SF patch #872326: Generator expression implementation

(Code contributed by Jiwon Seo.)

The documentation portion of the patch is being re-worked and will be
checked-in soon.  Likewise, PEP 289 will be updated to reflect Guido's
rationale for the design decisions on binding behavior (as described in
in his patch comments and in discussions on python-dev).

The test file, test_genexps.py, is written in doctest format and is
meant to exercise all aspects of the the patch.  Further additions are
welcome from everyone.  Please stress test this new feature as much as
possible before the alpha release.

Doc/whatsnew/whatsnew24.tex

@@ -88,6 +88,84 @@ Greg Wilson and ultimately implemented by Raymond Hettinger.}
 
 XXX write this.
 
+%======================================================================
+\section{PEP 229: Generator Expressions}
+
+Generator expressions create in-line generators using a syntax similar
+to list comprehensions but with parenthesis instead of the surrounding
+brackets.
+
+Genexps allow simple generators to be constructed without a separate function
+definition.  Writing:
+       
+\begin{verbatim}
+       g = (tgtexp  for var1 in exp1  for var2 in exp2 if exp3)
+\end{verbatim}
+       
+is equivalent to:
+
+\begin{verbatim}
+    def _generator(exp):
+        for var1 in exp:
+            for var2 in exp2:
+                if exp3:
+                    yield tgtexp
+    g = _generator(exp1)
+    del _generator
+\end{verbatim}
+
+The advantage over full generator definitions is in economy of
+expression.  Their advantage over list comprehensions is in saving
+memory by creating data only when it is needed rather than forming
+a whole list is memory all at once.  Applications using memory
+friendly generator expressions may scale-up to high volumes of data
+more readily than with list comprehensions.
+
+Generator expressions are intended to be used inside functions
+such as \function{sum()}, \function{min()}, \function{set()}, and
+\function{dict()}.  These functions consume their data all at once
+and would not benefit from having a full list instead of a generator
+an input:
+
+\begin{verbatim}
+>>> sum(i*i for i in range(10))
+285
+
+>>> sorted(set(i*i for i in xrange(-10, 11)))
+[0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100]
+
+>>> words = "Adam apple baker Bill Nancy NASA nut".split()
+>>> dict((word.lower(), word) for word in words)
+{'apple': 'apple', 'baker': 'baker', 'bill': 'Bill', 'nasa': 'NASA',
+ 'adam': 'Adam', 'nancy': 'Nancy', 'nut': 'nut'}
+
+>>> xvec = [10, 20, 30]
+>>> yvec = [7, 5, 3]
+>>> sum(x*y for x,y in itertools.izip(xvec, yvec))     # dot product
+260
+
+\end{verbatim}     
+
+These examples show the intended use for generator expressions
+in situations where the values get consumed immediately after the
+generator is created.  In these situations, they operate like
+memory efficient versions of list comprehensions.
+
+For more complex uses of generators, it is strongly recommended that
+the traditional full generator definitions be used instead.  In a
+generator expression, the first for-loop expression is evaluated
+as soon as the expression is defined while the other expressions do
+not get evaluated until the generator is run.  This nuance is never
+an issue when the generator is used immediately.  If it is not used
+right away, then it is better to write a full generator definition
+which more clearly reveals when the expressions are evaluated and is
+more obvious about the visibility and lifetime of its looping variables.
+
+\begin{seealso}
+\seepep{289}{Generator Expressions}{Proposed by Raymond Hettinger and
+implemented by Jiwon Seo with early efforts steered by Hye-Shik Chang.}
+\end{seealso}
+
 %======================================================================
 \section{PEP 322: Reverse Iteration}
 

Grammar/Grammar

@@ -80,8 +80,9 @@ arith_expr: term (('+'|'-') term)*
 term: factor (('*'|'/'|'%'|'//') factor)*
 factor: ('+'|'-'|'~') factor | power
 power: atom trailer* ['**' factor]
-atom: '(' [testlist] ')' | '[' [listmaker] ']' | '{' [dictmaker] '}' | '`' testlist1 '`' | NAME | NUMBER | STRING+
+atom: '(' [testlist_gexp] ')' | '[' [listmaker] ']' | '{' [dictmaker] '}' | '`' testlist1 '`' | NAME | NUMBER | STRING+
 listmaker: test ( list_for | (',' test)* [','] )
+testlist_gexp: test ( gen_for | (',' test)* [','] )
 lambdef: 'lambda' [varargslist] ':' test
 trailer: '(' [arglist] ')' | '[' subscriptlist ']' | '.' NAME
 subscriptlist: subscript (',' subscript)* [',']
@@ -95,12 +96,16 @@ dictmaker: test ':' test (',' test ':' test)* [',']
 classdef: 'class' NAME ['(' testlist ')'] ':' suite
 
 arglist: (argument ',')* (argument [',']| '*' test [',' '**' test] | '**' test)
-argument: [test '='] test	# Really [keyword '='] test
+argument: [test '='] test [gen_for] # Really [keyword '='] test
 
 list_iter: list_for | list_if
 list_for: 'for' exprlist 'in' testlist_safe [list_iter]
 list_if: 'if' test [list_iter]
 
+gen_iter: gen_for | gen_if
+gen_for: 'for' exprlist 'in' test [gen_iter]
+gen_if: 'if' test [gen_iter]
+
 testlist1: test (',' test)*
 
 # not used in grammar, but may appear in "node" passed from Parser to Compiler

Include/graminit.h

@@ -49,20 +49,24 @@
 #define power 304
 #define atom 305
 #define listmaker 306
-#define lambdef 307
-#define trailer 308
-#define subscriptlist 309
-#define subscript 310
-#define sliceop 311
-#define exprlist 312
-#define testlist 313
-#define testlist_safe 314
-#define dictmaker 315
-#define classdef 316
-#define arglist 317
-#define argument 318
-#define list_iter 319
-#define list_for 320
-#define list_if 321
-#define testlist1 322
-#define encoding_decl 323
+#define testlist_gexp 307
+#define lambdef 308
+#define trailer 309
+#define subscriptlist 310
+#define subscript 311
+#define sliceop 312
+#define exprlist 313
+#define testlist 314
+#define testlist_safe 315
+#define dictmaker 316
+#define classdef 317
+#define arglist 318
+#define argument 319
+#define list_iter 320
+#define list_for 321
+#define list_if 322
+#define gen_iter 323
+#define gen_for 324
+#define gen_if 325
+#define testlist1 326
+#define encoding_decl 327

Include/symtable.h

@@ -46,7 +46,7 @@ typedef struct _symtable_entry {
 				    including free refs to globals */
 	int ste_generator;       /* true if namespace is a generator */
 	int ste_opt_lineno;      /* lineno of last exec or import * */
-	int ste_tmpname;          /* temporary name counter */
+	int ste_tmpname;         /* temporary name counter */
 	struct symtable *ste_table;
 } PySymtableEntryObject;
 
@@ -93,6 +93,9 @@ PyAPI_FUNC(void) PySymtable_Free(struct symtable *);
 #define OPT_EXEC 2
 #define OPT_BARE_EXEC 4
 
+#define GENERATOR 1
+#define GENERATOR_EXPRESSION 2
+
 #ifdef __cplusplus
 }
 #endif

Lib/compiler/ast.py

@@ -1236,6 +1236,82 @@ class ListCompFor(Node):
     def __repr__(self):
         return "ListCompFor(%s, %s, %s)" % (repr(self.assign), repr(self.list), repr(self.ifs))
 
+class GenExpr(Node):
+    nodes["genexpr"] = "GenExpr"
+    def __init__(self, code):
+        self.code = code
+        self.argnames = ['[outmost-iterable]']
+        self.varargs = self.kwargs = None
+
+    def getChildren(self):
+        return self.code,
+
+    def getChildNodes(self):
+        return self.code,
+
+    def __repr__(self):
+        return "GenExpr(%s)" % (repr(self.code),)
+
+class GenExprInner(Node):
+    nodes["genexprinner"] = "GenExprInner"
+    def __init__(self, expr, quals):
+        self.expr = expr
+        self.quals = quals
+
+    def getChildren(self):
+        children = []
+        children.append(self.expr)
+        children.extend(flatten(self.quals))
+        return tuple(children)
+
+    def getChildNodes(self):
+        nodelist = []
+        nodelist.append(self.expr)
+        nodelist.extend(flatten_nodes(self.quals))
+        return tuple(nodelist)
+
+    def __repr__(self):
+        return "GenExprInner(%s, %s)" % (repr(self.expr), repr(self.quals))
+
+class GenExprFor(Node):
+    nodes["genexprfor"] = "GenExprFor"
+    def __init__(self, assign, iter, ifs):
+        self.assign = assign
+        self.iter = iter
+        self.ifs = ifs
+        self.is_outmost = False
+
+    def getChildren(self):
+        children = []
+        children.append(self.assign)
+        children.append(self.iter)
+        children.extend(flatten(self.ifs))
+        return tuple(children)
+
+    def getChildNodes(self):
+        nodelist = []
+        nodelist.append(self.assign)
+        nodelist.append(self.iter)
+        nodelist.extend(flatten_nodes(self.ifs))
+        return tuple(nodelist)
+
+    def __repr__(self):
+        return "GenExprFor(%s, %s, %s)" % (repr(self.assign), repr(self.iter), repr(self.ifs))
+
+class GenExprIf(Node):
+    nodes["genexprif"] = "GenExprIf"
+    def __init__(self, test):
+        self.test = test
+
+    def getChildren(self):
+        return self.test,
+
+    def getChildNodes(self):
+        return self.test,
+
+    def __repr__(self):
+        return "GenExprIf(%s)" % (repr(self.test),)
+
 klasses = globals()
 for k in nodes.keys():
     nodes[k] = klasses[nodes[k]]

Lib/compiler/pycodegen.py

@@ -619,6 +619,79 @@ class CodeGenerator:
         self.newBlock()
         self.emit('POP_TOP')
 
+    def visitGenExpr(self, node):
+        gen = GenExprCodeGenerator(node, self.scopes, self.class_name,
+                                   self.get_module())
+        walk(node.code, gen)
+        gen.finish()
+        self.set_lineno(node)
+        frees = gen.scope.get_free_vars()
+        if frees:
+            for name in frees:
+                self.emit('LOAD_CLOSURE', name)
+            self.emit('LOAD_CONST', gen)
+            self.emit('MAKE_CLOSURE', 0)
+        else:
+            self.emit('LOAD_CONST', gen)
+            self.emit('MAKE_FUNCTION', 0)
+
+        # precomputation of outmost iterable
+        self.visit(node.code.quals[0].iter)
+        self.emit('GET_ITER')
+        self.emit('CALL_FUNCTION', 1)
+
+    def visitGenExprInner(self, node):
+        self.set_lineno(node)
+        # setup list
+
+        stack = []
+        for i, for_ in zip(range(len(node.quals)), node.quals):
+            start, anchor = self.visit(for_)
+            cont = None
+            for if_ in for_.ifs:
+                if cont is None:
+                    cont = self.newBlock()
+                self.visit(if_, cont)
+            stack.insert(0, (start, cont, anchor))
+
+        self.visit(node.expr)
+        self.emit('YIELD_VALUE')
+
+        for start, cont, anchor in stack:
+            if cont:
+                skip_one = self.newBlock()
+                self.emit('JUMP_FORWARD', skip_one)
+                self.startBlock(cont)
+                self.emit('POP_TOP')
+                self.nextBlock(skip_one)
+            self.emit('JUMP_ABSOLUTE', start)
+            self.startBlock(anchor)
+        self.emit('LOAD_CONST', None)
+
+    def visitGenExprFor(self, node):
+        start = self.newBlock()
+        anchor = self.newBlock()
+        
+        if node.is_outmost:
+            self.loadName('[outmost-iterable]')
+        else:
+            self.visit(node.iter)
+            self.emit('GET_ITER')
+
+        self.nextBlock(start)
+        self.set_lineno(node, force=True)
+        self.emit('FOR_ITER', anchor)
+        self.nextBlock()
+        self.visit(node.assign)
+        return start, anchor
+
+    def visitGenExprIf(self, node, branch):
+        self.set_lineno(node, force=True)
+        self.visit(node.test)
+        self.emit('JUMP_IF_FALSE', branch)
+        self.newBlock()
+        self.emit('POP_TOP')
+
     # exception related
 
     def visitAssert(self, node):
@@ -1199,6 +1272,7 @@ class AbstractFunctionCode:
             klass.lambdaCount = klass.lambdaCount + 1
         else:
             name = func.name
+
         args, hasTupleArg = generateArgList(func.argnames)
         self.graph = pyassem.PyFlowGraph(name, func.filename, args,
                                          optimized=1)
@@ -1263,6 +1337,21 @@ class FunctionCodeGenerator(NestedScopeMixin, AbstractFunctionCode,
         if self.scope.generator is not None:
             self.graph.setFlag(CO_GENERATOR)
 
+class GenExprCodeGenerator(NestedScopeMixin, AbstractFunctionCode,
+                           CodeGenerator):
+    super_init = CodeGenerator.__init__ # call be other init
+    scopes = None
+
+    __super_init = AbstractFunctionCode.__init__
+
+    def __init__(self, gexp, scopes, class_name, mod):
+        self.scopes = scopes
+        self.scope = scopes[gexp]
+        self.__super_init(gexp, scopes, 1, class_name, mod)
+        self.graph.setFreeVars(self.scope.get_free_vars())
+        self.graph.setCellVars(self.scope.get_cell_vars())
+        self.graph.setFlag(CO_GENERATOR)
+
 class AbstractClassCode:
 
     def __init__(self, klass, scopes, module):

Lib/compiler/symbols.py

@@ -179,6 +179,21 @@ class ModuleScope(Scope):
 class FunctionScope(Scope):
     pass
 
+class GenExprScope(Scope):
+    __super_init = Scope.__init__
+
+    __counter = 1
+
+    def __init__(self, module, klass=None):
+        i = self.__counter
+        self.__counter += 1
+        self.__super_init("generator expression<%d>"%i, module, klass)
+        self.add_param('[outmost-iterable]')
+
+    def get_names(self):
+        keys = Scope.get_names()
+        return keys
+
 class LambdaScope(FunctionScope):
     __super_init = Scope.__init__
 
@@ -220,6 +235,32 @@ class SymbolVisitor:
         self.visit(node.code, scope)
         self.handle_free_vars(scope, parent)
 
+    def visitGenExpr(self, node, parent):
+        scope = GenExprScope(self.module, self.klass);
+        if parent.nested or isinstance(parent, FunctionScope) \
+                or isinstance(parent, GenExprScope):
+            scope.nested = 1
+
+        self.scopes[node] = scope
+        self.visit(node.code, scope)
+
+        self.handle_free_vars(scope, parent)
+
+    def visitGenExprInner(self, node, scope):
+        for genfor in node.quals:
+            self.visit(genfor, scope)
+
+        self.visit(node.expr, scope)
+
+    def visitGenExprFor(self, node, scope):
+        self.visit(node.assign, scope, 1)
+        self.visit(node.iter, scope)
+        for if_ in node.ifs:
+            self.visit(if_, scope)
+
+    def visitGenExprIf(self, node, scope):
+        self.visit(node.test, scope)
+        
     def visitLambda(self, node, parent, assign=0):
         # Lambda is an expression, so it could appear in an expression
         # context where assign is passed.  The transformer should catch

Lib/compiler/transformer.py

@@ -534,6 +534,12 @@ class Transformer:
     testlist1 = testlist
     exprlist = testlist
 
+    def testlist_gexp(self, nodelist):
+        if len(nodelist) == 2 and nodelist[1][0] == symbol.gen_for:
+            test = self.com_node(nodelist[0])
+            return self.com_generator_expression(test, nodelist[1])
+        return self.testlist(nodelist)
+
     def test(self, nodelist):
         # and_test ('or' and_test)* | lambdef
         if len(nodelist) == 1 and nodelist[0][0] == symbol.lambdef:
@@ -1085,6 +1091,48 @@ class Transformer:
                 values.append(self.com_node(nodelist[i]))
             return List(values)
 
+    if hasattr(symbol, 'gen_for'):
+        def com_generator_expression(self, expr, node):
+            # gen_iter: gen_for | gen_if
+            # gen_for: 'for' exprlist 'in' test [gen_iter]
+            # gen_if: 'if' test [gen_iter]
+
+            lineno = node[1][2]
+            fors = []
+            while node:
+                t = node[1][1]
+                if t == 'for':
+                    assignNode = self.com_assign(node[2], OP_ASSIGN)
+                    genNode = self.com_node(node[4])
+                    newfor = GenExprFor(assignNode, genNode, [])
+                    newfor.lineno = node[1][2]
+                    fors.append(newfor)
+                    if (len(node)) == 5:
+                        node = None
+                    else:
+                        node = self.com_gen_iter(node[5])
+                elif t == 'if':
+                    test = self.com_node(node[2])
+                    newif = GenExprIf(test)
+                    newif.lineno = node[1][2]
+                    newfor.ifs.append(newif)
+                    if len(node) == 3:
+                        node = None
+                    else:
+                        node = self.com_gen_iter(node[3])
+                else:
+                    raise SyntaxError, \
+                            ("unexpected generator expression element: %s %d"
+                             % (node, lineno))
+            fors[0].is_outmost = True
+            n = GenExpr(GenExprInner(expr, fors))
+            n.lineno = lineno
+            return n
+
+        def com_gen_iter(self, node):
+            assert node[0] == symbol.gen_iter
+            return node[1]
+
     def com_dictmaker(self, nodelist):
         # dictmaker: test ':' test (',' test ':' value)* [',']
         items = []
@@ -1122,6 +1170,8 @@ class Transformer:
             if node[0] == token.STAR or node[0] == token.DOUBLESTAR:
                 break
             kw, result = self.com_argument(node, kw)
+            if len_nodelist != 2 and isinstance(result, GenExpr):
+                raise SyntaxError, 'generator expression needs parenthesis'
             args.append(result)
         else:
             # No broken by star arg, so skip the last one we processed.
@@ -1148,6 +1198,9 @@ class Transformer:
         return CallFunc(primaryNode, args, star_node, dstar_node)
 
     def com_argument(self, nodelist, kw):
+        if len(nodelist) == 3 and nodelist[2][0] == symbol.gen_for:
+            test = self.com_node(nodelist[1])
+            return 0, self.com_generator_expression(test, nodelist[2])
         if len(nodelist) == 2:
             if kw:
                 raise SyntaxError, "non-keyword arg after keyword arg"

Lib/symbol.py

@@ -61,23 +61,27 @@ factor = 303
 power = 304
 atom = 305
 listmaker = 306
-lambdef = 307
-trailer = 308
-subscriptlist = 309
-subscript = 310
-sliceop = 311
-exprlist = 312
-testlist = 313
-testlist_safe = 314
-dictmaker = 315
-classdef = 316
-arglist = 317
-argument = 318
-list_iter = 319
-list_for = 320
-list_if = 321
-testlist1 = 322
-encoding_decl = 323
+testlist_gexp = 307
+lambdef = 308
+trailer = 309
+subscriptlist = 310
+subscript = 311
+sliceop = 312
+exprlist = 313
+testlist = 314
+testlist_safe = 315
+dictmaker = 316
+classdef = 317
+arglist = 318
+argument = 319
+list_iter = 320
+list_for = 321
+list_if = 322
+gen_iter = 323
+gen_for = 324
+gen_if = 325
+testlist1 = 326
+encoding_decl = 327
 #--end constants--
 
 sym_name = {}

Lib/test/test_deque.py

@@ -566,7 +566,7 @@ def test_main(verbose=None):
 
     # doctests
     from test import test_deque
-#    test_support.run_doctest(test_deque, verbose)
+    test_support.run_doctest(test_deque, verbose)
 
 if __name__ == "__main__":
     test_main(verbose=True)

Lib/test/test_genexps.py

+doctests = """
+
+Test simple loop with conditional
+
+    >>> sum(i*i for i in range(100) if i&1 == 1)
+    166650
+
+Test simple nesting
+
+    >>> list((i,j) for i in range(3) for j in range(4) )
+    [(0, 0), (0, 1), (0, 2), (0, 3), (1, 0), (1, 1), (1, 2), (1, 3), (2, 0), (2, 1), (2, 2), (2, 3)]
+
+Test nesting with the inner expression dependent on the outer
+
+    >>> list((i,j) for i in range(4) for j in range(i) )
+    [(1, 0), (2, 0), (2, 1), (3, 0), (3, 1), (3, 2)]
+
+Make sure the induction variable is not exposed
+
+    >>> i = 20
+    >>> sum(i*i for i in range(100))
+    328350
+    >>> i
+    20
+
+Test first class
+
+    >>> g = (i*i for i in range(4))
+    >>> type(g)
+    <type 'generator'>
+    >>> list(g)
+    [0, 1, 4, 9]
+
+Test direct calls to next()
+
+    >>> g = (i*i for i in range(3))
+    >>> g.next()
+    0
+    >>> g.next()
+    1
+    >>> g.next()
+    4
+    >>> g.next()
+    Traceback (most recent call last):
+      File "<pyshell#21>", line 1, in -toplevel-
+        g.next()
+    StopIteration
+
+Does it stay stopped?
+
+    >>> g.next()
+    Traceback (most recent call last):
+      File "<pyshell#21>", line 1, in -toplevel-
+        g.next()
+    StopIteration
+    >>> list(g)
+    []
+
+Test running gen when defining function is out of scope
+
+    >>> def f(n):
+    ...     return (i*i for i in xrange(n))
+    ...
+    >>> list(f(10))
+    [0, 1, 4, 9, 16, 25, 36, 49, 64, 81]
+
+    >>> def f(n):
+    ...     return ((i,j) for i in xrange(3) for j in xrange(n))
+    ...
+    >>> list(f(4))
+    [(0, 0), (0, 1), (0, 2), (0, 3), (1, 0), (1, 1), (1, 2), (1, 3), (2, 0), (2, 1), (2, 2), (2, 3)]
+    >>> def f(n):
+    ...     return ((i,j) for i in xrange(3) for j in xrange(4) if j in xrange(n))
+    ...
+    >>> list(f(4))
+    [(0, 0), (0, 1), (0, 2), (0, 3), (1, 0), (1, 1), (1, 2), (1, 3), (2, 0), (2, 1), (2, 2), (2, 3)]
+    >>> list(f(2))
+    [(0, 0), (0, 1), (1, 0), (1, 1), (2, 0), (2, 1)]
+
+#Verify that parenthesis are required in a statement
+#>>> def f(n):
+#...     return i*i for i in xrange(n)
+#...
+#SyntaxError: invalid syntax
+
+Verify early binding for the outermost for-expression
+
+    >>> x=10
+    >>> g = (i*i for i in range(x))
+    >>> x = 5
+    >>> list(g)
+    [0, 1, 4, 9, 16, 25, 36, 49, 64, 81]
+
+Verify late binding for the outermost if-expression
+
+    >>> include = (2,4,6,8)
+    >>> g = (i*i for i in range(10) if i in include)
+    >>> include = (1,3,5,7,9)
+    >>> list(g)
+    [1, 9, 25, 49, 81]
+
+Verify late binding for the innermost for-expression
+
+    >>> g = ((i,j) for i in range(3) for j in range(x))
+    >>> x = 4
+    >>> list(g)
+    [(0, 0), (0, 1), (0, 2), (0, 3), (1, 0), (1, 1), (1, 2), (1, 3), (2, 0), (2, 1), (2, 2), (2, 3)]
+
+Verify re-use of tuples (a side benefit of using genexps over listcomps)
+
+    >>> tupleids = map(id, ((i,i) for i in xrange(10)))
+    >>> max(tupleids) - min(tupleids)
+    0
+
+
+
+########### Tests borrowed from or inspired by test_generators.py ############
+
+Make a generator that acts like range()
+
+    >>> yrange = lambda n:  (i for i in xrange(n))
+    >>> list(yrange(10))
+    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+
+Generators always return to the most recent caller:
+
+    >>> def creator():
+    ...     r = yrange(5)
+    ...     print "creator", r.next()
+    ...     return r
+    ...
+    >>> def caller():
+    ...     r = creator()
+    ...     for i in r:
+    ...             print "caller", i
+    ...
+    >>> caller()
+    creator 0
+    caller 1
+    caller 2
+    caller 3
+    caller 4
+
+Generators can call other generators:
+
+    >>> def zrange(n):
+    ...     for i in yrange(n):
+    ...         yield i
+    ...
+    >>> list(zrange(5))
+    [0, 1, 2, 3, 4]
+
+
+Verify that a gen exp cannot be resumed while it is actively running:
+
+    >>> g = (me.next() for i in xrange(10))
+    >>> me = g
+    >>> me.next()
+    Traceback (most recent call last):
+      File "<pyshell#30>", line 1, in -toplevel-
+        me.next()
+      File "<pyshell#28>", line 1, in <generator expression>
+        g = (me.next() for i in xrange(10))
+    ValueError: generator already executing
+
+Verify exception propagation
+
+    >>> g = (10 // i for i in (5, 0, 2))
+    >>> g.next()
+    2
+    >>> g.next()
+    Traceback (most recent call last):
+      File "<pyshell#37>", line 1, in -toplevel-
+        g.next()
+      File "<pyshell#35>", line 1, in <generator expression>
+        g = (10 // i for i in (5, 0, 2))
+    ZeroDivisionError: integer division or modulo by zero
+    >>> g.next()
+    Traceback (most recent call last):
+      File "<pyshell#38>", line 1, in -toplevel-
+        g.next()
+    StopIteration
+
+Make sure that None is a valid return value
+
+    >>> list(None for i in xrange(10))
+    [None, None, None, None, None, None, None, None, None, None]
+
+Check that generator attributes are present
+
+    >>> g = (i*i for i in range(3))
+    >>> expected = set(['gi_frame', 'gi_running', 'next'])
+    >>> set(attr for attr in dir(g) if not attr.startswith('__')) >= expected
+    True
+
+    >>> print g.next.__doc__
+    x.next() -> the next value, or raise StopIteration
+    >>> import types
+    >>> isinstance(g, types.GeneratorType)
+    True
+
+Check the __iter__ slot is defined to return self
+
+    >>> iter(g) is g
+    True
+
+Verify that the running flag is set properly
+
+    >>> g = (me.gi_running for i in (0,1))
+    >>> me = g
+    >>> me.gi_running
+    0
+    >>> me.next()
+    1
+    >>> me.gi_running
+    0
+
+Verify that genexps are weakly referencable
+
+    >>> import weakref
+    >>> g = (i*i for i in range(4))
+    >>> wr = weakref.ref(g)
+    >>> wr() is g
+    True
+    >>> p = weakref.proxy(g)
+    >>> list(p)
+    [0, 1, 4, 9]
+
+
+"""
+
+
+__test__ = {'doctests' : doctests}
+
+def test_main(verbose=None):
+    import sys
+    from test import test_support
+    from test import test_genexps
+    test_support.run_doctest(test_genexps, verbose)
+
+    # verify reference counting
+    if verbose and hasattr(sys, "gettotalrefcount"):
+        import gc
+        counts = [None] * 5
+        for i in xrange(len(counts)):
+            test_support.run_doctest(test_genexps, verbose)
+            gc.collect()
+            counts[i] = sys.gettotalrefcount()
+        print counts
+
+if __name__ == "__main__":
+    test_main(verbose=True)
+
+
+
+
+
+

Lib/test/test_grammar.py

@@ -739,3 +739,46 @@ print [
         for (sp_sno, sp_pno) in suppart
           if sno == sp_sno and pno == sp_pno
 ]
+
+# generator expression tests
+g = ([x for x in range(10)] for x in range(1))
+verify(g.next() == [x for x in range(10)])
+try:
+    g.next()
+    raise TestFailed, 'should produce StopIteration exception'
+except StopIteration:
+    pass
+
+a = 1
+try:
+    g = (a for d in a)
+    g.next()
+    raise TestFailed, 'should produce TypeError'
+except TypeError:
+    pass
+
+verify(list((x, y) for x in 'abcd' for y in 'abcd') == [(x, y) for x in 'abcd' for y in 'abcd'])
+verify(list((x, y) for x in 'ab' for y in 'xy') == [(x, y) for x in 'ab' for y in 'xy'])
+
+a = [x for x in range(10)]
+b = (x for x in (y for y in a))
+verify(sum(b) == sum([x for x in range(10)]))
+
+verify(sum(x**2 for x in range(10)) == sum([x**2 for x in range(10)]))
+verify(sum(x*x for x in range(10) if x%2) == sum([x*x for x in range(10) if x%2]))
+verify(sum(x for x in (y for y in range(10))) == sum([x for x in range(10)]))
+verify(sum(x for x in (y for y in (z for z in range(10)))) == sum([x for x in range(10)]))
+verify(sum(x for x in [y for y in (z for z in range(10))]) == sum([x for x in range(10)]))
+verify(sum(x for x in (y for y in (z for z in range(10) if True)) if True) == sum([x for x in range(10)]))
+verify(sum(x for x in (y for y in (z for z in range(10) if True) if False) if True) == 0)
+check_syntax("foo(x for x in range(10), 100)")
+check_syntax("foo(100, x for x in range(10))")
+
+# test for outmost iterable precomputation
+x = 10; g = (i for i in range(x)); x = 5
+verify(len(list(g)) == 10)
+
+# This should hold, since we're only precomputing outmost iterable.
+x = 10; t = False; g = ((i,j) for i in range(x) if t for j in range(x))
+x = 5; t = True;
+verify([(i,j) for i in range(10) for j in range(5)] == list(g))

Lib/test/test_parser.py

@@ -67,6 +67,8 @@ class RoundtripLegalSyntaxTestCase(unittest.TestCase):
         self.check_expr("lambda foo=bar, blaz=blat+2, **z: 0")
         self.check_expr("lambda foo=bar, blaz=blat+2, *y, **z: 0")
         self.check_expr("lambda x, *y, **z: 0")
+        self.check_expr("(x for x in range(10))")
+        self.check_expr("foo(x for x in range(10))")
 
     def test_print(self):
         self.check_suite("print")

Misc/ACKS

@@ -500,6 +500,7 @@ Barry Scott
 Steven Scott
 Nick Seidenman
 Fred Sells
+Jiwon Seo
 Denis Severson
 Ha Shao
 Bruce Sherwood

Misc/NEWS

@@ -12,6 +12,8 @@ What's New in Python 2.4 alpha 1?
 Core and builtins
 -----------------
 
+- Implemented generator expressions (PEP 289).  Coded by Jiwon Seo.
+
 - Enabled the profiling of C extension functions (and builtins) - check
   new documentation and modified profiler and bdb modules for more details
 

Modules/parsermodule.c

@@ -855,7 +855,9 @@ VALIDATER(subscriptlist);       VALIDATER(sliceop);
 VALIDATER(exprlist);            VALIDATER(dictmaker);
 VALIDATER(arglist);             VALIDATER(argument);
 VALIDATER(listmaker);           VALIDATER(yield_stmt);
-VALIDATER(testlist1);
+VALIDATER(testlist1);           VALIDATER(gen_for);
+VALIDATER(gen_iter);            VALIDATER(gen_if);
+VALIDATER(testlist_gexp);
 
 #undef VALIDATER
 
@@ -1246,6 +1248,21 @@ validate_list_iter(node *tree)
     return res;
 }
 
+/*  gen_iter:  gen_for | gen_if
+ */
+static int
+validate_gen_iter(node *tree)
+{
+    int res = (validate_ntype(tree, gen_iter)
+               && validate_numnodes(tree, 1, "gen_iter"));
+    if (res && TYPE(CHILD(tree, 0)) == gen_for)
+        res = validate_gen_for(CHILD(tree, 0));
+    else
+        res = validate_gen_if(CHILD(tree, 0));
+
+    return res;
+}
+
 /*  list_for:  'for' exprlist 'in' testlist [list_iter]
  */
 static int
@@ -1268,6 +1285,28 @@ validate_list_for(node *tree)
     return res;
 }
 
+/*  gen_for:  'for' exprlist 'in' test [gen_iter]
+ */
+static int
+validate_gen_for(node *tree)
+{
+    int nch = NCH(tree);
+    int res;
+
+    if (nch == 5)
+        res = validate_gen_iter(CHILD(tree, 4));
+    else
+        res = validate_numnodes(tree, 4, "gen_for");
+
+    if (res)
+        res = (validate_name(CHILD(tree, 0), "for")
+               && validate_exprlist(CHILD(tree, 1))
+               && validate_name(CHILD(tree, 2), "in")
+               && validate_test(CHILD(tree, 3)));
+
+    return res;
+}
+
 /*  list_if:  'if' test [list_iter]
  */
 static int
@@ -1288,6 +1327,25 @@ validate_list_if(node *tree)
     return res;
 }
 
+/*  gen_if:  'if' test [gen_iter]
+ */
+static int
+validate_gen_if(node *tree)
+{
+    int nch = NCH(tree);
+    int res;
+
+    if (nch == 3)
+        res = validate_gen_iter(CHILD(tree, 2));
+    else
+        res = validate_numnodes(tree, 2, "gen_if");
+    
+    if (res)
+        res = (validate_name(CHILD(tree, 0), "if")
+               && validate_test(CHILD(tree, 1)));
+
+    return res;
+}
 
 /*  validate_fpdef()
  *
@@ -2187,7 +2245,7 @@ validate_atom(node *tree)
                    && (validate_rparen(CHILD(tree, nch - 1))));
 
             if (res && (nch == 3))
-                res = validate_testlist(CHILD(tree, 1));
+                res = validate_testlist_gexp(CHILD(tree, 1));
             break;
           case LSQB:
             if (nch == 2)
@@ -2244,7 +2302,7 @@ validate_listmaker(node *tree)
         ok = validate_test(CHILD(tree, 0));
 
     /*
-     *  list_iter | (',' test)* [',']
+     *  list_for | (',' test)* [',']
      */
     if (nch == 2 && TYPE(CHILD(tree, 1)) == list_for)
         ok = validate_list_for(CHILD(tree, 1));
@@ -2266,6 +2324,43 @@ validate_listmaker(node *tree)
     return ok;
 }
 
+/*  testlist_gexp:
+ *    test ( gen_for | (',' test)* [','] )
+ */
+static int
+validate_testlist_gexp(node *tree)
+{
+    int nch = NCH(tree);
+    int ok = nch;
+
+    if (nch == 0)
+        err_string("missing child nodes of testlist_gexp");
+    else {
+        ok = validate_test(CHILD(tree, 0));
+    }
+
+    /*
+     *  gen_for | (',' test)* [',']
+     */
+    if (nch == 2 && TYPE(CHILD(tree, 1)) == gen_for)
+        ok = validate_gen_for(CHILD(tree, 1));
+    else {
+        /*  (',' test)* [',']  */
+        int i = 1;
+        while (ok && nch - i >= 2) {
+            ok = (validate_comma(CHILD(tree, i))
+                  && validate_test(CHILD(tree, i+1)));
+            i += 2;
+        }
+        if (ok && i == nch-1)
+            ok = validate_comma(CHILD(tree, i));
+        else if (i != nch) {
+            ok = 0;
+            err_string("illegal trailing nodes for testlist_gexp");
+        }
+    }
+    return ok;
+}
 
 /*  funcdef:
  *      'def' NAME parameters ':' suite
@@ -2318,6 +2413,18 @@ validate_arglist(node *tree)
         /* raise the right error from having an invalid number of children */
         return validate_numnodes(tree, nch + 1, "arglist");
 
+    if (nch > 1) {
+        for (i=0; i<nch; i++) {
+            if (TYPE(CHILD(tree, i)) == argument) {
+                node *ch = CHILD(tree, i);
+                if (NCH(ch) == 2 && TYPE(CHILD(ch, 1)) == gen_for) {
+                    err_string("need '(', ')' for generator expression");
+                    return 0;
+                }
+            }
+        }
+    }
+
     while (ok && nch-i >= 2) {
         /* skip leading (argument ',') */
         ok = (validate_argument(CHILD(tree, i))
@@ -2377,17 +2484,19 @@ validate_arglist(node *tree)
 
 /*  argument:
  *
- *  [test '='] test
+ *  [test '='] test [gen_for]
  */
 static int
 validate_argument(node *tree)
 {
     int nch = NCH(tree);
     int res = (validate_ntype(tree, argument)
-               && ((nch == 1) || (nch == 3))
+               && ((nch == 1) || (nch == 2) || (nch == 3))
                && validate_test(CHILD(tree, 0)));
 
-    if (res && (nch == 3))
+    if (res && (nch == 2))
+        res = validate_gen_for(CHILD(tree, 1));
+    else if (res && (nch == 3))
         res = (validate_equal(CHILD(tree, 1))
                && validate_test(CHILD(tree, 2)));
 

Python/symtable.c

@@ -66,6 +66,8 @@ PySymtableEntry_New(struct symtable *st, char *name, int type, int lineno)
 	switch (type) {
 	case funcdef:
 	case lambdef:
+	case testlist_gexp: /* generator expression */
+	case argument:      /* generator expression */
 		ste->ste_type = TYPE_FUNCTION;
 		break;
 	case classdef:

Tools/compiler/ast.txt

@@ -38,6 +38,10 @@ AssAttr: expr, attrname*, flags*
 ListComp: expr, quals!
 ListCompFor: assign, list, ifs!
 ListCompIf: test
+GenExpr: code 
+GenExprInner: expr, quals!
+GenExprFor: assign, iter, ifs!
+GenExprIf: test
 List: nodes!
 Dict: items!
 Not: expr
@@ -85,3 +89,10 @@ init(Lambda):
         self.varargs = 1
     if flags & CO_VARKEYWORDS:
         self.kwargs = 1
+
+init(GenExpr):
+    self.argnames = ['[outmost-iterable]']
+    self.varargs = self.kwargs = None
+
+init(GenExprFor):
+    self.is_outmost = False