Ideally this would have been implemented via desugaring at parse-time,
but yield* is an expression, and its desugaring includes statements like
while and try/catch. We'd have to have BlockExpression in the AST to
support that, and it's not worth it for this feature.
So instead we implement all of the logic in
FullCodeGenerator::VisitYield. Delegating yield AST nodes now have a
try handler index, for the try/catch. Otherwise the implementation is
straightforward.
R=rossberg@chromium.org
BUG=v8:2355
TEST=mjsunit/harmony/generators-iteration
Review URL: https://codereview.chromium.org/14582007
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14669 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Ideally this would have been implemented via desugaring at parse-time,
but yield* is an expression, and its desugaring includes statements like
while and try/catch. We'd have to have BlockExpression in the AST to
support that, and it's not worth it for this feature.
So instead we implement all of the logic in
FullCodeGenerator::VisitYield. Delegating yield AST nodes now have a
try handler index, for the try/catch. Otherwise the implementation is
straightforward.
R=mstarzinger@chromium.org
BUG=v8:2355
TEST=mjsunit/harmony/generators-iteration
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14666 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
ES3 specified that functions created via Function() would have
enumerable prototypes, unlike function literals. For this reason, V8
has always had two prototypes for functions: "function_map" for
literals, and "function_instance_map" for "function instances": those
functions created by Function().
However, since 2009 or so, both maps have been the same! Both have had
writable, non-enumerable prototypes. Moreover, ES5 changed to specify
that function instances would have non-enumerable prototypes.
This patch removes the separate maps for function instances in sloppy
and strict mode.
R=mstarzinger@chromium.org
TEST=mjsunit/function-prototype
BUG=
Review URL: https://codereview.chromium.org/14829005
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14619 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This CL extends the generator suspend and resume implementation to
capture values on the operand stack.
It factors out some helpers to measure and access the operand stack into
the JavaScriptFrame class. It also refactors the suspend and resume
helpers to avoid handle allocation.
BUG=v8:2355
TEST=mjsunit/harmony/generators-iteration
Review URL: https://codereview.chromium.org/14348003
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14458 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This patch makes it so that suspending generators always saves the
context. Previously we erroneously assumed that if the operand stack
was empty, that the context would be unchanged, but that is not the case
with "with".
Fixing this brought out an interesting bug in the variable allocator.
Yield inside with will reference a context-allocated temporary holding
the generator object. Before the fix, this object was looked up in the
with context instead of the function context, because with contexts were
not being simulated during full-codegen. Previously this was OK as all
variables would be given LOOKUP allocation instead of CONTEXT, but the
context-allocated temporary invalidated this assumption. The fix is to
simulate the context chain more accurately in full-codegen.
R=mstarzinger@chromium.org
BUG=v8:2355
TEST=mjsunit/harmony/generators-iteration
Review URL: https://codereview.chromium.org/14416011
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14454 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
ia32, unlike the other architectures, includes a --debug-code check that
asserts that runtime functions do not return the hole. However the new
SuspendJSGeneratorObject runtime does return the hole at times.
This CL adds a wee hack that only signals an error if the callee was not
SuspendJSGeneratorObject.
R=mstarzinger@chromium.org
BUG=
Review URL: https://codereview.chromium.org/13856011
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14437 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Previously there has been no reason to context-allocate the receiver, so
access to the receiver always goes through the stack. This was failing
with generators, which assumed that forcing context allocation would
relieve the need of storing anything but the context and the function on
the stack.
This CL adds a slot in generator objects to capture the receiver, and
restores it when resuming a generator.
BUG=v8:2355
TEST=mjsunit/harmony/generators-iteration
Review URL: https://codereview.chromium.org/14158006
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14434 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
The generator object methods "next", "send", and "throw" now
include some inline assembly to set up a resumed stack frame. In some
common cases, we can just jump back into the frame to resume it.
Otherwise the resume code calls out to a runtime to fill in the operand
stack, rewind the handlers, and possibly to throw an exception.
BUG=v8:2355
TESTS=mjsunit/harmony/generators-iteration
Review URL: https://codereview.chromium.org/14066016
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14415 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Revert part of r14179. From the regression test's comment:
Currently, the gap resolver doesn't handle moves from a ConstantOperand to a
DoubleRegister, but these kind of moves appeared when HConstant::EmitAtUses
was changed to allow special double values (-0, NaN, hole). So we should
either enhance the gap resolver or make sure that such moves don't happen.
BUG=chrome:234101
Review URL: https://codereview.chromium.org/14429002
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14394 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
* src/ast.h:
* src/parser.cc: Differentiate between the different kinds of yields, in
anticipation of boxing return values. Parse `return' into `yield' in
a generator.
* src/runtime.h:
* src/runtime.cc (Runtime_SuspendJSGeneratorObject): New horrible
runtime function: saves continuation, context, and operands into the
generator object.
* src/arm/full-codegen-arm.cc (VisitYield):
* src/ia32/full-codegen-ia32.cc (VisitYield):
* src/x64/full-codegen-x64.cc (VisitYield): Arrange to call
SuspendJSGeneratorObject. If the call returns the hole, we suspend.
Otherwise we resume.
BUG=v8:2355
TEST=These codepaths are tested when the generator is first invoked, and so
are covered by mjsunit/harmony/generators-objects.js.
Review URL: https://codereview.chromium.org/13704010
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14353 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Generator object maps now link to their constructors, which are created
with a "Generator" class name. This does not cause a per-generator
constructor property to be set.
BUG=v8:2355
TEST=mjsunit/harmony/generators-objects
Review URL: https://codereview.chromium.org/14262004
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14309 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
When defining an indexed property on an Array object, the object's
length property should (perhaps) be updated. This was done for any
property for which
ToUInt32(name) == ToNumber(name)
was true, meaning any property name that, when converted to a number,
was an integer in the range [0, 2^32). The detection should be more
strict; an indexed property is one for which
ToString(ToUInt32(name)) == name
is true only.
Review URL: https://codereview.chromium.org/13914003
Patch from Jens Lindström <jl@opera.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14242 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
* src/contexts.h:
* src/bootstrapper.cc (InitializeExperimentalGlobal): Make generator
meta-objects, and store maps for constructing generator functions
and their prototypes.
* src/factory.h:
* src/factory.cc (MapForNewFunction): New helper.
(NewFunctionFromSharedFunctionInfo): Use the new helper.
* src/heap.cc (AllocateFunctionPrototype, AllocateInitialMap): For
generators, allocate appropriate prototypes and maps.
* src/code-stubs.h:
* src/arm/code-stubs-arm.h:
* src/arm/full-codegen-arm.h:
* src/ia32/code-stubs-ia32.h:
* src/ia32/full-codegen-ia32.h:
* src/x64/code-stubs-x64.h:
* src/x64/full-codegen-x64.h: Allow fast closure creation for generators,
using the appropriate map.
* test/mjsunit/harmony/builtins.js: Add a special case for
GeneratorFunctionPrototype.prototype.__proto__.
BUG=
TEST=mjsunit/harmony/generators-runtime
Review URL: https://codereview.chromium.org/13192004
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14236 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Currently, when a long series of removable simulates are merged, we do
this by merging them one by one as we find them. As we merge the value
value lists of the simulates, those lists snowball so that we get a
quadratic complexity wrt runtime and memory consumption.
Instead, we gather simulates that need to be merged, and merge them
backwards starting from the last simulate.
R=jkummerow@chromium.org
BUG=v8:2612
Review URL: https://chromiumcodereview.appspot.com/13649003
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14169 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
* src/scopes.h (ForceContextAllocation, has_forced_context_allocation):
New interface to force context allocation for an entire function's
scope.
* src/scopes.cc: Unless a new scope is a function scope, if its outer
scope has forced context allocation, it should also force context
allocation.
(MustAllocateInContext): Return true if the scope as a whole has
forced context allocation.
(CollectStackAndContextLocals): Allow temporaries to be
context-allocated.
* src/parser.cc (ParseFunctionLiteral): Force context allocation for
generator scopes.
* src/v8globals.h (VariableMode): Update comment on TEMPORARY.
* src/arm/full-codegen-arm.cc (Generate):
* src/ia32/full-codegen-ia32.cc (Generate):
* src/x64/full-codegen-x64.cc (Generate): Assert that generators have no
stack slots.
* test/mjsunit/harmony/generators-instantiation.js: New test.
BUG=v8:2355
TEST=mjsunit/harmony/generators-instantiation
Review URL: https://codereview.chromium.org/13408005
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14152 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This patchset begins by adding support for "yield", which is unlike other tokens
in JS. In a generator, whether strict or classic, it is a syntactic keyword.
In classic mode it is an identifier. In strict mode it is reserved.
This patch adds YIELD as a token to the scanner, and adapts the preparser and
parser appropriately. It also parses "function*", indicating that a function is
actually a generator, for both eagerly and lazily parsed functions.
Currently "yield" just compiles as "return".
BUG=v8:2355
TEST=mjsunit/harmony/generators-parsing
Review URL: https://codereview.chromium.org/12646003
Patch from Andy Wingo <wingo@igalia.com>.
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14116 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
If we find a property in the prototype-chain that we can overwrite, and
we have a transition, keep the holder in the lookup-result as the actual
holder. We will need it for the consistency-check in GenerateStoreField.
By directly checking the entire chain we avoid having to lazily bail out
to a copy of the miss stub while generating the Field Store IC.
Currently this CL disallows a normal non-receiver holder, given that
that would require a positive lookup + details verification to ensure
the property did not become read-only. This fixes the regressions in the
attached tests.
Review URL: https://chromiumcodereview.appspot.com/12810006
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14061 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This is one step in the direction of fixing a range of small bugs in the array methods when dealing with non-standard element attributes.
Added tests exercising this behavior for shift and unshift.
For Proxies and Interceptors, the behavior of %GetArrayKeys is now to just return an interval, rather than trying to list all their indexed properties. In the Proxy case, this seems like the only way to avoid an observable difference between smart and non-smart array methods. For Interceptors, the usual case (in WebKit, anyway) is for them to have all indices in [0, length), so enumerating them won't be any better than simply iterating over that range.
Review URL: https://codereview.chromium.org/12653010
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14057 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
(qua last week's TC39)
Specifically:
- Install Symbol constructor function on the global object.
- Adjust code generation for typeof.
- Remove IsSymbol built-in, IS_SYMBOL macro now defined using typeof.
- Remove hack that allowed symbols as constructor results, and some other special cases.
- Remove symbol_delegate and GetDelegate function.
- Extend ToBoolean stub to handle symbols.
- Extend ToNumber to return NaN on symbols.
- Poison symbol's toString function, and thereby ToString on symbols.
R=mstarzinger@chromium.org
BUG=v8:2158
Review URL: https://codereview.chromium.org/12957004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14051 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
- Addition of a compiled hydrogen stub for KeyedStores.
- Inlining of "grow" stubs into OPTIMIZED_FUNCTIONs
- Addition of new "ignore OOB" ic stub that silently swallows out-of-bounds stores to external typed arrays.
- Addition of new "copy-on-write" ic stub that inlines allocation and copying operations for cow array
- New stub are generated with Crankshaft, so they are automatically inlined into OPTIMIZED_FUNCTIONs
Review URL: https://codereview.chromium.org/12221064
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@14001 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
We cannot rely on C++ compilers inferring the int-type from the enum
value range. Whereas Linux/OSX find uint32_t as type for [0,MaxUInt32],
Windows insists it's int.
Update the test to execute its original intent on all platforms: 1 value
larger than max arguments, 1 smaller than max arguments (on all
platforms). This makes the test run a lot faster.
BUG=chromium:194749
Review URL: https://chromiumcodereview.appspot.com/12507010
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13988 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
The predicate CanBeSpilled had a bug, prohibiting the necessary spilling and
correct splitting of live ranges. Removed a redundant assertion immediately done
by the callee anyway.
Thanks to Slava for help with that issue and the entertaining historical
background of the whole story... ;-)
BUG=177883
Review URL: https://codereview.chromium.org/12631012
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13891 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This includes:
- actually release handles kept by compilation info when compilation completes.
- do not use parallel recompilation on single core CPUs.
- artificially delay parallel recompilation for debugging.
- fix outdated assertions wrt optimization status.
- add "parallel" option to %OptimizeFunctionOnNextCall.
R=jkummerow@chromium.org
BUG=
Review URL: https://chromiumcodereview.appspot.com/12442002
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13827 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This is a temporary workaround when the __proto__ property is being
redefined (e.g. by Object.freeze()) to not loose the foreign callback.
Once the __proto__ property is a real JavaScript accessor this hack is
no longer necessary. This change also makes __proto__ configurable.
R=rossberg@chromium.org
BUG=v8:2565
TEST=mjsunit/regress/regress-2565
Review URL: https://codereview.chromium.org/12398010
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13817 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Since symbols and strings share a common representation, most of this change is about consistently replacing 'String' with 'Name' in all places where property names are expected. In particular, no new logic at all is necessary for maps, property dictionaries, or transitions. :) The only places where an actual case distinction is needed have to do with generated type checks, and with conversions of names to strings (especially in logger and profiler).
Left in some TODOs wrt to the API: interceptors and native getters don't accept symbols as property names yet, because that would require extending the external v8.h.
(Baseline CL: https://codereview.chromium.org/12296026/)
R=verwaest@chromium.org,mstarzinger@chromium.org
BUG=v8:2158
Review URL: https://codereview.chromium.org/12330012
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13811 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
- Add --harmony-symbols flag.
- Add Symbol constructor; allow symbols as (unreplaced) return value from constructors.
- Introduce %CreateSymbol and %_IsSymbol natives and respective instructions.
- Extend 'typeof' code generation to handle symbols.
- Extend CompareIC with a UNIQUE_NAMES state that (uniformly) handles internalized strings and symbols.
- Property lookup delegates to SymbolDelegate object for symbols, which only carries the toString method.
- Extend Object.prototype.toString to recognise symbols.
Per the current draft spec, symbols are actually pseudo objects that are frozen with a null prototype and only one property (toString). For simplicity, we do not treat them as proper objects for now, although typeof will return "object". Only property access works as if they were (frozen) objects (via the internal delegate object).
(Baseline CL: https://codereview.chromium.org/12223071/)
R=mstarzinger@chromium.org
BUG=v8:2158
Review URL: https://codereview.chromium.org/12296026
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13786 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This fixes the deoptimizer to materialize arguments objects of correct
length even in cases where the actual argument values are unknown and
were optimized away by Crankshaft. This can happen if only the length
property or the identity of an arguments object is used.
R=svenpanne@chromium.org
BUG=chromium:163530
TEST=mjsunit/regress/regress-crbug-163530
Review URL: https://codereview.chromium.org/12335132
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13763 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This moves the __proto__ property to Object.prototype and turns it into
a callback property actually present in the descriptor array as opposed
to a hack in the properties lookup. For now it still is a "magic" data
property using foreign callbacks and not an accessor property visible to
JavaScript.
The second effect of this change is that JSON.parse() no longer treats
the __proto__ property specially, it will be defined as any other data
property. Note that object literals still have their special handling.
R=rossberg@chromium.org
BUG=v8:621,v8:1949,v8:2441
TEST=mjsunit,cctest,test262
Review URL: https://codereview.chromium.org/12212011
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13728 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This allows Crankshaft to completely inline a f.apply() dispatch if the
exact number of arguments is known and the function is constant. The
deoptimizer doesn't generate the f.apply() frame during deoptimization,
so the materialized frames look like f.apply() did a tailcall.
R=jkummerow@chromium.org
TEST=mjsunit/compiler/inline-function-apply
Review URL: https://codereview.chromium.org/12263004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13665 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This includes:
* Adding support for saving callee-clobbered double registers in Crankshaft code.
* Adding a new "HTrapAllocationMemento" hydrogen instruction to handle AllocationSiteInfo data in crankshafted stubs.
* Adding a new "HAllocate" hydrogen instruction that can allocate raw memory from the GC in crankshafted code.
* Support for manipulation of the hole in HChange instructions for Crankshafted stubs.
* Utility routines to manually build loops and if statements containing hydrogen code.
Review URL: https://codereview.chromium.org/11659022
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13585 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
When a data property has its attributes changed but its value remains the same,
don't emit an oldValue. This makes the API more consistent by only emitting
oldValue when the value of a property has actually changed (or been removed,
in the case of a reconfiguration as an accessor property or a deletion).
Review URL: https://codereview.chromium.org/11820004
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13565 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
HCheckPrototypeMaps currently records the prototype and the holder of the
prototype chain (both ends of the chain) and assumes that the chain elements
and their maps did not change in during the entirety of Crankshaft. The actual
traversal of the prototype chain happens in Lithium at code generation.
With parallel compilation, this assumption is not longer correct.
R=mstarzinger@chromium.org
BUG=
Review URL: https://chromiumcodereview.appspot.com/11864013
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13454 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Flattening strings is relatively costly and by doing it after every duplication
we avoid combinatorial explosion.
Note that flattening could have been done by e.g. using a regular expression,
too, but this is just another implementation detail and %FlattenString seems
general enough to be useful in other tests, too.
Review URL: https://codereview.chromium.org/11828014
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13337 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
