Give meaning to the oparg for BUILD_MAP: estimated size of the dictionary.

Allows dictionaries to be pre-sized (upto 255 elements) saving time lost
to re-sizes with their attendant mallocs and re-insertions.

Has zero effect on small dictionaries (5 elements or fewer), a slight
benefit for dicts upto 22 elements (because they had to resize once
anyway), and more benefit for dicts upto 255 elements (saving multiple
resizes during the build-up and reducing the number of collisions on
the first insertions).  Beyond 255 elements, there is no addional benefit.

Include/dictobject.h

@@ -110,6 +110,7 @@ PyAPI_FUNC(Py_ssize_t) PyDict_Size(PyObject *mp);
 PyAPI_FUNC(PyObject *) PyDict_Copy(PyObject *mp);
 PyAPI_FUNC(int) PyDict_Contains(PyObject *mp, PyObject *key);
 PyAPI_FUNC(int) _PyDict_Contains(PyObject *mp, PyObject *key, long hash);
+PyAPI_FUNC(PyObject *) _PyDict_NewPresized(Py_ssize_t minused);
 
 /* PyDict_Update(mp, other) is equivalent to PyDict_Merge(mp, other, 1). */
 PyAPI_FUNC(int) PyDict_Update(PyObject *mp, PyObject *other);

Lib/opcode.py

@@ -139,7 +139,7 @@ def jabs_op(name, op):
 name_op('LOAD_NAME', 101)       # Index in name list
 def_op('BUILD_TUPLE', 102)      # Number of tuple items
 def_op('BUILD_LIST', 103)       # Number of list items
-def_op('BUILD_MAP', 104)        # Always zero for now
+def_op('BUILD_MAP', 104)        # Number of dict entries (upto 255)
 name_op('LOAD_ATTR', 105)       # Index in name list
 def_op('COMPARE_OP', 106)       # Comparison operator
 hascompare.append(106)

Misc/NEWS

@@ -12,6 +12,10 @@ What's New in Python 2.6 alpha 1?
 Core and builtins
 -----------------
 
+- Compiler now generates simpler and faster code for dictionary literals.
+  The oparg for BUILD_MAP now indicates an estimated dictionary size.
+  There is a new opcode, STORE_MAP, for adding entries to the dictionary.
+
 - Issue #1638: %zd configure test fails on Linux
 
 - Issue #1620: New property decorator syntax was modifying the decorator

Objects/dictobject.c

@@ -549,6 +549,23 @@ dictresize(PyDictObject *mp, Py_ssize_t minused)
 	return 0;
 }
 
+/* Create a new dictionary pre-sized to hold an estimated number of elements.
+   Underestimates are okay because the dictionary will resize as necessary.
+   Overestimates just mean the dictionary will be more sparse than usual.
+*/
+
+PyObject *
+_PyDict_NewPresized(Py_ssize_t minused)
+{
+	PyObject *op = PyDict_New();
+
+	if (minused>5 && op != NULL && dictresize((PyDictObject *)op, minused) == -1) {
+		Py_DECREF(op);
+		return NULL;
+	}
+	return op;
+}
+
 /* Note that, for historical reasons, PyDict_GetItem() suppresses all errors
  * that may occur (originally dicts supported only string keys, and exceptions
  * weren't possible).  So, while the original intent was that a NULL return

Python/ceval.c

@@ -1997,7 +1997,7 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
 			break;
 
 		case BUILD_MAP:
-			x = PyDict_New();
+			x = _PyDict_NewPresized((Py_ssize_t)oparg);
 			PUSH(x);
 			if (x != NULL) continue;
 			break;

Python/compile.c

@@ -2922,11 +2922,9 @@ compiler_visit_expr(struct compiler *c, expr_ty e)
 	case IfExp_kind:
 		return compiler_ifexp(c, e);
 	case Dict_kind:
-		/* XXX get rid of arg? */
-		ADDOP_I(c, BUILD_MAP, 0);
 		n = asdl_seq_LEN(e->v.Dict.values);
-		/* We must arrange things just right for STORE_SUBSCR.
+		ADDOP_I(c, BUILD_MAP, (n>255 ? 255 : n));
-		   It wants the stack to look like (value) (dict) (key) */
+		n = asdl_seq_LEN(e->v.Dict.values);
 		for (i = 0; i < n; i++) {
 			VISIT(c, expr, 
 				(expr_ty)asdl_seq_GET(e->v.Dict.values, i));