GH-98522: Add version number to code objects. (GH-98525)

* Add version number to code object for better versioning of functions.

* Improves specialization for closures and list comprehensions.

Author: markshannon

Include/cpython/code.h

@@ -87,6 +87,7 @@ typedef struct {
     int co_nplaincellvars;        /* number of non-arg cell variables */       \
     int co_ncellvars;             /* total number of cell variables */         \
     int co_nfreevars;             /* number of free variables */               \
+    uint32_t co_version;          /* version number */                         \
                                                                                \
     PyObject *co_localsplusnames; /* tuple mapping offsets to names */         \
     PyObject *co_localspluskinds; /* Bytes mapping to local kinds (one byte    \

Include/internal/pycore_code.h

@@ -474,6 +474,8 @@ typedef struct _PyShimCodeDef {
 extern PyCodeObject *
 _Py_MakeShimCode(const _PyShimCodeDef *code);
 
+extern uint32_t _Py_next_func_version;
+
 
 #ifdef __cplusplus
 }

Misc/NEWS.d/next/Core and Builtins/2022-10-21-16-10-39.gh-issue-98522.s_SixG.rst

@@ -0,0 +1,3 @@
+Add an internal version number to code objects, to give better versioning of
+inner functions and comprehensions, and thus better specialization of those
+functions. This change is invisible to both Python and C extensions.

Objects/codeobject.c

@@ -11,7 +11,6 @@
 #include "pycore_tuple.h"         // _PyTuple_ITEMS()
 #include "clinic/codeobject.c.h"
 
-
 static void
 notify_code_watchers(PyCodeEvent event, PyCodeObject *co)
 {
@@ -398,7 +397,10 @@ init_code(PyCodeObject *co, struct _PyCodeConstructor *con)
     co->co_nplaincellvars = nplaincellvars;
     co->co_ncellvars = ncellvars;
     co->co_nfreevars = nfreevars;
-
+    co->co_version = _Py_next_func_version;
+    if (_Py_next_func_version != 0) {
+        _Py_next_func_version++;
+    }
     /* not set */
     co->co_weakreflist = NULL;
     co->co_extra = NULL;

Objects/funcobject.c

@@ -3,7 +3,7 @@
 
 #include "Python.h"
 #include "pycore_ceval.h"         // _PyEval_BuiltinsFromGlobals()
-#include "pycore_function.h"      // FUNC_MAX_WATCHERS
+#include "pycore_code.h"          // _Py_next_func_version
 #include "pycore_object.h"        // _PyObject_GC_UNTRACK()
 #include "pycore_pyerrors.h"      // _PyErr_Occurred()
 #include "structmember.h"         // PyMemberDef
@@ -64,7 +64,6 @@ PyFunction_ClearWatcher(int watcher_id)
     interp->active_func_watchers &= ~(1 << watcher_id);
     return 0;
 }
-
 PyFunctionObject *
 _PyFunction_FromConstructor(PyFrameConstructor *constr)
 {

Programs/_bootstrap_python.c

@@ -14,6 +14,8 @@
 #include "Python/frozen_modules/importlib._bootstrap_external.h"
 /* End includes */
 
+uint32_t _Py_next_func_version = 1;
+
 /* Empty initializer for deepfrozen modules */
 int _Py_Deepfreeze_Init(void)
 {

Programs/_freeze_module.c

@@ -9,6 +9,7 @@
    Keep this file in sync with Programs/_freeze_module.py.
 */
 
+
 #include <Python.h>
 #include <marshal.h>
 #include "pycore_fileutils.h"     // _Py_stat_struct
@@ -22,6 +23,8 @@
 #include <unistd.h>
 #endif
 
+uint32_t _Py_next_func_version = 1;
+
 /* Empty initializer for deepfrozen modules */
 int _Py_Deepfreeze_Init(void)
 {

Python/bytecodes.c

@@ -3452,6 +3452,7 @@ dummy_func(
                 func->func_defaults = POP();
             }
 
+            func->func_version = ((PyCodeObject *)codeobj)->co_version;
             PUSH((PyObject *)func);
         }
 

Python/generated_cases.c.h

@@ -44,6 +44,7 @@ def make_string_literal(b: bytes) -> str:
 CO_FAST_CELL = 0x40
 CO_FAST_FREE = 0x80
 
+next_code_version = 1
 
 def get_localsplus(code: types.CodeType):
     a = collections.defaultdict(int)
@@ -227,6 +228,7 @@ def generate_unicode(self, name: str, s: str) -> str:
 
 
     def generate_code(self, name: str, code: types.CodeType) -> str:
+        global next_code_version
         # The ordering here matches PyCode_NewWithPosOnlyArgs()
         # (but see below).
         co_consts = self.generate(name + "_consts", code.co_consts)
@@ -268,6 +270,8 @@ def generate_code(self, name: str, code: types.CodeType) -> str:
             self.write(f".co_nplaincellvars = {nplaincellvars},")
             self.write(f".co_ncellvars = {ncellvars},")
             self.write(f".co_nfreevars = {nfreevars},")
+            self.write(f".co_version = {next_code_version},")
+            next_code_version += 1
             self.write(f".co_localsplusnames = {co_localsplusnames},")
             self.write(f".co_localspluskinds = {co_localspluskinds},")
             self.write(f".co_filename = {co_filename},")
@@ -461,6 +465,7 @@ def generate(args: list[str], output: TextIO) -> None:
             with printer.block(f"if ({p} < 0)"):
                 printer.write("return -1;")
         printer.write("return 0;")
+    printer.write(f"\nuint32_t _Py_next_func_version = {next_code_version};\n")
     if verbose:
         print(f"Cache hits: {printer.hits}, misses: {printer.misses}")