This makes back pointers in the map transition tree explicit by having
accurate back pointers throughout the lifetime of maps instead of
establishing and destroying back pointers before and after each marking
phase. This is a prerequisite for being able to clear map transitions
during incremental marking.
R=vegorov@chromium.org
BUG=v8:1465
Review URL: https://chromiumcodereview.appspot.com/10381053
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11528 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
AccessorPairs can now contain map transitions, which is similar to our current
handling of CONSTANT_FUNCTION/CONSTANT_TRANSITION, but generalized to a pair for
holding info about the getter and the setter. This way we can achieve map
sharing for objects with accessor properties, which is a prerequisite for making
them fast via inlining. We fall back to the previous way of handling accessor
properties when sharing is not possible or we don't handle a special case.
Note: When an exisiting accessor property is redefined we could in principle
move the AccessorPair out of the descriptor into the object itself (again just
like the way we do something similar for CONSTANT_FUNCTION/CONSTANT_TRANSITION),
but this would require a new property kind for holding a pair of values. Perhaps
we can implement this later, but for now this hopefully rare case is handled
like before, losing map sharing and potentially creating more maps than strictly
necessary.
Review URL: https://chromiumcodereview.appspot.com/10238005
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11496 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Constructs the (generally cyclic) graph of module instance objects
and populates their exports. Any exports other than nested modules
are currently set to 'undefined' (but already present as properties).
Details:
- Added new type JSModule for instance objects: a JSObject carrying a context.
- Statically allocate instance objects for all module literals (in parser 8-}).
- Extend interfaces to record and unify concrete instance objects,
and to support iteration over members.
- Introduce new runtime function for pushing module contexts.
- Generate code for allocating, initializing, and setting module contexts,
and for populating instance objects from module literals.
Currently, all non-module exports are still initialized with 'undefined'.
- Module aliases are resolved statically, so no special code is required.
- Make sure that code containing module constructs is never optimized
(macrofy AST node construction flag setting while we're at it).
- Add test case checking linkage.
Baseline: http://codereview.chromium.org/9722043/R=svenpanne@chromium.org,mstarzinger@chromium.org
BUG=
TEST=
Review URL: https://chromiumcodereview.appspot.com/9844002
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11336 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Although things are currently OK here, in the future it won't be enough to check
for the existence of a CALLBACKS result, we must additionally check that it
actually contains an accessor. In a nutshell: 'sed s/IsFound/IsProperty/' once
again...
Additionally, the control flow in DefinePropertyAccessor has been simplified by
using a helper function.
Review URL: https://chromiumcodereview.appspot.com/10071009
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11305 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Now the whole getter/setter/attributes triple gets created/set together,
avoiding any hacks regarding previous values/attributes, making things a lot
simpler.
While doing this, an interesting problem surfaced, which has been there for a
long time: After adding/changing acessors in slow mode, we could potentially
fail going back to fast mode because of a failed memory allocation, signaling
the need for a GC. But we have already changed the object in slow mode, so we
are not idempotent and the retry would trigger a newly inserted assertion
(namely, that the code obeys access restrictions). This has been solved by
splitting the transformation to fast mode from the actual setting of the
accessors.
Review URL: https://chromiumcodereview.appspot.com/9716035
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11112 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Marking aligned frames with a marker can produce false positives since
an optimized frame spill slot may be mistakenly seen as a marker value.
It also breaks the debugger reproducably: Tested when enabling alignment
for all functions and running the debugger unit tests.
BUG=v8:2009
TEST=no crashes in EarleyBoyer
Review URL: https://chromiumcodereview.appspot.com/9703110
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11075 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Previously, there were 1 or 2 calls to the runtime when accessors were changed
or set. This doesn't really work well with property attributes, leading to some
hacks and complicates things even further when trying to share maps in presence
of accessors. Therefore, the runtime entry now takes the full triple (getter,
setter, attributes), where the getter and/or the setter can be null in case they
shouldn't be changed.
For now, we do basically the same on the native side as we did before on the
JavaScript side, but this will change in future CLs, the current CL is already
large enough.
Note that object literals with a getter and a setter for the same property still
do 2 calls, but this is a little bit more tricky to fix and will be handled in a
separate CL.
Review URL: https://chromiumcodereview.appspot.com/9616016
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10956 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Main change from the original CL: Call::ComputeTarget does not use IsProperty
anymore, because this would potentially need a holder, which we don't have
here. Using Map::LookupInDescriptors with a NULL holder is a bit fishy in
general, because one has to be *extremely* careful when using its LookupResult.
The original CL made Chrome's NetInternalsTest.netInternalsTourTabs browser test
fail, but it's a mystery how this could happen: We should never reach
Call::ComputeTarget via Call::RecordTypeFeedback with a CALLBACKS property,
because we never consider calls to them monomorphic, which is in turn because of
the stub cache leaving them in the pre-monomorphic state. Therefore, I don't
have a clue how to write a regression test for this...
As an additional tiny bonus, the --trace-opt output for deoptimizations has been
improved.
Review URL: https://chromiumcodereview.appspot.com/9584003
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10906 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
With transitions in AccessorPairs, it is not enough to look at the PropertyType
alone to decide whether we look at a property or not: For objects with
JavaScript accessors, we have to look into the AccessorPair itself and see if
one of its 2 parts is actually a JavaScript accessor. Therefore, a predicate
with a PropertyType argument alone doesn't make sense anymore, we might need the
associated value, too.
Things are complicated by the fact that the holder in a LookupResult can be
NULL, so we must be careful to retrieve its value only when it is really
needed. To achieve the needed call-by-name semantics, a new Entry is introduced,
which is basically a closure over a DescriptorArray and an index into this array
(C++0x to the rescue!). GCC is clever enough to inline this class, so we pay no
runtime penalty for this abstraction.
It's all a bit ugly, but this is caused by the current structure of Descriptor,
DescriptorArray and LookupResult: Things would be much easier if DescriptorArray
were, well, an array of Descriptors, and LookupResult were a 'Maybe Descriptor'
(in Haskell-terms).
Review URL: https://chromiumcodereview.appspot.com/9466047
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10847 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This allows elements of the non-strict arguments object to be redefined
with custom attributes and still maintain an alias into the context.
Such a slow alias is maintained by placing a special marker into the
dictionary backing store of the arguments object.
R=rossberg@chromium.org
BUG=v8:1772
TEST=test262,mjsunit/object-define-property
Review URL: https://chromiumcodereview.appspot.com/9460004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10827 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This refactors the way we (re)define elements to perform normalization
and attribute updating at a much deeper level, thereby removing some
bogus special cases in upper runtime layers.
Most element setters take an indicator flag that distinguishes between
setting and defining. Setting of an element causes attributes to remain
unchanged, writability to be checked and callbacks to be called.
Defining of an element causes attributes to be updated and callbacks to
be overridden. The same approach could be taken for properties.
R=svenpanne@chromium.org
BUG=v8:1772
TEST=test262,test262/15.2.3.6-4-333-11
Review URL: https://chromiumcodereview.appspot.com/9443014
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10808 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Only JSObject enumerables with enum cache (fast case properties, no interceptors, no enumerable properties on the prototype) are supported.
HLoadKeyedGeneric with keys produced by for-in enumeration are recognized and rewritten into direct property load by index. For this enum-cache was extended to store property indices in a separate array (see handles.cc).
New hydrogen instructions:
- HForInPrepareMap: checks for-in fast case preconditions and returns map that contains enum-cache;
- HForInCacheArray: extracts enum-cache array from the map;
- HCheckMapValue: map check with HValue map instead of immediate;
- HLoadFieldByIndex: load fast property by it's index, positive indexes denote in-object properties, negative - out of object properties;
Changed hydrogen instructions:
- HLoadKeyedFastElement: added hole check suppression for loads from internal FixedArrays that are knows to have no holes inside.
R=fschneider@chromium.org
BUG=
TEST=
Review URL: https://chromiumcodereview.appspot.com/9425045
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10794 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This CL is an intermediate step only, in the end we need to have a single
DefineOrRedefineAccessorProperty call for a single Object.defineProperty
call. Currently we can end up making two such calls, making the necessary access
checks extremely ugly and hard (impossible?) to get right for complete spec
conformance.
The bulk of the change is quite mechanical:
* Prepare an AccessorPair *before* we add it to our data structures,
eliminating the previous voodoo-like threading of a placeholder.
* The previous item makes it possible to activate our check that we do not
share AccessorPairs by accident.
* Split a monster method into 2 quite unrelated methods.
* Use templated To method in a few places.
Review URL: https://chromiumcodereview.appspot.com/9428026
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10788 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This change enables optimization of top-level and eval-code. For this to work, it adds
support for declaring global variables in optimized code.
At the same time it disables the eager generation of deoptimization support data
in the full code generator (originally introduced in
r10040). This speeds up initial compilation and saves
memory for functions that won't be optimized. It requires
recompiling the function with deoptimization
support when we decide to optimize it.
Review URL: https://chromiumcodereview.appspot.com/9187005
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10700 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
With the upcoming changes to CALLBACKS properties, a predicate on the transition
type alone doesn't make sense anymore: For CALLBACKS one has to look into the
property's value to decide, and there is even the possibility of having a an
accessor function *and* a transition in the same property.
I am not completely happy with some parts of this CL, because they contain
redundant code, but given the various representations we currently have for
property type/value pairs, I can see no easy way around that. Perhaps one can
improve this a bit in a different CL, the current diversity really, really hurts
productivity...
As a bonus, this CL includes a few minor things:
* CaseClause::RecordTypeFeedback has been cleaned up and it handles the
NULL_DESCRIPTOR case correctly now. Under some (very unlikely) circumstances,
we previously missed some opportunities for monomorphic calls. In general, it
is rather unfortunate that NULL_DESCRIPTOR "shines through", it is just a
hack for the inability to remove a descriptor entry during GC, something
callers shouldn't have to be aware of.
* DescriptorArray::CopyInsert has been cleaned up a bit, preparing it for later
CALLBACKS-related changes.
* LookupResult::Print is now more informative for CONSTANT_TRANSITION.
Review URL: https://chromiumcodereview.appspot.com/9320066
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10600 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Loosen the requirement for Map equivalency on several map checks, including checks up the prototype chain, that are not sensitive to ElementsKinds. These selected map checks should also match against FAST_DOUBLE_ELEMENT and FAST_ELEMENT transitions of the original map. This specifically helps all variants of transitioned JSArrays to still efficiently call builtins like push, pop and sort.
BUG=none
TEST=none
Committed: http://code.google.com/p/v8/source/detail?r=10331
Review URL: http://codereview.chromium.org/9015020
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10356 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Loosen the requirement for Map equivalency on several map checks, including checks up the prototype chain, that are not sensitive to ElementsKinds. These selected map checks should also match against FAST_DOUBLE_ELEMENT and FAST_ELEMENT transitions of the original map. This specifically helps all variants of transitioned JSArrays to still efficiently call builtins like push, pop and sort.
BUG=none
TEST=none
Review URL: http://codereview.chromium.org/9015020
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10331 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
CPU-eating DOS attacks against node.js servers. Based on code from
Bert Belder. This version only solves the issue for those that compile
V8 themselves or those that do not use snapshots. A snapshot-based
precompiled V8 will still have predictable string hash codes.
Review URL: http://codereview.chromium.org/9086006
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10330 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This introduces an additional check into the StoreIC_ArrayLength builtin
checking that the array still has fast properties. Redifinitions of the
length property that would cause it's type or attributes to change, will
switch to slow properties, thereby invalidating said optimization.
R=svenpanne@chromium.org
BUG=v8:1756
TEST=test262
Review URL: http://codereview.chromium.org/8895025
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10254 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This shaves 416+ KB, just under 1% off the size of the debug d8 executable
on Linux (mostly because the CheckHelper functions for assertions were
getting separate copies for each compilation unit). The difference in
release builds is negligible---a size reduction of 0.1%.
Also, change namespace-level 'static const' variables to remove the static
storage class as it's the default.
R=danno@chromium.org
BUG=
TEST=
Review URL: http://codereview.chromium.org/8680013
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10083 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This CL introduces a third mode next to the non-strict
(henceforth called 'classic mode') and 'strict mode'
which is called 'extended mode' as in the current
ES.next specification drafts. The extended mode is based on
the 'strict mode' and adds new functionality to it. This
means that most of the semantics of these two modes
coincide.
The 'extended mode' is entered instead of the 'strict mode'
during parsing when using the 'strict mode' directive
"use strict" and when the the harmony-scoping flag is
active. This should be changed once it is fully specified how the 'extended mode' is entered.
This change introduces a new 3 valued enum LanguageMode
(see globals.h) corresponding to the modes which is mostly
used by the frontend code. This includes the following
components:
* (Pre)Parser
* Compiler
* SharedFunctionInfo, Scope and ScopeInfo
* runtime functions: StoreContextSlot,
ResolvePossiblyDirectEval, InitializeVarGlobal,
DeclareGlobals
The old enum StrictModeFlag is still used in the backend
when the distinction between the 'strict mode' and the 'extended mode' does not matter. This includes:
* SetProperty runtime function, Delete builtin
* StoreIC and KeyedStoreIC
* StubCache
Review URL: http://codereview.chromium.org/8417035
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10062 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Changes the way we do lazy deoptimization:
1. For side-effect instructions, we insert the lazy-deopt call at
the following LLazyBailout instruction.
CALL
GAP
LAZY-BAILOUT ==> lazy-deopt-call
2. For other instructions (StackCheck) we insert it right after the
instruction since the deopt targets an earlier deoptimization environment.
STACK-CHECK
GAP ==> lazy-deopt-call
The pc of the lazy-deopt call that will be patched in is recorded in the
deoptimization input data. Each Lithium instruction can have 0..n safepoints.
All safepoints get the deoptimization index of the associated LAZY-BAILOUT
instruction. On lazy deoptimization we use the return-pc to find the safepoint.
The safepoint tells us the deoptimization index, which in turn finds us the
PC where to insert the lazy-deopt-call.
Additional changes:
* RegExpLiteral marked it as having side-effects so that it
gets an explicitlazy-bailout instruction (instead of
treating it specially like stack-checks)
* Enable target recording CallFunctionStub to achieve
more inlining on optimized code.
BUG=v8:1789
TEST=jslint and uglify run without crashing, mjsunit/compiler/regress-lazy-deopt.js
Review URL: http://codereview.chromium.org/8492004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10006 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Original commit message:
Add a level of indirection to exception handler addresses.
To support deoptimization of exception handlers, the handler address in the
stack is converted to a pair of code object and an index into a separate
table of code offsets. The index part is invariant under deoptimization.
The index is packed into the handler state field so that handler size does
not change.
R=vegorov@chromium.org
BUG=
TEST=
Review URL: http://codereview.chromium.org/8538011
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9977 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
To support deoptimization of exception handlers, the handler address in the
stack is converted to a pair of code object and an index into a separate
table of code offsets. The index part is invariant under deoptimization.
The index is packed into the handler state field so that handler size does
not change.
R=vegorov@chromium.org,fschneider@chromium.org
BUG=
TEST=
Review URL: http://codereview.chromium.org/8462010
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9975 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This also refactors the way we set the length of an arrays' backing
store to use the new elements accessor interface. The actual fix is in
DictionaryElementsAccessor::SetLengthWithoutNormalize() where we first
search for non-deletable elements according to ES5 section 15.4.5.2
specifications.
Snippet from the specification: Attempting to set the length property of
an Array object to a value that is numerically less than or equal to the
largest numeric property name of an existing array indexed non-deletable
property of the array will result in the length being set to a numeric
value that is one greater than that largest numeric property name.
R=danno@chromium.org
TEST=test262/15.4.4.??-7-b-16
Review URL: http://codereview.chromium.org/8372064
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9911 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Do not rely on 'default' clauses or 'if's when analysing a PropertyType, because
this makes it hard to find the relevant places when a new type is added. Note
that the detection of "phantom property types" is left untouched, because this
might have a performance impact, especially for the GC (to be investigated).
This is a preliminary step for introducing a new kind of map transition.
Review URL: http://codereview.chromium.org/8491016
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9900 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This implementation extends the internal ObjectHashTable to be able to
hold arbitrary objects (e.g. Smis, Strings, ...) as keys by applying
specialized hashing functions to primitive types. Equality of keys is
defined using the internal SameValue function.
R=rossberg@chromium.org
BUG=v8:1622
TEST=mjsunit/harmony/collections
Review URL: http://codereview.chromium.org/8372027
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9777 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This function relies on a number of helpers for checking prototypes and
probing dictionaries. It is not possible to wrap these helpers to retry
after allocation failure in a safe way---the assembler has no way to undo
what it has already assembled.
These functions have all been duplicated with handle and raw versions. The
raw versions will eventually be removed completely.
R=ulan@chromium.org,vegorov@chromium.org
BUG=
TEST=
Review URL: http://codereview.chromium.org/8332003
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9769 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This implements block scoped 'const' declared variables in harmony mode. They
have a temporal dead zone semantics similar to 'let' bindings, i.e. accessing
uninitialized 'const' bindings in throws a ReferenceError.
As for 'let' bindings, the semantics of 'const' bindings in global scope is not
correctly implemented yet. Furthermore assignments to 'const's are silently
ignored. Another CL will introduce treatment of those assignments as early
errors.
Review URL: http://codereview.chromium.org/7992005
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9764 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This also includes the two fixes from r9674 and r9675. Here's the diff
to the previous CL.
--- a/src/runtime.cc
+++ b/src/runtime.cc
@@ -11133,17 +11133,26 @@ class ScopeIterator {
context_(Context::cast(frame->context())),
nested_scope_chain_(4) {
+ // Catch the case when the debugger stops in an internal function.
+ Handle<SharedFunctionInfo> shared_info(function_->shared());
+ if (shared_info->script() == isolate->heap()->undefined_value()) {
+ if (shared_info->scope_info()->HasContext()) Next();
+ return;
+ }
+
// Check whether we are in global code or function code. If there is a stack
// slot for .result then this function has been created for evaluating
// global code and it is not a real function.
// Checking for the existence of .result seems fragile, but the scope info
// saved with the code object does not otherwise have that information.
- int index = function_->shared()->scope_info()->
+ int index = shared_info->scope_info()->
StackSlotIndex(isolate_->heap()->result_symbol());
// Reparse the code and analyze the scopes.
ZoneScope zone_scope(isolate, DELETE_ON_EXIT);
- Handle<SharedFunctionInfo> shared_info(function_->shared());
Handle<Script> script(Script::cast(shared_info->script()));
Scope* scope;
if (index >= 0) {
Review URL: http://codereview.chromium.org/8344046
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9734 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This is reapplying r9501 with this single change which seemed to be causing most (all) of the failures for r9501.
--- a/src/debug.cc
+++ b/src/debug.cc
@@ -2230,6 +2230,7 @@ Debugger::Debugger(Isolate* isolate)
compiling_natives_(false),
is_loading_debugger_(false),
never_unload_debugger_(false),
+ force_debugger_active_(true),
message_handler_(NULL),
debugger_unload_pending_(false),
host_dispatch_handler_(NULL),
R=kmillikin@chromium.org
BUG=
TEST=
Review URL: http://codereview.chromium.org//8337009
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9684 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This refactoring (almost) gets rid of the requirement to get the target
object address for an object pointer embedded in code objects. This is
not possible on MIPS as pointers are encoded using two instructions. All
usages of RelocInfo::target_object_address() are (almost) obsoleted by
this change. The serializer still uses it, so MIPS will not yet work
with snapshots turned on.
R=danno@chromium.org,vegorov@chromium.org
Review URL: http://codereview.chromium.org/8245007
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9597 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This eliminates compile-errors when assigning Handle<SerializedScopeInfo> to
Handle<Object> in a place where the declaration was not available because
variables.h was not included.
As a result I had to also move the enum Variable::Mode to v8globals.h and
rename it to VariableMode.
Review URL: http://codereview.chromium.org/8221004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9575 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Previously, the logic using the hidden properties backing object was
spread accross use sites. Now it's all contained in JSObject, with
only simple accessors available.
Also change the backing object to be a StringDictionary rather than a JSObject.
There's still room for improvement by making a hash-table that don't
store property details as well.
Review URL: http://codereview.chromium.org/8050013
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9510 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
