gh-145685: per-type method cache implementation

Include/cpython/object.h

@@ -246,6 +246,8 @@ struct _typeobject {
       * This function must escape to any code that can result in
       * the GC being run, such as Py_DECREF.  */
     _Py_iteritemfunc _tp_iteritem;
+
+    void *_tp_cache;
 };
 
 #define _Py_ATTR_CACHE_UNUSED (30000)  // (see tp_versions_used)

Include/internal/pycore_interp_structs.h

@@ -548,23 +548,6 @@ struct _types_runtime_state {
 };
 
 
-// Type attribute lookup cache: speed up attribute and method lookups,
-// see _PyType_Lookup().
-struct type_cache_entry {
-    unsigned int version;  // initialized from type->tp_version_tag
-#ifdef Py_GIL_DISABLED
-   _PySeqLock sequence;
-#endif
-    PyObject *name;        // reference to exactly a str or None
-    PyObject *value;       // borrowed reference or NULL
-};
-
-#define MCACHE_SIZE_EXP 12
-
-struct type_cache {
-    struct type_cache_entry hashtable[1 << MCACHE_SIZE_EXP];
-};
-
 typedef struct {
     PyTypeObject *type;
     int isbuiltin;
@@ -579,6 +562,10 @@ typedef struct {
        are also some diagnostic uses for the list of weakrefs,
        so we still keep it. */
     PyObject *tp_weaklist;
+    /* Per-interpreter attribute lookup cache (struct type_cache *).
+       For static builtin types the cache must be per-interpreter
+       because tp_dict and the values it stores are per-interpreter. */
+    void *_tp_cache;
 } managed_static_type_state;
 
 #define TYPE_VERSION_CACHE_SIZE (1<<12)  /* Must be a power of 2 */
@@ -589,8 +576,6 @@ struct types_state {
        where all those lower numbers are used for core static types. */
     unsigned int next_version_tag;
 
-    struct type_cache type_cache;
-
     /* Every static builtin type is initialized for each interpreter
        during its own initialization, including for the main interpreter
        during global runtime initialization.  This is done by calling

Include/internal/pycore_object.h

@@ -291,8 +291,6 @@ _PyType_HasFeature(PyTypeObject *type, unsigned long feature) {
     return ((type->tp_flags) & feature) != 0;
 }
 
-extern void _PyType_InitCache(PyInterpreterState *interp);
-
 extern PyStatus _PyObject_InitState(PyInterpreterState *interp);
 extern void _PyObject_FiniState(PyInterpreterState *interp);
 extern bool _PyRefchain_IsTraced(PyInterpreterState *interp, PyObject *obj);

Include/internal/pycore_typecache.h

@@ -0,0 +1,44 @@
+#ifndef PY_INTERNAL_TYPECACHE_H
+#define PY_INTERNAL_TYPECACHE_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+#include "pycore_stackref.h"
+
+
+#define _Py_TYPECACHE_MINSIZE 8
+
+struct type_cache_entry {
+    PyObject *name;
+    PyObject *value;
+};
+
+struct type_cache {
+    uint32_t mask;
+    uint32_t version_tag;
+    uint32_t available;
+    uint32_t used;
+    struct type_cache_entry hashtable[1];
+};
+
+struct _PyTypeCacheLookupResult {
+    _PyStackRef value;
+    int cache_hit;
+    uint32_t version_tag;
+};
+
+
+extern void _PyTypeCache_InitType(PyTypeObject *type);
+extern void _PyTypeCache_Insert(PyTypeObject *type, PyObject *name, PyObject *value);
+extern struct _PyTypeCacheLookupResult _PyTypeCache_Lookup(PyTypeObject *type, PyObject *name);
+extern void _PyTypeCache_Invalidate(PyTypeObject *type);
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* PY_INTERNAL_TYPECACHE_H */

Lib/test/test_free_threading/test_type.py

@@ -84,6 +84,24 @@ def reader_func():
 
         self.run_one(writer_func, reader_func)
 
+    def test_attr_cache_mortal(self):
+        class C:
+            x = object()
+
+        class D(C):
+            pass
+
+        def writer_func():
+            for _ in range(3000):
+                C.x = object()
+
+        def reader_func():
+            for _ in range(3000):
+                C.x
+                D.x
+
+        self.run_one(writer_func, reader_func)
+
     def test___class___modification(self):
         loops = 200
 

Lib/test/test_sys.py

@@ -1788,7 +1788,7 @@ def delx(self): del self.__x
         check((1,2,3), vsize('') + self.P + 3*self.P)
         # type
         # static type: PyTypeObject
-        fmt = 'P2nPI13Pl4Pn9Pn12PI2Pc'
+        fmt = 'P2nPI13Pl4Pn9Pn12PI2PcP'
         s = vsize(fmt)
         check(int, s)
         typeid = 'n' if support.Py_GIL_DISABLED else ''

Makefile.pre.in

@@ -506,6 +506,7 @@ PYTHON_OBJS=	\
 		Python/thread.o \
 		Python/traceback.o \
 		Python/tracemalloc.o \
+		Python/typecache.o \
 		Python/uniqueid.o \
 		Python/getopt.o \
 		Python/pystrcmp.o \
@@ -1411,6 +1412,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_tracemalloc.h \
 		$(srcdir)/Include/internal/pycore_tstate.h \
 		$(srcdir)/Include/internal/pycore_tuple.h \
+		$(srcdir)/Include/internal/pycore_typecache.h \
 		$(srcdir)/Include/internal/pycore_typedefs.h \
 		$(srcdir)/Include/internal/pycore_typeobject.h \
 		$(srcdir)/Include/internal/pycore_typevarobject.h \