Issue #2183: Simplify and optimize bytecode for list comprehensions.

git-svn-id: http://svn.python.org/projects/python/trunk@67818 6015fed2-1504-0410-9fe1-9d1591cc4771

Author: antoine.pitrou

Doc/library/dis.rst

@@ -463,9 +463,11 @@ Miscellaneous opcodes.
    address to jump to (which should be a ``FOR_ITER`` instruction).
 
 
-.. opcode:: LIST_APPEND ()
+.. opcode:: LIST_APPEND (i)
 
-   Calls ``list.append(TOS1, TOS)``.  Used to implement list comprehensions.
+   Calls ``list.append(TOS[-i], TOS)``.  Used to implement list comprehensions.
+   While the appended value is popped off, the list object remains on the
+   stack so that it is available for further iterations of the loop.
 
 
 .. opcode:: LOAD_LOCALS ()

Include/opcode.h

@@ -22,7 +22,6 @@ extern "C" {
 
 #define UNARY_INVERT	15
 
-#define LIST_APPEND	18
 #define BINARY_POWER	19
 
 #define BINARY_MULTIPLY	20
@@ -89,6 +88,7 @@ extern "C" {
 #define DELETE_NAME	91	/* "" */
 #define UNPACK_SEQUENCE	92	/* Number of sequence items */
 #define FOR_ITER	93
+#define LIST_APPEND	94
 
 #define STORE_ATTR	95	/* Index in name list */
 #define DELETE_ATTR	96	/* "" */

Lib/compiler/pycodegen.py

@@ -570,16 +570,10 @@ def visitCompare(self, node):
             self.nextBlock(end)
 
     # list comprehensions
-    __list_count = 0
-
     def visitListComp(self, node):
         self.set_lineno(node)
         # setup list
-        tmpname = "$list%d" % self.__list_count
-        self.__list_count = self.__list_count + 1
         self.emit('BUILD_LIST', 0)
-        self.emit('DUP_TOP')
-        self._implicitNameOp('STORE', tmpname)
 
         stack = []
         for i, for_ in zip(range(len(node.quals)), node.quals):
@@ -591,9 +585,8 @@ def visitListComp(self, node):
                 self.visit(if_, cont)
             stack.insert(0, (start, cont, anchor))
 
-        self._implicitNameOp('LOAD', tmpname)
         self.visit(node.expr)
-        self.emit('LIST_APPEND')
+        self.emit('LIST_APPEND', len(node.quals) + 1)
 
         for start, cont, anchor in stack:
             if cont:
@@ -604,9 +597,6 @@ def visitListComp(self, node):
                 self.nextBlock(skip_one)
             self.emit('JUMP_ABSOLUTE', start)
             self.startBlock(anchor)
-        self._implicitNameOp('DELETE', tmpname)
-
-        self.__list_count = self.__list_count - 1
 
     def visitListCompFor(self, node):
         start = self.newBlock()

Lib/opcode.py

@@ -58,7 +58,6 @@ def jabs_op(name, op):
 
 def_op('UNARY_INVERT', 15)
 
-def_op('LIST_APPEND', 18)
 def_op('BINARY_POWER', 19)
 def_op('BINARY_MULTIPLY', 20)
 def_op('BINARY_DIVIDE', 21)
@@ -128,7 +127,7 @@ def jabs_op(name, op):
 name_op('DELETE_NAME', 91)      # ""
 def_op('UNPACK_SEQUENCE', 92)   # Number of tuple items
 jrel_op('FOR_ITER', 93)
-
+def_op('LIST_APPEND', 94)
 name_op('STORE_ATTR', 95)       # Index in name list
 name_op('DELETE_ATTR', 96)      # ""
 name_op('STORE_GLOBAL', 97)     # ""

Lib/test/test_dis.py

@@ -54,29 +54,25 @@ def bug1333982(x=[]):
 
 dis_bug1333982 = """\
  %-4d         0 LOAD_CONST               1 (0)
-              3 JUMP_IF_TRUE            41 (to 47)
+              3 JUMP_IF_TRUE            33 (to 39)
               6 POP_TOP
               7 LOAD_GLOBAL              0 (AssertionError)
              10 BUILD_LIST               0
-             13 DUP_TOP
-             14 STORE_FAST               1 (_[1])
-             17 LOAD_FAST                0 (x)
-             20 GET_ITER
-        >>   21 FOR_ITER                13 (to 37)
-             24 STORE_FAST               2 (s)
-             27 LOAD_FAST                1 (_[1])
-             30 LOAD_FAST                2 (s)
-             33 LIST_APPEND
-             34 JUMP_ABSOLUTE           21
-        >>   37 DELETE_FAST              1 (_[1])
-
- %-4d        40 LOAD_CONST               2 (1)
-             43 BINARY_ADD
-             44 RAISE_VARARGS            2
-        >>   47 POP_TOP
-
- %-4d        48 LOAD_CONST               0 (None)
-             51 RETURN_VALUE
+             13 LOAD_FAST                0 (x)
+             16 GET_ITER
+        >>   17 FOR_ITER                12 (to 32)
+             20 STORE_FAST               1 (s)
+             23 LOAD_FAST                1 (s)
+             26 LIST_APPEND              2
+             29 JUMP_ABSOLUTE           17
+
+ %-4d   >>   32 LOAD_CONST               2 (1)
+             35 BINARY_ADD
+             36 RAISE_VARARGS            2
+        >>   39 POP_TOP
+
+ %-4d        40 LOAD_CONST               0 (None)
+             43 RETURN_VALUE
 """%(bug1333982.func_code.co_firstlineno + 1,
      bug1333982.func_code.co_firstlineno + 2,
      bug1333982.func_code.co_firstlineno + 3)

Misc/NEWS

@@ -12,6 +12,9 @@ What's New in Python 2.7 alpha 1
 Core and Builtins
 -----------------
 
+- Issue #2183: Simplify and optimize bytecode for list comprehensions.
+  Original patch by Neal Norwitz.
+
 - Issue #4597: Fixed exception handling when the __exit__ function of a
   context manager returns a value that cannot be converted to a bool.
 

Python/ceval.c

@@ -1294,9 +1294,8 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
 
 		case LIST_APPEND:
 			w = POP();
-			v = POP();
+			v = stack_pointer[-oparg];
 			err = PyList_Append(v, w);
-			Py_DECREF(v);
 			Py_DECREF(w);
 			if (err == 0) {
 				PREDICT(JUMP_ABSOLUTE);

Python/compile.c

@@ -693,7 +693,7 @@ opcode_stack_effect(int opcode, int oparg)
 			return 0;
 
 		case LIST_APPEND:
-			return -2;
+			return -1;
 
 		case BINARY_POWER:
 		case BINARY_MULTIPLY:
@@ -2599,9 +2599,8 @@ compiler_call(struct compiler *c, expr_ty e)
 }
 
 static int
-compiler_listcomp_generator(struct compiler *c, PyObject *tmpname,
-			    asdl_seq *generators, int gen_index, 
-			    expr_ty elt)
+compiler_listcomp_generator(struct compiler *c, asdl_seq *generators,
+			    int gen_index, expr_ty elt)
 {
 	/* generate code for the iterator, then each of the ifs,
 	   and then write to the element */
@@ -2638,16 +2637,13 @@ compiler_listcomp_generator(struct compiler *c, PyObject *tmpname,
 	} 
 
 	if (++gen_index < asdl_seq_LEN(generators))
-	    if (!compiler_listcomp_generator(c, tmpname, 
-					     generators, gen_index, elt))
+	    if (!compiler_listcomp_generator(c, generators, gen_index, elt))
 		return 0;
 
 	/* only append after the last for generator */
 	if (gen_index >= asdl_seq_LEN(generators)) {
-	    if (!compiler_nameop(c, tmpname, Load))
-		return 0;
 	    VISIT(c, expr, elt);
-	    ADDOP(c, LIST_APPEND);
+	    ADDOP_I(c, LIST_APPEND, gen_index+1);
 
 	    compiler_use_next_block(c, skip);
 	}
@@ -2659,32 +2655,17 @@ compiler_listcomp_generator(struct compiler *c, PyObject *tmpname,
 	} 
 	ADDOP_JABS(c, JUMP_ABSOLUTE, start);
 	compiler_use_next_block(c, anchor);
-	/* delete the temporary list name added to locals */
-	if (gen_index == 1)
-	    if (!compiler_nameop(c, tmpname, Del))
-		return 0;
 
 	return 1;
 }
 
 static int
 compiler_listcomp(struct compiler *c, expr_ty e)
 {
-	identifier tmp;
-	int rc = 0;
-	asdl_seq *generators = e->v.ListComp.generators;
-
 	assert(e->kind == ListComp_kind);
-	tmp = compiler_new_tmpname(c);
-	if (!tmp)
-		return 0;
 	ADDOP_I(c, BUILD_LIST, 0);
-	ADDOP(c, DUP_TOP);
-	if (compiler_nameop(c, tmp, Store))
-	    rc = compiler_listcomp_generator(c, tmp, generators, 0, 
-					     e->v.ListComp.elt);
-	Py_DECREF(tmp);
-	return rc;
+	return compiler_listcomp_generator(c, e->v.ListComp.generators, 0,
+					   e->v.ListComp.elt);
 }
 
 static int

Python/import.c

@@ -73,9 +73,10 @@ extern time_t PyOS_GetLastModificationTime(char *, FILE *);
        Python 2.5c2: 62131 (fix wrong code: for x, in ... in listcomp/genexp)
        Python 2.6a0: 62151 (peephole optimizations and STORE_MAP opcode)
        Python 2.6a1: 62161 (WITH_CLEANUP optimization)
+       Python 2.7a0: 62171 (optimize list comprehensions/change LIST_APPEND)
 .
 */
-#define MAGIC (62161 | ((long)'\r'<<16) | ((long)'\n'<<24))
+#define MAGIC (62171 | ((long)'\r'<<16) | ((long)'\n'<<24))
 
 /* Magic word as global; note that _PyImport_Init() can change the
    value of this global to accommodate for alterations of how the