GH-77273: Better bytecodes for f-strings (GH-6132)

Doc/library/dis.rst

@@ -1465,26 +1465,47 @@ iterations of the loop.
    an argument from two-byte to four-byte.
 
 
-.. opcode:: FORMAT_VALUE (flags)
+.. opcode:: CONVERT_VALUE (oparg)
 
-   Used for implementing formatted literal strings (f-strings).  Pops
+   Convert value to a string, depending on ``oparg``::
-   an optional *fmt_spec* from the stack, then a required *value*.
-   *flags* is interpreted as follows:
 
-   * ``(flags & 0x03) == 0x00``: *value* is formatted as-is.
+      value = STACK.pop()
-   * ``(flags & 0x03) == 0x01``: call :func:`str` on *value* before
+      result = func(value)
-     formatting it.
+      STACK.push(result)
-   * ``(flags & 0x03) == 0x02``: call :func:`repr` on *value* before
-     formatting it.
-   * ``(flags & 0x03) == 0x03``: call :func:`ascii` on *value* before
-     formatting it.
-   * ``(flags & 0x04) == 0x04``: pop *fmt_spec* from the stack and use
-     it, else use an empty *fmt_spec*.
 
-   Formatting is performed using :c:func:`PyObject_Format`.  The
+   * ``oparg == 1``: call :func:`str` on *value*
-   result is pushed on the stack.
+   * ``oparg == 2``: call :func:`repr` on *value*
+   * ``oparg == 3``: call :func:`ascii` on *value*
 
-   .. versionadded:: 3.6
+   Used for implementing formatted literal strings (f-strings).
+
+   .. versionadded:: 3.13
+
+
+.. opcode:: FORMAT_SIMPLE
+
+   Formats the value on top of stack::
+
+      value = STACK.pop()
+      result = value.__format__("")
+      STACK.push(result)
+
+   Used for implementing formatted literal strings (f-strings).
+
+   .. versionadded:: 3.13
+
+.. opcode:: FORMAT_SPEC
+
+   Formats the given value with the given format spec::
+
+      spec = STACK.pop()
+      value = STACK.pop()
+      result = value.__format__(spec)
+      STACK.push(result)
+
+   Used for implementing formatted literal strings (f-strings).
+
+   .. versionadded:: 3.13
 
 
 .. opcode:: MATCH_CLASS (count)

Include/internal/pycore_opcode.h

@@ -102,6 +102,7 @@ const uint8_t _PyOpcode_Deopt[256] = {
     [COMPARE_OP_INT] = COMPARE_OP,
     [COMPARE_OP_STR] = COMPARE_OP,
     [CONTAINS_OP] = CONTAINS_OP,
+    [CONVERT_VALUE] = CONVERT_VALUE,
     [COPY] = COPY,
     [COPY_FREE_VARS] = COPY_FREE_VARS,
     [DELETE_ATTR] = DELETE_ATTR,
@@ -117,7 +118,8 @@ const uint8_t _PyOpcode_Deopt[256] = {
     [END_SEND] = END_SEND,
     [ENTER_EXECUTOR] = ENTER_EXECUTOR,
     [EXTENDED_ARG] = EXTENDED_ARG,
-    [FORMAT_VALUE] = FORMAT_VALUE,
+    [FORMAT_SIMPLE] = FORMAT_SIMPLE,
+    [FORMAT_WITH_SPEC] = FORMAT_WITH_SPEC,
     [FOR_ITER] = FOR_ITER,
     [FOR_ITER_GEN] = FOR_ITER,
     [FOR_ITER_LIST] = FOR_ITER,
@@ -286,6 +288,8 @@ static const char *const _PyOpcode_OpName[267] = {
     [CHECK_EG_MATCH] = "CHECK_EG_MATCH",
     [CALL_BUILTIN_FAST_WITH_KEYWORDS] = "CALL_BUILTIN_FAST_WITH_KEYWORDS",
     [CALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS] = "CALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS",
+    [FORMAT_SIMPLE] = "FORMAT_SIMPLE",
+    [FORMAT_WITH_SPEC] = "FORMAT_WITH_SPEC",
     [CALL_NO_KW_BUILTIN_FAST] = "CALL_NO_KW_BUILTIN_FAST",
     [CALL_NO_KW_BUILTIN_O] = "CALL_NO_KW_BUILTIN_O",
     [CALL_NO_KW_ISINSTANCE] = "CALL_NO_KW_ISINSTANCE",
@@ -293,8 +297,6 @@ static const char *const _PyOpcode_OpName[267] = {
     [CALL_NO_KW_LIST_APPEND] = "CALL_NO_KW_LIST_APPEND",
     [CALL_NO_KW_METHOD_DESCRIPTOR_FAST] = "CALL_NO_KW_METHOD_DESCRIPTOR_FAST",
     [CALL_NO_KW_METHOD_DESCRIPTOR_NOARGS] = "CALL_NO_KW_METHOD_DESCRIPTOR_NOARGS",
-    [CALL_NO_KW_METHOD_DESCRIPTOR_O] = "CALL_NO_KW_METHOD_DESCRIPTOR_O",
-    [CALL_NO_KW_STR_1] = "CALL_NO_KW_STR_1",
     [WITH_EXCEPT_START] = "WITH_EXCEPT_START",
     [GET_AITER] = "GET_AITER",
     [GET_ANEXT] = "GET_ANEXT",
@@ -302,39 +304,39 @@ static const char *const _PyOpcode_OpName[267] = {
     [BEFORE_WITH] = "BEFORE_WITH",
     [END_ASYNC_FOR] = "END_ASYNC_FOR",
     [CLEANUP_THROW] = "CLEANUP_THROW",
+    [CALL_NO_KW_METHOD_DESCRIPTOR_O] = "CALL_NO_KW_METHOD_DESCRIPTOR_O",
+    [CALL_NO_KW_STR_1] = "CALL_NO_KW_STR_1",
     [CALL_NO_KW_TUPLE_1] = "CALL_NO_KW_TUPLE_1",
     [CALL_NO_KW_TYPE_1] = "CALL_NO_KW_TYPE_1",
-    [COMPARE_OP_FLOAT] = "COMPARE_OP_FLOAT",
-    [COMPARE_OP_INT] = "COMPARE_OP_INT",
     [STORE_SUBSCR] = "STORE_SUBSCR",
     [DELETE_SUBSCR] = "DELETE_SUBSCR",
+    [COMPARE_OP_FLOAT] = "COMPARE_OP_FLOAT",
+    [COMPARE_OP_INT] = "COMPARE_OP_INT",
     [COMPARE_OP_STR] = "COMPARE_OP_STR",
     [FOR_ITER_LIST] = "FOR_ITER_LIST",
     [FOR_ITER_TUPLE] = "FOR_ITER_TUPLE",
     [FOR_ITER_RANGE] = "FOR_ITER_RANGE",
-    [FOR_ITER_GEN] = "FOR_ITER_GEN",
-    [LOAD_SUPER_ATTR_ATTR] = "LOAD_SUPER_ATTR_ATTR",
     [GET_ITER] = "GET_ITER",
     [GET_YIELD_FROM_ITER] = "GET_YIELD_FROM_ITER",
-    [LOAD_SUPER_ATTR_METHOD] = "LOAD_SUPER_ATTR_METHOD",
+    [FOR_ITER_GEN] = "FOR_ITER_GEN",
     [LOAD_BUILD_CLASS] = "LOAD_BUILD_CLASS",
-    [LOAD_ATTR_CLASS] = "LOAD_ATTR_CLASS",
+    [LOAD_SUPER_ATTR_ATTR] = "LOAD_SUPER_ATTR_ATTR",
-    [LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN] = "LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN",
+    [LOAD_SUPER_ATTR_METHOD] = "LOAD_SUPER_ATTR_METHOD",
     [LOAD_ASSERTION_ERROR] = "LOAD_ASSERTION_ERROR",
     [RETURN_GENERATOR] = "RETURN_GENERATOR",
+    [LOAD_ATTR_CLASS] = "LOAD_ATTR_CLASS",
+    [LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN] = "LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN",
     [LOAD_ATTR_INSTANCE_VALUE] = "LOAD_ATTR_INSTANCE_VALUE",
     [LOAD_ATTR_MODULE] = "LOAD_ATTR_MODULE",
     [LOAD_ATTR_PROPERTY] = "LOAD_ATTR_PROPERTY",
     [LOAD_ATTR_SLOT] = "LOAD_ATTR_SLOT",
     [LOAD_ATTR_WITH_HINT] = "LOAD_ATTR_WITH_HINT",
-    [LOAD_ATTR_METHOD_LAZY_DICT] = "LOAD_ATTR_METHOD_LAZY_DICT",
-    [LOAD_ATTR_METHOD_NO_DICT] = "LOAD_ATTR_METHOD_NO_DICT",
     [RETURN_VALUE] = "RETURN_VALUE",
-    [LOAD_ATTR_METHOD_WITH_VALUES] = "LOAD_ATTR_METHOD_WITH_VALUES",
+    [LOAD_ATTR_METHOD_LAZY_DICT] = "LOAD_ATTR_METHOD_LAZY_DICT",
     [SETUP_ANNOTATIONS] = "SETUP_ANNOTATIONS",
-    [LOAD_GLOBAL_BUILTIN] = "LOAD_GLOBAL_BUILTIN",
+    [LOAD_ATTR_METHOD_NO_DICT] = "LOAD_ATTR_METHOD_NO_DICT",
     [LOAD_LOCALS] = "LOAD_LOCALS",
-    [LOAD_GLOBAL_MODULE] = "LOAD_GLOBAL_MODULE",
+    [LOAD_ATTR_METHOD_WITH_VALUES] = "LOAD_ATTR_METHOD_WITH_VALUES",
     [POP_EXCEPT] = "POP_EXCEPT",
     [STORE_NAME] = "STORE_NAME",
     [DELETE_NAME] = "DELETE_NAME",
@@ -357,9 +359,9 @@ static const char *const _PyOpcode_OpName[267] = {
     [IMPORT_NAME] = "IMPORT_NAME",
     [IMPORT_FROM] = "IMPORT_FROM",
     [JUMP_FORWARD] = "JUMP_FORWARD",
+    [LOAD_GLOBAL_BUILTIN] = "LOAD_GLOBAL_BUILTIN",
+    [LOAD_GLOBAL_MODULE] = "LOAD_GLOBAL_MODULE",
     [STORE_ATTR_INSTANCE_VALUE] = "STORE_ATTR_INSTANCE_VALUE",
-    [STORE_ATTR_SLOT] = "STORE_ATTR_SLOT",
-    [STORE_ATTR_WITH_HINT] = "STORE_ATTR_WITH_HINT",
     [POP_JUMP_IF_FALSE] = "POP_JUMP_IF_FALSE",
     [POP_JUMP_IF_TRUE] = "POP_JUMP_IF_TRUE",
     [LOAD_GLOBAL] = "LOAD_GLOBAL",
@@ -378,7 +380,7 @@ static const char *const _PyOpcode_OpName[267] = {
     [POP_JUMP_IF_NONE] = "POP_JUMP_IF_NONE",
     [RAISE_VARARGS] = "RAISE_VARARGS",
     [GET_AWAITABLE] = "GET_AWAITABLE",
-    [STORE_SUBSCR_DICT] = "STORE_SUBSCR_DICT",
+    [STORE_ATTR_SLOT] = "STORE_ATTR_SLOT",
     [BUILD_SLICE] = "BUILD_SLICE",
     [JUMP_BACKWARD_NO_INTERRUPT] = "JUMP_BACKWARD_NO_INTERRUPT",
     [MAKE_CELL] = "MAKE_CELL",
@@ -394,20 +396,20 @@ static const char *const _PyOpcode_OpName[267] = {
     [LIST_APPEND] = "LIST_APPEND",
     [SET_ADD] = "SET_ADD",
     [MAP_ADD] = "MAP_ADD",
-    [STORE_SUBSCR_LIST_INT] = "STORE_SUBSCR_LIST_INT",
+    [STORE_ATTR_WITH_HINT] = "STORE_ATTR_WITH_HINT",
     [COPY_FREE_VARS] = "COPY_FREE_VARS",
     [YIELD_VALUE] = "YIELD_VALUE",
     [RESUME] = "RESUME",
     [MATCH_CLASS] = "MATCH_CLASS",
+    [STORE_SUBSCR_DICT] = "STORE_SUBSCR_DICT",
+    [STORE_SUBSCR_LIST_INT] = "STORE_SUBSCR_LIST_INT",
     [UNPACK_SEQUENCE_LIST] = "UNPACK_SEQUENCE_LIST",
-    [UNPACK_SEQUENCE_TUPLE] = "UNPACK_SEQUENCE_TUPLE",
-    [FORMAT_VALUE] = "FORMAT_VALUE",
     [BUILD_CONST_KEY_MAP] = "BUILD_CONST_KEY_MAP",
     [BUILD_STRING] = "BUILD_STRING",
+    [CONVERT_VALUE] = "CONVERT_VALUE",
+    [UNPACK_SEQUENCE_TUPLE] = "UNPACK_SEQUENCE_TUPLE",
     [UNPACK_SEQUENCE_TWO_TUPLE] = "UNPACK_SEQUENCE_TWO_TUPLE",
     [SEND_GEN] = "SEND_GEN",
-    [160] = "<160>",
-    [161] = "<161>",
     [LIST_EXTEND] = "LIST_EXTEND",
     [SET_UPDATE] = "SET_UPDATE",
     [DICT_MERGE] = "DICT_MERGE",
@@ -517,8 +519,6 @@ static const char *const _PyOpcode_OpName[267] = {
 #endif
 
 #define EXTRA_CASES \
-    case 160: \
-    case 161: \
     case 166: \
     case 167: \
     case 178: \

Include/opcode.h

@@ -30,6 +30,8 @@ extern "C" {
 #define PUSH_EXC_INFO                           35
 #define CHECK_EXC_MATCH                         36
 #define CHECK_EG_MATCH                          37
+#define FORMAT_SIMPLE                           40
+#define FORMAT_WITH_SPEC                        41
 #define WITH_EXCEPT_START                       49
 #define GET_AITER                               50
 #define GET_ANEXT                               51
@@ -107,9 +109,9 @@ extern "C" {
 #define YIELD_VALUE                            150
 #define RESUME                                 151
 #define MATCH_CLASS                            152
-#define FORMAT_VALUE                           155
 #define BUILD_CONST_KEY_MAP                    156
 #define BUILD_STRING                           157
+#define CONVERT_VALUE                          158
 #define LIST_EXTEND                            162
 #define SET_UPDATE                             163
 #define DICT_MERGE                             164
@@ -175,47 +177,47 @@ extern "C" {
 #define CALL_BUILTIN_CLASS                      34
 #define CALL_BUILTIN_FAST_WITH_KEYWORDS         38
 #define CALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS  39
-#define CALL_NO_KW_BUILTIN_FAST                 40
+#define CALL_NO_KW_BUILTIN_FAST                 42
-#define CALL_NO_KW_BUILTIN_O                    41
+#define CALL_NO_KW_BUILTIN_O                    43
-#define CALL_NO_KW_ISINSTANCE                   42
+#define CALL_NO_KW_ISINSTANCE                   44
-#define CALL_NO_KW_LEN                          43
+#define CALL_NO_KW_LEN                          45
-#define CALL_NO_KW_LIST_APPEND                  44
+#define CALL_NO_KW_LIST_APPEND                  46
-#define CALL_NO_KW_METHOD_DESCRIPTOR_FAST       45
+#define CALL_NO_KW_METHOD_DESCRIPTOR_FAST       47
-#define CALL_NO_KW_METHOD_DESCRIPTOR_NOARGS     46
+#define CALL_NO_KW_METHOD_DESCRIPTOR_NOARGS     48
-#define CALL_NO_KW_METHOD_DESCRIPTOR_O          47
+#define CALL_NO_KW_METHOD_DESCRIPTOR_O          56
-#define CALL_NO_KW_STR_1                        48
+#define CALL_NO_KW_STR_1                        57
-#define CALL_NO_KW_TUPLE_1                      56
+#define CALL_NO_KW_TUPLE_1                      58
-#define CALL_NO_KW_TYPE_1                       57
+#define CALL_NO_KW_TYPE_1                       59
-#define COMPARE_OP_FLOAT                        58
+#define COMPARE_OP_FLOAT                        62
-#define COMPARE_OP_INT                          59
+#define COMPARE_OP_INT                          63
-#define COMPARE_OP_STR                          62
+#define COMPARE_OP_STR                          64
-#define FOR_ITER_LIST                           63
+#define FOR_ITER_LIST                           65
-#define FOR_ITER_TUPLE                          64
+#define FOR_ITER_TUPLE                          66
-#define FOR_ITER_RANGE                          65
+#define FOR_ITER_RANGE                          67
-#define FOR_ITER_GEN                            66
+#define FOR_ITER_GEN                            70
-#define LOAD_SUPER_ATTR_ATTR                    67
+#define LOAD_SUPER_ATTR_ATTR                    72
-#define LOAD_SUPER_ATTR_METHOD                  70
+#define LOAD_SUPER_ATTR_METHOD                  73
-#define LOAD_ATTR_CLASS                         72
+#define LOAD_ATTR_CLASS                         76
-#define LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN       73
+#define LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN       77
-#define LOAD_ATTR_INSTANCE_VALUE                76
+#define LOAD_ATTR_INSTANCE_VALUE                78
-#define LOAD_ATTR_MODULE                        77
+#define LOAD_ATTR_MODULE                        79
-#define LOAD_ATTR_PROPERTY                      78
+#define LOAD_ATTR_PROPERTY                      80
-#define LOAD_ATTR_SLOT                          79
+#define LOAD_ATTR_SLOT                          81
-#define LOAD_ATTR_WITH_HINT                     80
+#define LOAD_ATTR_WITH_HINT                     82
-#define LOAD_ATTR_METHOD_LAZY_DICT              81
+#define LOAD_ATTR_METHOD_LAZY_DICT              84
-#define LOAD_ATTR_METHOD_NO_DICT                82
+#define LOAD_ATTR_METHOD_NO_DICT                86
-#define LOAD_ATTR_METHOD_WITH_VALUES            84
+#define LOAD_ATTR_METHOD_WITH_VALUES            88
-#define LOAD_GLOBAL_BUILTIN                     86
+#define LOAD_GLOBAL_BUILTIN                    111
-#define LOAD_GLOBAL_MODULE                      88
+#define LOAD_GLOBAL_MODULE                     112
-#define STORE_ATTR_INSTANCE_VALUE              111
+#define STORE_ATTR_INSTANCE_VALUE              113
-#define STORE_ATTR_SLOT                        112
+#define STORE_ATTR_SLOT                        132
-#define STORE_ATTR_WITH_HINT                   113
+#define STORE_ATTR_WITH_HINT                   148
-#define STORE_SUBSCR_DICT                      132
+#define STORE_SUBSCR_DICT                      153
-#define STORE_SUBSCR_LIST_INT                  148
+#define STORE_SUBSCR_LIST_INT                  154
-#define UNPACK_SEQUENCE_LIST                   153
+#define UNPACK_SEQUENCE_LIST                   155
-#define UNPACK_SEQUENCE_TUPLE                  154
+#define UNPACK_SEQUENCE_TUPLE                  159
-#define UNPACK_SEQUENCE_TWO_TUPLE              158
+#define UNPACK_SEQUENCE_TWO_TUPLE              160
-#define SEND_GEN                               159
+#define SEND_GEN                               161
 
 #define HAS_ARG(op) ((((op) >= HAVE_ARGUMENT) && (!IS_PSEUDO_OPCODE(op)))\
     || ((op) == JUMP) \

Lib/dis.py

@@ -25,13 +25,8 @@ del _opcodes_all
 _have_code = (types.MethodType, types.FunctionType, types.CodeType,
               classmethod, staticmethod, type)
 
-FORMAT_VALUE = opmap['FORMAT_VALUE']
+CONVERT_VALUE = opmap['CONVERT_VALUE']
-FORMAT_VALUE_CONVERTERS = (
+
-    (None, ''),
-    (str, 'str'),
-    (repr, 'repr'),
-    (ascii, 'ascii'),
-)
 SET_FUNCTION_ATTRIBUTE = opmap['SET_FUNCTION_ATTRIBUTE']
 FUNCTION_ATTR_FLAGS = ('defaults', 'kwdefaults', 'annotations', 'closure')
 
@@ -579,13 +574,9 @@ def _get_instructions_bytes(code, varname_from_oparg=None,
             elif deop in hascompare:
                 argval = cmp_op[arg>>4]
                 argrepr = argval
-            elif deop == FORMAT_VALUE:
+            elif deop == CONVERT_VALUE:
-                argval, argrepr = FORMAT_VALUE_CONVERTERS[arg & 0x3]
+                argval = (None, str, repr, ascii)[arg]
-                argval = (argval, bool(arg & 0x4))
+                argrepr = ('', 'str', 'repr', 'ascii')[arg]
-                if argval[1]:
-                    if argrepr:
-                        argrepr += ', '
-                    argrepr += 'with format'
             elif deop == SET_FUNCTION_ATTRIBUTE:
                 argrepr = ', '.join(s for i, s in enumerate(FUNCTION_ATTR_FLAGS)
                                     if arg & (1<<i))

Lib/importlib/_bootstrap_external.py

@@ -447,7 +447,9 @@ _code_type = type(_write_atomic.__code__)
 #     Python 3.12b1 3531 (Add PEP 695 changes)
 #     Python 3.13a1 3550 (Plugin optimizer support)
 #     Python 3.13a1 3551 (Compact superinstructions)
-#     Python 3.13a1 3552 (Add SET_FUNCTION_ATTRIBUTE)
+#     Python 3.13a1 3552 (Remove LOAD_FAST__LOAD_CONST and LOAD_CONST__LOAD_FAST)
+#     Python 3.13a1 3553 (Add SET_FUNCTION_ATTRIBUTE)
+#     Python 3.13a1 3554 (more efficient bytecodes for f-strings)
 
 #     Python 3.14 will start with 3600
 
@@ -464,7 +466,7 @@ _code_type = type(_write_atomic.__code__)
 # Whenever MAGIC_NUMBER is changed, the ranges in the magic_values array
 # in PC/launcher.c must also be updated.
 
-MAGIC_NUMBER = (3552).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3554).to_bytes(2, 'little') + b'\r\n'
 
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 

Lib/opcode.py

@@ -110,6 +110,9 @@ def_op('PUSH_EXC_INFO', 35)
 def_op('CHECK_EXC_MATCH', 36)
 def_op('CHECK_EG_MATCH', 37)
 
+def_op('FORMAT_SIMPLE', 40)
+def_op('FORMAT_WITH_SPEC', 41)
+
 def_op('WITH_EXCEPT_START', 49)
 def_op('GET_AITER', 50)
 def_op('GET_ANEXT', 51)
@@ -213,9 +216,9 @@ def_op('YIELD_VALUE', 150)
 def_op('RESUME', 151)   # This must be kept in sync with deepfreeze.py
 def_op('MATCH_CLASS', 152)
 
-def_op('FORMAT_VALUE', 155)
 def_op('BUILD_CONST_KEY_MAP', 156)
 def_op('BUILD_STRING', 157)
+def_op('CONVERT_VALUE', 158)
 
 def_op('LIST_EXTEND', 162)
 def_op('SET_UPDATE', 163)

Lib/test/test_dis.py

@@ -443,18 +443,20 @@ dis_fstring = """\
 %3d        RESUME                   0
 
 %3d        LOAD_FAST                0 (a)
-           FORMAT_VALUE             0
+           FORMAT_SIMPLE
            LOAD_CONST               1 (' ')
            LOAD_FAST                1 (b)
            LOAD_CONST               2 ('4')
-           FORMAT_VALUE             4 (with format)
+           FORMAT_WITH_SPEC
            LOAD_CONST               1 (' ')
            LOAD_FAST                2 (c)
-           FORMAT_VALUE             2 (repr)
+           CONVERT_VALUE            2 (repr)
+           FORMAT_SIMPLE
            LOAD_CONST               1 (' ')
            LOAD_FAST                3 (d)
+           CONVERT_VALUE            2 (repr)
            LOAD_CONST               2 ('4')
-           FORMAT_VALUE             6 (repr, with format)
+           FORMAT_WITH_SPEC
            BUILD_STRING             7
            RETURN_VALUE
 """ % (_fstring.__code__.co_firstlineno, _fstring.__code__.co_firstlineno + 1)

Misc/NEWS.d/next/Core and Builtins/2023-06-05-08-30-49.gh-issue-33092.hZ0xSI.rst

@@ -0,0 +1,3 @@
+Simplify and speed up interpreter for f-strings. Removes ``FORMAT_VALUE``
+opcode. Add ``CONVERT_VALUE``, ``FORMAT_SIMPLE`` and ``FORMAT_WITH_SPEC``
+opcode. Compiler emits more efficient sequence for each format expression.

Programs/test_frozenmain.h

@@ -9,8 +9,8 @@ unsigned char M_test_frozenmain[] = {
     1,0,2,0,101,1,106,8,0,0,0,0,0,0,0,0,
     0,0,0,0,0,0,0,0,0,0,171,0,0,0,0,0,
     0,0,100,4,25,0,0,0,90,5,100,5,68,0,93,20,
-    0,0,90,6,2,0,101,2,100,6,101,6,155,0,100,7,
+    0,0,90,6,2,0,101,2,100,6,101,6,40,0,100,7,
-    101,5,101,6,25,0,0,0,155,0,157,4,171,1,0,0,
+    101,5,101,6,25,0,0,0,40,0,157,4,171,1,0,0,
     0,0,0,0,1,0,140,22,0,0,4,0,121,1,41,8,
     233,0,0,0,0,78,122,18,70,114,111,122,101,110,32,72,
     101,108,108,111,32,87,111,114,108,100,122,8,115,121,115,46,

Python/bytecodes.c

@@ -52,6 +52,8 @@
 #define family(name, ...) static int family_##name
 #define pseudo(name) static int pseudo_##name
 
+typedef PyObject *(*convertion_func_ptr)(PyObject *);
+
 // Dummy variables for stack effects.
 static PyObject *value, *value1, *value2, *left, *right, *res, *sum, *prod, *sub;
 static PyObject *container, *start, *stop, *v, *lhs, *rhs, *res2;
@@ -3367,43 +3369,35 @@ dummy_func(
             ERROR_IF(slice == NULL, error);
         }
 
-        inst(FORMAT_VALUE, (value, fmt_spec if ((oparg & FVS_MASK) == FVS_HAVE_SPEC) -- result)) {
+        inst(CONVERT_VALUE, (value -- result)) {
-            /* Handles f-string value formatting. */
+            convertion_func_ptr  conv_fn;
-            PyObject *(*conv_fn)(PyObject *);
+            assert(oparg >= FVC_STR && oparg <= FVC_ASCII);
-            int which_conversion = oparg & FVC_MASK;
+            conv_fn = CONVERSION_FUNCTIONS[oparg];
-
+            result = conv_fn(value);
-            /* See if any conversion is specified. */
-            switch (which_conversion) {
-            case FVC_NONE:  conv_fn = NULL;           break;
-            case FVC_STR:   conv_fn = PyObject_Str;   break;
-            case FVC_REPR:  conv_fn = PyObject_Repr;  break;
-            case FVC_ASCII: conv_fn = PyObject_ASCII; break;
-            default:
-                _PyErr_Format(tstate, PyExc_SystemError,
-                              "unexpected conversion flag %d",
-                              which_conversion);
-                goto error;
-            }
-
-            /* If there's a conversion function, call it and replace
-               value with that result. Otherwise, just use value,
-               without conversion. */
-            if (conv_fn != NULL) {
-                result = conv_fn(value);
-                Py_DECREF(value);
-                if (result == NULL) {
-                    Py_XDECREF(fmt_spec);
-                    ERROR_IF(true, error);
-                }
-                value = result;
-            }
-
-            result = PyObject_Format(value, fmt_spec);
             Py_DECREF(value);
-            Py_XDECREF(fmt_spec);
             ERROR_IF(result == NULL, error);
         }
 
+        inst(FORMAT_SIMPLE, (value -- res)) {
+            /* If value is a unicode object, then we know the result
+             * of format(value) is value itself. */
+            if (!PyUnicode_CheckExact(value)) {
+                res = PyObject_Format(value, NULL);
+                Py_DECREF(value);
+                ERROR_IF(res == NULL, error);
+            }
+            else {
+                res = value;
+            }
+        }
+
+        inst(FORMAT_WITH_SPEC, (value, fmt_spec -- res)) {
+            res = PyObject_Format(value, fmt_spec);
+            Py_DECREF(value);
+            Py_DECREF(fmt_spec);
+            ERROR_IF(res == NULL, error);
+        }
+
         inst(COPY, (bottom, unused[oparg-1] -- bottom, unused[oparg-1], top)) {
             assert(oparg > 0);
             top = Py_NewRef(bottom);

Python/ceval.c

@@ -222,6 +222,14 @@ _PyEvalFramePushAndInit_Ex(PyThreadState *tstate, PyFunctionObject *func,
 static void
 _PyEvalFrameClearAndPop(PyThreadState *tstate, _PyInterpreterFrame *frame);
 
+typedef PyObject *(*convertion_func_ptr)(PyObject *);
+
+static const convertion_func_ptr CONVERSION_FUNCTIONS[4] = {
+    [FVC_STR] = PyObject_Str,
+    [FVC_REPR] = PyObject_Repr,
+    [FVC_ASCII] = PyObject_ASCII
+};
+
 #define UNBOUNDLOCAL_ERROR_MSG \
     "cannot access local variable '%s' where it is not associated with a value"
 #define UNBOUNDFREE_ERROR_MSG \

Python/compile.c

@@ -4985,26 +4985,26 @@ compiler_formatted_value(struct compiler *c, expr_ty e)
     /* The expression to be formatted. */
     VISIT(c, expr, e->v.FormattedValue.value);
 
-    switch (conversion) {
+    location loc = LOC(e);
-    case 's': oparg = FVC_STR;   break;
+    if (conversion != -1) {
-    case 'r': oparg = FVC_REPR;  break;
+        switch (conversion) {
-    case 'a': oparg = FVC_ASCII; break;
+        case 's': oparg = FVC_STR;   break;
-    case -1:  oparg = FVC_NONE;  break;
+        case 'r': oparg = FVC_REPR;  break;
-    default:
+        case 'a': oparg = FVC_ASCII; break;
-        PyErr_Format(PyExc_SystemError,
+        default:
+            PyErr_Format(PyExc_SystemError,
                      "Unrecognized conversion character %d", conversion);
-        return ERROR;
+            return ERROR;
+        }
+        ADDOP_I(c, loc, CONVERT_VALUE, oparg);
     }
     if (e->v.FormattedValue.format_spec) {
         /* Evaluate the format spec, and update our opcode arg. */
         VISIT(c, expr, e->v.FormattedValue.format_spec);
-        oparg |= FVS_HAVE_SPEC;
+        ADDOP(c, loc, FORMAT_WITH_SPEC);
+    } else {
+        ADDOP(c, loc, FORMAT_SIMPLE);
     }
-
-    /* And push our opcode and oparg */
-    location loc = LOC(e);
-    ADDOP_I(c, loc, FORMAT_VALUE, oparg);
-
     return SUCCESS;
 }
 

Python/opcode_metadata.h

@@ -402,8 +402,12 @@ _PyOpcode_num_popped(int opcode, int oparg, bool jump) {
             return 0;
         case BUILD_SLICE:
             return ((oparg == 3) ? 1 : 0) + 2;
-        case FORMAT_VALUE:
+        case CONVERT_VALUE:
-            return (((oparg & FVS_MASK) == FVS_HAVE_SPEC) ? 1 : 0) + 1;
+            return 1;
+        case FORMAT_SIMPLE:
+            return 1;
+        case FORMAT_WITH_SPEC:
+            return 2;
         case COPY:
             return (oparg-1) + 1;
         case BINARY_OP:
@@ -820,7 +824,11 @@ _PyOpcode_num_pushed(int opcode, int oparg, bool jump) {
             return 0;
         case BUILD_SLICE:
             return 1;
-        case FORMAT_VALUE:
+        case CONVERT_VALUE:
+            return 1;
+        case FORMAT_SIMPLE:
+            return 1;
+        case FORMAT_WITH_SPEC:
             return 1;
         case COPY:
             return (oparg-1) + 2;
@@ -1064,7 +1072,9 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[512] = {
     [SET_FUNCTION_ATTRIBUTE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG },
     [RETURN_GENERATOR] = { true, INSTR_FMT_IX, 0 },
     [BUILD_SLICE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG },
-    [FORMAT_VALUE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG },
+    [CONVERT_VALUE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG },
+    [FORMAT_SIMPLE] = { true, INSTR_FMT_IX, 0 },
+    [FORMAT_WITH_SPEC] = { true, INSTR_FMT_IX, 0 },
     [COPY] = { true, INSTR_FMT_IB, HAS_ARG_FLAG },
     [BINARY_OP] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG },
     [SWAP] = { true, INSTR_FMT_IB, HAS_ARG_FLAG },

Python/opcode_targets.h

@@ -39,6 +39,8 @@ static void *opcode_targets[256] = {
     &&TARGET_CHECK_EG_MATCH,
     &&TARGET_CALL_BUILTIN_FAST_WITH_KEYWORDS,
     &&TARGET_CALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS,
+    &&TARGET_FORMAT_SIMPLE,
+    &&TARGET_FORMAT_WITH_SPEC,
     &&TARGET_CALL_NO_KW_BUILTIN_FAST,
     &&TARGET_CALL_NO_KW_BUILTIN_O,
     &&TARGET_CALL_NO_KW_ISINSTANCE,
@@ -46,8 +48,6 @@ static void *opcode_targets[256] = {
     &&TARGET_CALL_NO_KW_LIST_APPEND,
     &&TARGET_CALL_NO_KW_METHOD_DESCRIPTOR_FAST,
     &&TARGET_CALL_NO_KW_METHOD_DESCRIPTOR_NOARGS,
-    &&TARGET_CALL_NO_KW_METHOD_DESCRIPTOR_O,
-    &&TARGET_CALL_NO_KW_STR_1,
     &&TARGET_WITH_EXCEPT_START,
     &&TARGET_GET_AITER,
     &&TARGET_GET_ANEXT,
@@ -55,39 +55,39 @@ static void *opcode_targets[256] = {
     &&TARGET_BEFORE_WITH,
     &&TARGET_END_ASYNC_FOR,
     &&TARGET_CLEANUP_THROW,
+    &&TARGET_CALL_NO_KW_METHOD_DESCRIPTOR_O,
+    &&TARGET_CALL_NO_KW_STR_1,
     &&TARGET_CALL_NO_KW_TUPLE_1,
     &&TARGET_CALL_NO_KW_TYPE_1,
-    &&TARGET_COMPARE_OP_FLOAT,
-    &&TARGET_COMPARE_OP_INT,
     &&TARGET_STORE_SUBSCR,
     &&TARGET_DELETE_SUBSCR,
+    &&TARGET_COMPARE_OP_FLOAT,
+    &&TARGET_COMPARE_OP_INT,
     &&TARGET_COMPARE_OP_STR,
     &&TARGET_FOR_ITER_LIST,
     &&TARGET_FOR_ITER_TUPLE,
     &&TARGET_FOR_ITER_RANGE,
-    &&TARGET_FOR_ITER_GEN,
-    &&TARGET_LOAD_SUPER_ATTR_ATTR,
     &&TARGET_GET_ITER,
     &&TARGET_GET_YIELD_FROM_ITER,
-    &&TARGET_LOAD_SUPER_ATTR_METHOD,
+    &&TARGET_FOR_ITER_GEN,
     &&TARGET_LOAD_BUILD_CLASS,
-    &&TARGET_LOAD_ATTR_CLASS,
+    &&TARGET_LOAD_SUPER_ATTR_ATTR,
-    &&TARGET_LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN,
+    &&TARGET_LOAD_SUPER_ATTR_METHOD,
     &&TARGET_LOAD_ASSERTION_ERROR,
     &&TARGET_RETURN_GENERATOR,
+    &&TARGET_LOAD_ATTR_CLASS,
+    &&TARGET_LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN,
     &&TARGET_LOAD_ATTR_INSTANCE_VALUE,
     &&TARGET_LOAD_ATTR_MODULE,
     &&TARGET_LOAD_ATTR_PROPERTY,
     &&TARGET_LOAD_ATTR_SLOT,
     &&TARGET_LOAD_ATTR_WITH_HINT,
-    &&TARGET_LOAD_ATTR_METHOD_LAZY_DICT,
-    &&TARGET_LOAD_ATTR_METHOD_NO_DICT,
     &&TARGET_RETURN_VALUE,
-    &&TARGET_LOAD_ATTR_METHOD_WITH_VALUES,
+    &&TARGET_LOAD_ATTR_METHOD_LAZY_DICT,
     &&TARGET_SETUP_ANNOTATIONS,
-    &&TARGET_LOAD_GLOBAL_BUILTIN,
+    &&TARGET_LOAD_ATTR_METHOD_NO_DICT,
     &&TARGET_LOAD_LOCALS,
-    &&TARGET_LOAD_GLOBAL_MODULE,
+    &&TARGET_LOAD_ATTR_METHOD_WITH_VALUES,
     &&TARGET_POP_EXCEPT,
     &&TARGET_STORE_NAME,
     &&TARGET_DELETE_NAME,
@@ -110,9 +110,9 @@ static void *opcode_targets[256] = {
     &&TARGET_IMPORT_NAME,
     &&TARGET_IMPORT_FROM,
     &&TARGET_JUMP_FORWARD,
+    &&TARGET_LOAD_GLOBAL_BUILTIN,
+    &&TARGET_LOAD_GLOBAL_MODULE,
     &&TARGET_STORE_ATTR_INSTANCE_VALUE,
-    &&TARGET_STORE_ATTR_SLOT,
-    &&TARGET_STORE_ATTR_WITH_HINT,
     &&TARGET_POP_JUMP_IF_FALSE,
     &&TARGET_POP_JUMP_IF_TRUE,
     &&TARGET_LOAD_GLOBAL,
@@ -131,7 +131,7 @@ static void *opcode_targets[256] = {
     &&TARGET_POP_JUMP_IF_NONE,
     &&TARGET_RAISE_VARARGS,
     &&TARGET_GET_AWAITABLE,
-    &&TARGET_STORE_SUBSCR_DICT,
+    &&TARGET_STORE_ATTR_SLOT,
     &&TARGET_BUILD_SLICE,
     &&TARGET_JUMP_BACKWARD_NO_INTERRUPT,
     &&TARGET_MAKE_CELL,
@@ -147,20 +147,20 @@ static void *opcode_targets[256] = {
     &&TARGET_LIST_APPEND,
     &&TARGET_SET_ADD,
     &&TARGET_MAP_ADD,
-    &&TARGET_STORE_SUBSCR_LIST_INT,
+    &&TARGET_STORE_ATTR_WITH_HINT,
     &&TARGET_COPY_FREE_VARS,
     &&TARGET_YIELD_VALUE,
     &&TARGET_RESUME,
     &&TARGET_MATCH_CLASS,
+    &&TARGET_STORE_SUBSCR_DICT,
+    &&TARGET_STORE_SUBSCR_LIST_INT,
     &&TARGET_UNPACK_SEQUENCE_LIST,
-    &&TARGET_UNPACK_SEQUENCE_TUPLE,
-    &&TARGET_FORMAT_VALUE,
     &&TARGET_BUILD_CONST_KEY_MAP,
     &&TARGET_BUILD_STRING,
+    &&TARGET_CONVERT_VALUE,
+    &&TARGET_UNPACK_SEQUENCE_TUPLE,
     &&TARGET_UNPACK_SEQUENCE_TWO_TUPLE,
     &&TARGET_SEND_GEN,
-    &&_unknown_opcode,
-    &&_unknown_opcode,
     &&TARGET_LIST_EXTEND,
     &&TARGET_SET_UPDATE,
     &&TARGET_DICT_MERGE,