gh-92031: Deoptimize Static Code at Finalization (GH-92039)

5 files changed, 225 insertions, 10 deletions

diff --git a/Include/internal/pycore_opcode.h b/Include/internal/pycore_opcode.h
index eadcba1..09f6501 100644
--- a/Include/internal/pycore_opcode.h
+++ b/Include/internal/pycore_opcode.h
@@ -16,6 +16,8 @@ extern const uint8_t _PyOpcode_Caches[256];
 
 extern const uint8_t _PyOpcode_Deopt[256];
 
+extern const uint8_t _PyOpcode_Original[256];
+
 #ifdef NEED_OPCODE_TABLES
 static const uint32_t _PyOpcode_RelativeJump[8] = {
     0U,
@@ -235,6 +237,190 @@ const uint8_t _PyOpcode_Deopt[256] = {
     [WITH_EXCEPT_START] = WITH_EXCEPT_START,
     [YIELD_VALUE] = YIELD_VALUE,
 };
+
+const uint8_t _PyOpcode_Original[256] = {
+    [ASYNC_GEN_WRAP] = ASYNC_GEN_WRAP,
+    [BEFORE_ASYNC_WITH] = BEFORE_ASYNC_WITH,
+    [BEFORE_WITH] = BEFORE_WITH,
+    [BINARY_OP] = BINARY_OP,
+    [BINARY_OP_ADAPTIVE] = BINARY_OP,
+    [BINARY_OP_ADD_FLOAT] = BINARY_OP,
+    [BINARY_OP_ADD_INT] = BINARY_OP,
+    [BINARY_OP_ADD_UNICODE] = BINARY_OP,
+    [BINARY_OP_INPLACE_ADD_UNICODE] = BINARY_OP,
+    [BINARY_OP_MULTIPLY_FLOAT] = BINARY_OP,
+    [BINARY_OP_MULTIPLY_INT] = BINARY_OP,
+    [BINARY_OP_SUBTRACT_FLOAT] = BINARY_OP,
+    [BINARY_OP_SUBTRACT_INT] = BINARY_OP,
+    [BINARY_SUBSCR] = BINARY_SUBSCR,
+    [BINARY_SUBSCR_ADAPTIVE] = BINARY_SUBSCR,
+    [BINARY_SUBSCR_DICT] = BINARY_SUBSCR,
+    [BINARY_SUBSCR_GETITEM] = BINARY_SUBSCR,
+    [BINARY_SUBSCR_LIST_INT] = BINARY_SUBSCR,
+    [BINARY_SUBSCR_TUPLE_INT] = BINARY_SUBSCR,
+    [BUILD_CONST_KEY_MAP] = BUILD_CONST_KEY_MAP,
+    [BUILD_LIST] = BUILD_LIST,
+    [BUILD_MAP] = BUILD_MAP,
+    [BUILD_SET] = BUILD_SET,
+    [BUILD_SLICE] = BUILD_SLICE,
+    [BUILD_STRING] = BUILD_STRING,
+    [BUILD_TUPLE] = BUILD_TUPLE,
+    [CACHE] = CACHE,
+    [CALL] = CALL,
+    [CALL_ADAPTIVE] = CALL,
+    [CALL_FUNCTION_EX] = CALL_FUNCTION_EX,
+    [CALL_PY_EXACT_ARGS] = CALL,
+    [CALL_PY_WITH_DEFAULTS] = CALL,
+    [CHECK_EG_MATCH] = CHECK_EG_MATCH,
+    [CHECK_EXC_MATCH] = CHECK_EXC_MATCH,
+    [COMPARE_OP] = COMPARE_OP,
+    [COMPARE_OP_ADAPTIVE] = COMPARE_OP,
+    [COMPARE_OP_FLOAT_JUMP] = COMPARE_OP,
+    [COMPARE_OP_INT_JUMP] = COMPARE_OP,
+    [COMPARE_OP_STR_JUMP] = COMPARE_OP,
+    [CONTAINS_OP] = CONTAINS_OP,
+    [COPY] = COPY,
+    [COPY_FREE_VARS] = COPY_FREE_VARS,
+    [DELETE_ATTR] = DELETE_ATTR,
+    [DELETE_DEREF] = DELETE_DEREF,
+    [DELETE_FAST] = DELETE_FAST,
+    [DELETE_GLOBAL] = DELETE_GLOBAL,
+    [DELETE_NAME] = DELETE_NAME,
+    [DELETE_SUBSCR] = DELETE_SUBSCR,
+    [DICT_MERGE] = DICT_MERGE,
+    [DICT_UPDATE] = DICT_UPDATE,
+    [END_ASYNC_FOR] = END_ASYNC_FOR,
+    [EXTENDED_ARG] = EXTENDED_ARG_QUICK,
+    [EXTENDED_ARG_QUICK] = EXTENDED_ARG_QUICK,
+    [FORMAT_VALUE] = FORMAT_VALUE,
+    [FOR_ITER] = FOR_ITER,
+    [GET_AITER] = GET_AITER,
+    [GET_ANEXT] = GET_ANEXT,
+    [GET_AWAITABLE] = GET_AWAITABLE,
+    [GET_ITER] = GET_ITER,
+    [GET_LEN] = GET_LEN,
+    [GET_YIELD_FROM_ITER] = GET_YIELD_FROM_ITER,
+    [IMPORT_FROM] = IMPORT_FROM,
+    [IMPORT_NAME] = IMPORT_NAME,
+    [IMPORT_STAR] = IMPORT_STAR,
+    [IS_OP] = IS_OP,
+    [JUMP_BACKWARD] = JUMP_BACKWARD,
+    [JUMP_BACKWARD_NO_INTERRUPT] = JUMP_BACKWARD_NO_INTERRUPT,
+    [JUMP_BACKWARD_QUICK] = JUMP_BACKWARD,
+    [JUMP_FORWARD] = JUMP_FORWARD,
+    [JUMP_IF_FALSE_OR_POP] = JUMP_IF_FALSE_OR_POP,
+    [JUMP_IF_TRUE_OR_POP] = JUMP_IF_TRUE_OR_POP,
+    [KW_NAMES] = KW_NAMES,
+    [LIST_APPEND] = LIST_APPEND,
+    [LIST_EXTEND] = LIST_EXTEND,
+    [LIST_TO_TUPLE] = LIST_TO_TUPLE,
+    [LOAD_ASSERTION_ERROR] = LOAD_ASSERTION_ERROR,
+    [LOAD_ATTR] = LOAD_ATTR,
+    [LOAD_ATTR_ADAPTIVE] = LOAD_ATTR,
+    [LOAD_ATTR_INSTANCE_VALUE] = LOAD_ATTR,
+    [LOAD_ATTR_MODULE] = LOAD_ATTR,
+    [LOAD_ATTR_SLOT] = LOAD_ATTR,
+    [LOAD_ATTR_WITH_HINT] = LOAD_ATTR,
+    [LOAD_BUILD_CLASS] = LOAD_BUILD_CLASS,
+    [LOAD_CLASSDEREF] = LOAD_CLASSDEREF,
+    [LOAD_CLOSURE] = LOAD_CLOSURE,
+    [LOAD_CONST] = LOAD_CONST,
+    [LOAD_CONST__LOAD_FAST] = LOAD_CONST,
+    [LOAD_DEREF] = LOAD_DEREF,
+    [LOAD_FAST] = LOAD_FAST,
+    [LOAD_FAST__LOAD_CONST] = LOAD_FAST,
+    [LOAD_FAST__LOAD_FAST] = LOAD_FAST,
+    [LOAD_GLOBAL] = LOAD_GLOBAL,
+    [LOAD_GLOBAL_ADAPTIVE] = LOAD_GLOBAL,
+    [LOAD_GLOBAL_BUILTIN] = LOAD_GLOBAL,
+    [LOAD_GLOBAL_MODULE] = LOAD_GLOBAL,
+    [LOAD_METHOD] = LOAD_METHOD,
+    [LOAD_METHOD_ADAPTIVE] = LOAD_METHOD,
+    [LOAD_METHOD_CLASS] = LOAD_METHOD,
+    [LOAD_METHOD_MODULE] = LOAD_METHOD,
+    [LOAD_METHOD_NO_DICT] = LOAD_METHOD,
+    [LOAD_METHOD_WITH_DICT] = LOAD_METHOD,
+    [LOAD_METHOD_WITH_VALUES] = LOAD_METHOD,
+    [LOAD_NAME] = LOAD_NAME,
+    [MAKE_CELL] = MAKE_CELL,
+    [MAKE_FUNCTION] = MAKE_FUNCTION,
+    [MAP_ADD] = MAP_ADD,
+    [MATCH_CLASS] = MATCH_CLASS,
+    [MATCH_KEYS] = MATCH_KEYS,
+    [MATCH_MAPPING] = MATCH_MAPPING,
+    [MATCH_SEQUENCE] = MATCH_SEQUENCE,
+    [NOP] = NOP,
+    [POP_EXCEPT] = POP_EXCEPT,
+    [POP_JUMP_BACKWARD_IF_FALSE] = POP_JUMP_BACKWARD_IF_FALSE,
+    [POP_JUMP_BACKWARD_IF_NONE] = POP_JUMP_BACKWARD_IF_NONE,
+    [POP_JUMP_BACKWARD_IF_NOT_NONE] = POP_JUMP_BACKWARD_IF_NOT_NONE,
+    [POP_JUMP_BACKWARD_IF_TRUE] = POP_JUMP_BACKWARD_IF_TRUE,
+    [POP_JUMP_FORWARD_IF_FALSE] = POP_JUMP_FORWARD_IF_FALSE,
+    [POP_JUMP_FORWARD_IF_NONE] = POP_JUMP_FORWARD_IF_NONE,
+    [POP_JUMP_FORWARD_IF_NOT_NONE] = POP_JUMP_FORWARD_IF_NOT_NONE,
+    [POP_JUMP_FORWARD_IF_TRUE] = POP_JUMP_FORWARD_IF_TRUE,
+    [POP_TOP] = POP_TOP,
+    [PRECALL] = PRECALL,
+    [PRECALL_ADAPTIVE] = PRECALL,
+    [PRECALL_BOUND_METHOD] = PRECALL,
+    [PRECALL_BUILTIN_CLASS] = PRECALL,
+    [PRECALL_BUILTIN_FAST_WITH_KEYWORDS] = PRECALL,
+    [PRECALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS] = PRECALL,
+    [PRECALL_NO_KW_BUILTIN_FAST] = PRECALL,
+    [PRECALL_NO_KW_BUILTIN_O] = PRECALL,
+    [PRECALL_NO_KW_ISINSTANCE] = PRECALL,
+    [PRECALL_NO_KW_LEN] = PRECALL,
+    [PRECALL_NO_KW_LIST_APPEND] = PRECALL,
+    [PRECALL_NO_KW_METHOD_DESCRIPTOR_FAST] = PRECALL,
+    [PRECALL_NO_KW_METHOD_DESCRIPTOR_NOARGS] = PRECALL,
+    [PRECALL_NO_KW_METHOD_DESCRIPTOR_O] = PRECALL,
+    [PRECALL_NO_KW_STR_1] = PRECALL,
+    [PRECALL_NO_KW_TUPLE_1] = PRECALL,
+    [PRECALL_NO_KW_TYPE_1] = PRECALL,
+    [PRECALL_PYFUNC] = PRECALL,
+    [PREP_RERAISE_STAR] = PREP_RERAISE_STAR,
+    [PRINT_EXPR] = PRINT_EXPR,
+    [PUSH_EXC_INFO] = PUSH_EXC_INFO,
+    [PUSH_NULL] = PUSH_NULL,
+    [RAISE_VARARGS] = RAISE_VARARGS,
+    [RERAISE] = RERAISE,
+    [RESUME] = RESUME,
+    [RESUME_QUICK] = RESUME,
+    [RETURN_GENERATOR] = RETURN_GENERATOR,
+    [RETURN_VALUE] = RETURN_VALUE,
+    [SEND] = SEND,
+    [SETUP_ANNOTATIONS] = SETUP_ANNOTATIONS,
+    [SET_ADD] = SET_ADD,
+    [SET_UPDATE] = SET_UPDATE,
+    [STORE_ATTR] = STORE_ATTR,
+    [STORE_ATTR_ADAPTIVE] = STORE_ATTR,
+    [STORE_ATTR_INSTANCE_VALUE] = STORE_ATTR,
+    [STORE_ATTR_SLOT] = STORE_ATTR,
+    [STORE_ATTR_WITH_HINT] = STORE_ATTR,
+    [STORE_DEREF] = STORE_DEREF,
+    [STORE_FAST] = STORE_FAST,
+    [STORE_FAST__LOAD_FAST] = STORE_FAST,
+    [STORE_FAST__STORE_FAST] = STORE_FAST,
+    [STORE_GLOBAL] = STORE_GLOBAL,
+    [STORE_NAME] = STORE_NAME,
+    [STORE_SUBSCR] = STORE_SUBSCR,
+    [STORE_SUBSCR_ADAPTIVE] = STORE_SUBSCR,
+    [STORE_SUBSCR_DICT] = STORE_SUBSCR,
+    [STORE_SUBSCR_LIST_INT] = STORE_SUBSCR,
+    [SWAP] = SWAP,
+    [UNARY_INVERT] = UNARY_INVERT,
+    [UNARY_NEGATIVE] = UNARY_NEGATIVE,
+    [UNARY_NOT] = UNARY_NOT,
+    [UNARY_POSITIVE] = UNARY_POSITIVE,
+    [UNPACK_EX] = UNPACK_EX,
+    [UNPACK_SEQUENCE] = UNPACK_SEQUENCE,
+    [UNPACK_SEQUENCE_ADAPTIVE] = UNPACK_SEQUENCE,
+    [UNPACK_SEQUENCE_LIST] = UNPACK_SEQUENCE,
+    [UNPACK_SEQUENCE_TUPLE] = UNPACK_SEQUENCE,
+    [UNPACK_SEQUENCE_TWO_TUPLE] = UNPACK_SEQUENCE,
+    [WITH_EXCEPT_START] = WITH_EXCEPT_START,
+    [YIELD_VALUE] = YIELD_VALUE,
+};
 #endif   // NEED_OPCODE_TABLES
 
 #ifdef Py_DEBUG
diff --git a/Lib/test/test_embed.py b/Lib/test/test_embed.py
index 16d1c7d..9bb4bd7 100644
--- a/Lib/test/test_embed.py
+++ b/Lib/test/test_embed.py
@@ -343,6 +343,20 @@ class EmbeddingTests(EmbeddingTestsMixin, unittest.TestCase):
         out, err = self.run_embedded_interpreter("test_repeated_init_exec", code)
         self.assertEqual(out, 'Tests passed\n' * INIT_LOOPS)
 
+    @support.skip_if_pgo_task
+    def test_quickened_static_code_gets_unquickened_at_Py_FINALIZE(self):
+        # https://github.com/python/cpython/issues/92031
+        code = """if 1:
+            from importlib._bootstrap import _handle_fromlist
+            import dis
+            for name in dis.opmap:
+                # quicken this frozen code object.
+                _handle_fromlist(dis, [name], lambda *args: None)
+        """
+        run = self.run_embedded_interpreter
+        for i in range(50):
+            out, err = run("test_repeated_init_exec", code, timeout=60)
+
     def test_ucnhash_capi_reset(self):
         # bpo-47182: unicodeobject.c:ucnhash_capi was not reset on shutdown.
         code = "print('\\N{digit nine}')"
diff --git a/Misc/NEWS.d/next/Core and Builtins/2022-04-29-02-50-41.gh-issue-92031.2PpaIN.rst b/Misc/NEWS.d/next/Core and Builtins/2022-04-29-02-50-41.gh-issue-92031.2PpaIN.rst
new file mode 100644
index 0000000..6bdefb8
--- /dev/null
+++ b/Misc/NEWS.d/next/Core and Builtins/2022-04-29-02-50-41.gh-issue-92031.2PpaIN.rst
@@ -0,0 +1 @@
+Deoptimize statically-allocated code objects during ``Py_FINALIZE()`` so that future ``_PyCode_Quicken`` calls always start with unquickened code.
diff --git a/Objects/codeobject.c b/Objects/codeobject.c
index 9f922da..c2b29be 100644
--- a/Objects/codeobject.c
+++ b/Objects/codeobject.c
@@ -1350,23 +1350,29 @@ _PyCode_GetFreevars(PyCodeObject *co)
     return get_localsplus_names(co, CO_FAST_FREE, co->co_nfreevars);
 }
 
-PyObject *
-_PyCode_GetCode(PyCodeObject *co)
+static void
+deopt_code(_Py_CODEUNIT *instructions, Py_ssize_t len)
 {
-    PyObject *code = PyBytes_FromStringAndSize(NULL, _PyCode_NBYTES(co));
-    if (code == NULL) {
-        return NULL;
-    }
-    _Py_CODEUNIT *instructions = (_Py_CODEUNIT *)PyBytes_AS_STRING(code);
-    for (int i = 0; i < Py_SIZE(co); i++) {
-        _Py_CODEUNIT instruction = _PyCode_CODE(co)[i];
-        int opcode = _PyOpcode_Deopt[_Py_OPCODE(instruction)];
+    for (int i = 0; i < len; i++) {
+        _Py_CODEUNIT instruction = instructions[i];
+        int opcode = _PyOpcode_Original[_Py_OPCODE(instruction)];
         int caches = _PyOpcode_Caches[opcode];
         instructions[i] = _Py_MAKECODEUNIT(opcode, _Py_OPARG(instruction));
         while (caches--) {
             instructions[++i] = _Py_MAKECODEUNIT(CACHE, 0);
         }
     }
+}
+
+PyObject *
+_PyCode_GetCode(PyCodeObject *co)
+{
+    PyObject *code = PyBytes_FromStringAndSize((const char *)_PyCode_CODE(co),
+                                               _PyCode_NBYTES(co));
+    if (code == NULL) {
+        return NULL;
+    }
+    deopt_code((_Py_CODEUNIT *)PyBytes_AS_STRING(code), Py_SIZE(co));
     return code;
 }
 
@@ -2076,6 +2082,7 @@ _PyStaticCode_Dealloc(PyCodeObject *co)
     if (co->co_warmup == 0) {
          _Py_QuickenedCount--;
     }
+    deopt_code(_PyCode_CODE(co), Py_SIZE(co));
     co->co_warmup = QUICKENING_INITIAL_WARMUP_VALUE;
     PyMem_Free(co->co_extra);
     co->co_extra = NULL;
diff --git a/Tools/scripts/generate_opcode_h.py b/Tools/scripts/generate_opcode_h.py
index 6a04297..e1f4f01 100644
--- a/Tools/scripts/generate_opcode_h.py
+++ b/Tools/scripts/generate_opcode_h.py
@@ -117,6 +117,7 @@ def main(opcode_py, outfile='Include/opcode.h', internaloutfile='Include/interna
 
         iobj.write("\nextern const uint8_t _PyOpcode_Caches[256];\n")
         iobj.write("\nextern const uint8_t _PyOpcode_Deopt[256];\n")
+        iobj.write("\nextern const uint8_t _PyOpcode_Original[256];\n")
         iobj.write("\n#ifdef NEED_OPCODE_TABLES\n")
         write_int_array_from_ops("_PyOpcode_RelativeJump", opcode['hasjrel'], iobj)
         write_int_array_from_ops("_PyOpcode_Jump", opcode['hasjrel'] + opcode['hasjabs'], iobj)
@@ -137,6 +138,12 @@ def main(opcode_py, outfile='Include/opcode.h', internaloutfile='Include/interna
         for opt, deopt in sorted(deoptcodes.items()):
             iobj.write(f"    [{opt}] = {deopt},\n")
         iobj.write("};\n")
+        iobj.write("\nconst uint8_t _PyOpcode_Original[256] = {\n")
+        for opt, deopt in sorted(deoptcodes.items()):
+            if opt.startswith("EXTENDED_ARG"):
+                deopt = "EXTENDED_ARG_QUICK"
+            iobj.write(f"    [{opt}] = {deopt},\n")
+        iobj.write("};\n")
         iobj.write("#endif   // NEED_OPCODE_TABLES\n")
 
         fobj.write("\n")