloat_ExactDealloc(PyFloatObject *op);
     /* _PyFloat_{Pack,Unpack}{4,8}
      *
      * The struct and pickle (at least) modules need an efficient platform-
@@ Expand Down @@
Original file line number	Diff line number	Diff line change
Expand Up		@@ -10,6 +10,7 @@ extern "C" {

		#include "pycore_fileutils.h" // _Py_error_handler

		void _PyUnicode_ExactDealloc(PyUnicodeObject *op);

		/* runtime lifecycle */

Expand Down
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		Add type-specialized versions of the ``Py_DECREF()``, and use them for ``float``, ``int``, ``str``, ``bool``, and ``None`` to avoid pointer-chasing at runtime where types are known at C compile time.