bpo-44511: Improve the bytecode for class and mapping patterns (GH-26922)

* Refactor mapping patterns and speed up class patterns.

* Simplify MATCH_KEYS and MATCH_CLASS.

* Add COPY opcode.

Author: brandtbucher

Doc/library/dis.rst

@@ -755,15 +755,6 @@ iterations of the loop.
    .. versionadded:: 3.11
 
 
-.. opcode:: COPY_DICT_WITHOUT_KEYS
-
-   TOS is a tuple of mapping keys, and TOS1 is the match subject.  Replace TOS
-   with a :class:`dict` formed from the items of TOS1, but without any of the
-   keys in TOS.
-
-   .. versionadded:: 3.10
-
-
 .. opcode:: GET_LEN
 
    Push ``len(TOS)`` onto the stack.
@@ -795,11 +786,14 @@ iterations of the loop.
 
    TOS is a tuple of mapping keys, and TOS1 is the match subject.  If TOS1
    contains all of the keys in TOS, push a :class:`tuple` containing the
-   corresponding values, followed by ``True``. Otherwise, push ``None``,
-   followed by ``False``.
+   corresponding values. Otherwise, push ``None``.
 
    .. versionadded:: 3.10
 
+   .. versionchanged:: 3.11
+      Previously, this instruction also pushed a boolean value indicating
+      success (``True``) or failure (``False``).
+
 
 All of the following opcodes use their arguments.
 
@@ -1277,12 +1271,16 @@ All of the following opcodes use their arguments.
    against, and TOS2 is the match subject.  *count* is the number of positional
    sub-patterns.
 
-   Pop TOS.  If TOS2 is an instance of TOS1 and has the positional and keyword
-   attributes required by *count* and TOS, set TOS to ``True`` and TOS1 to a
-   tuple of extracted attributes.  Otherwise, set TOS to ``False``.
+   Pop TOS, TOS1, and TOS2.  If TOS2 is an instance of TOS1 and has the
+   positional and keyword attributes required by *count* and TOS, push a tuple
+   of extracted attributes.  Otherwise, push ``None``.
 
    .. versionadded:: 3.10
 
+   .. versionchanged:: 3.11
+      Previously, this instruction also pushed a boolean value indicating
+      success (``True``) or failure (``False``).
+
 .. opcode:: GEN_START (kind)
 
     Pops TOS. If TOS was not ``None``, raises an exception. The ``kind``
@@ -1301,6 +1299,14 @@ All of the following opcodes use their arguments.
    .. versionadded:: 3.10
 
 
+.. opcode:: COPY (i)
+
+   Push the *i*-th item to the top of the stack. The item is not removed from its
+   original location.
+
+   .. versionadded:: 3.11
+
+
 .. opcode:: HAVE_ARGUMENT
 
    This is not really an opcode.  It identifies the dividing line between

Doc/whatsnew/3.11.rst

@@ -308,6 +308,16 @@ CPython bytecode changes
   fashion as :opcode:`CALL_METHOD`, but also supports keyword arguments.  Works
   in tandem with :opcode:`LOAD_METHOD`.
 
+* Removed ``COPY_DICT_WITHOUT_KEYS``.
+
+* :opcode:`MATCH_CLASS` and :opcode:`MATCH_KEYS` no longer push an additional
+  boolean value indicating whether the match succeeded or failed. Instead, they
+  indicate failure with :const:`None` (where a tuple of extracted values would
+  otherwise be).
+
+* Added :opcode:`COPY`, which pushes the *i*-th item to the top of the stack.
+  The item is not removed from its original location.
+
 
 Deprecated
 ==========

Include/opcode.h

@@ -364,6 +364,8 @@ def _write_atomic(path, data, mode=0o666):
 #     Python 3.11a1 3459 (PEP 657: add end line numbers and column offsets for instructions)
 #     Python 3.11a1 3460 (Add co_qualname field to PyCodeObject bpo-44530)
 #     Python 3.11a1 3461 (JUMP_ABSOLUTE must jump backwards)
+#     Python 3.11a2 3462 (bpo-44511: remove COPY_DICT_WITHOUT_KEYS, change
+#                         MATCH_CLASS and MATCH_KEYS, and add COPY)
 
 #
 # MAGIC must change whenever the bytecode emitted by the compiler may no
@@ -373,7 +375,7 @@ def _write_atomic(path, data, mode=0o666):
 # Whenever MAGIC_NUMBER is changed, the ranges in the magic_values array
 # in PC/launcher.c must also be updated.
 
-MAGIC_NUMBER = (3461).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3462).to_bytes(2, 'little') + b'\r\n'
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 
 _PYCACHE = '__pycache__'

Lib/importlib/_bootstrap_external.py

@@ -85,7 +85,7 @@ def jabs_op(name, op):
 def_op('MATCH_MAPPING', 31)
 def_op('MATCH_SEQUENCE', 32)
 def_op('MATCH_KEYS', 33)
-def_op('COPY_DICT_WITHOUT_KEYS', 34)
+
 def_op('PUSH_EXC_INFO', 35)
 
 def_op('POP_EXCEPT_AND_RERAISE', 37)
@@ -165,7 +165,7 @@ def jabs_op(name, op):
 def_op('IS_OP', 117)
 def_op('CONTAINS_OP', 118)
 def_op('RERAISE', 119)
-
+def_op('COPY', 120)
 jabs_op('JUMP_IF_NOT_EXC_MATCH', 121)
 
 def_op('LOAD_FAST', 124)        # Local variable number

Lib/opcode.py

@@ -0,0 +1 @@
+Improve the generated bytecode for class and mapping patterns.

Misc/NEWS.d/next/Core and Builtins/2021-06-26-16-55-08.bpo-44511.k8sMvV.rst

@@ -4143,25 +4143,30 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
         }
 
         TARGET(MATCH_CLASS) {
-            // Pop TOS. On success, set TOS to True and TOS1 to a tuple of
-            // attributes. On failure, set TOS to False.
+            // Pop TOS and TOS1. Set TOS to a tuple of attributes on success, or
+            // None on failure.
             PyObject *names = POP();
-            PyObject *type = TOP();
-            PyObject *subject = SECOND();
+            PyObject *type = POP();
+            PyObject *subject = TOP();
             assert(PyTuple_CheckExact(names));
             PyObject *attrs = match_class(tstate, subject, type, oparg, names);
             Py_DECREF(names);
+            Py_DECREF(type);
             if (attrs) {
                 // Success!
                 assert(PyTuple_CheckExact(attrs));
-                Py_DECREF(subject);
-                SET_SECOND(attrs);
+                SET_TOP(attrs);
             }
             else if (_PyErr_Occurred(tstate)) {
+                // Error!
                 goto error;
             }
-            Py_DECREF(type);
-            SET_TOP(PyBool_FromLong(!!attrs));
+            else {
+                // Failure!
+                Py_INCREF(Py_None);
+                SET_TOP(Py_None);
+            }
+            Py_DECREF(subject);
             DISPATCH();
         }
 
@@ -4171,6 +4176,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             PyObject *res = match ? Py_True : Py_False;
             Py_INCREF(res);
             PUSH(res);
+            PREDICT(POP_JUMP_IF_FALSE);
             DISPATCH();
         }
 
@@ -4180,53 +4186,19 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             PyObject *res = match ? Py_True : Py_False;
             Py_INCREF(res);
             PUSH(res);
+            PREDICT(POP_JUMP_IF_FALSE);
             DISPATCH();
         }
 
         TARGET(MATCH_KEYS) {
-            // On successful match for all keys, PUSH(values) and PUSH(True).
-            // Otherwise, PUSH(None) and PUSH(False).
+            // On successful match, PUSH(values). Otherwise, PUSH(None).
             PyObject *keys = TOP();
             PyObject *subject = SECOND();
             PyObject *values_or_none = match_keys(tstate, subject, keys);
             if (values_or_none == NULL) {
                 goto error;
             }
             PUSH(values_or_none);
-            if (Py_IsNone(values_or_none)) {
-                Py_INCREF(Py_False);
-                PUSH(Py_False);
-                DISPATCH();
-            }
-            assert(PyTuple_CheckExact(values_or_none));
-            Py_INCREF(Py_True);
-            PUSH(Py_True);
-            DISPATCH();
-        }
-
-        TARGET(COPY_DICT_WITHOUT_KEYS) {
-            // rest = dict(TOS1)
-            // for key in TOS:
-            //     del rest[key]
-            // SET_TOP(rest)
-            PyObject *keys = TOP();
-            PyObject *subject = SECOND();
-            PyObject *rest = PyDict_New();
-            if (rest == NULL || PyDict_Update(rest, subject)) {
-                Py_XDECREF(rest);
-                goto error;
-            }
-            // This may seem a bit inefficient, but keys is rarely big enough to
-            // actually impact runtime.
-            assert(PyTuple_CheckExact(keys));
-            for (Py_ssize_t i = 0; i < PyTuple_GET_SIZE(keys); i++) {
-                if (PyDict_DelItem(rest, PyTuple_GET_ITEM(keys, i))) {
-                    Py_DECREF(rest);
-                    goto error;
-                }
-            }
-            Py_DECREF(keys);
-            SET_TOP(rest);
             DISPATCH();
         }
 
@@ -5027,6 +4999,14 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             DISPATCH();
         }
 
+        TARGET(COPY) {
+            assert(oparg != 0);
+            PyObject *peek = PEEK(oparg);
+            Py_INCREF(peek);
+            PUSH(peek);
+            DISPATCH();
+        }
+
         TARGET(EXTENDED_ARG) {
             int oldoparg = oparg;
             NEXTOPARG();