JavaScriptCore:
2008-06-03 Darin Adler <Darin Adler>
Reviewed by Geoff.
Makes standalone SunSpider 1.025x as fast as before.
The getOwnPropertySlot virtual function now takes care of the toObject call
for get. Similarly, the put function (and later deleteProperty) does the
same for those operations. To do this, the virtual functions were moved from
the JSObject class to the JSCell class. Also, since the caller no longer knows
the identity of the "original object", which is used by JavaScript-function
based getters, changed the PropertySlot class so the original object is
already stored in the slot when getOwnPropertySlot is called, if the caller
intends to call getValue.
This affected the old interpreter code enough that the easiest thing for me
was to just delete it. While I am not certain the mysterious slowdown is not
still occurring, the net change is definitely a significant speedup.
- VM/Machine.cpp: Moved the UNLIKELY macro into AlwaysInline.h.
(KJS::resolve): Set up the originalObject in the PropertySlot before
calling getPropertySlot. Also removed the originalObject argument from
getValue.
(KJS::resolve_skip): Ditto.
(KJS::resolveBaseAndProperty): Ditto.
(KJS::resolveBaseAndFunc): Ditto.
(KJS::Machine::privateExecute): Removed the toObject calls from the get and
put functions where possible, instead calling directly with JSValue and letting
the JSValue and JSCell calls handle toObject. Same for toThisObject.
- kjs/ExecState.h: Removed OldInterpreterExecState.
- API/JSBase.cpp: Updated includes.
- kjs/LocalStorageEntry.h: Removed contents. Later we can remove the file too.
- kjs/array_instance.cpp:
(KJS::ArrayInstance::lengthGetter): Removed originalObject argumet.
(KJS::ArrayInstance::inlineGetOwnPropertySlot): Don't pass a base value to
setValueSlot. Also use UNLIKELY around the "getting elements past the end of
the array" code path; less common than successfully getting an element.
- kjs/array_object.cpp:
(KJS::getProperty): Initialize the PropertySlot with the original object.
Don't pass the original object to the get function.
(KJS::arrayProtoFuncFilter): Ditto.
(KJS::arrayProtoFuncMap): Ditto.
(KJS::arrayProtoFuncEvery): Ditto.
(KJS::arrayProtoFuncForEach): Ditto.
(KJS::arrayProtoFuncSome): Ditto.
- kjs/function_object.cpp:
(KJS::FunctionObjectImp::construct): Removed an obsolete comment.
- kjs/grammar.y: Eliminated support for some of the node types that were
used to optimize executing from the syntax tree.
- kjs/internal.cpp:
(KJS::StringImp::toThisObject): Added. Same as toObject.
(KJS::NumberImp::toThisObject): Ditto.
(KJS::GetterSetterImp::getOwnPropertySlot): Added. Not reached.
(KJS::GetterSetterImp::put): Ditto.
(KJS::GetterSetterImp::toThisObject): Ditto.
- kjs/internal.h: Added toThisObject to NumberImp for speed.
- kjs/lexer.cpp:
(KJS::Lexer::shift): Changed shift to just do a single character, to unroll
the loop and especially to make the one character case faster.
(KJS::Lexer::setCode): Call shift multiple times instead of passing a number.
(KJS::Lexer::lex): Ditto.
(KJS::Lexer::matchPunctuator): Ditto. Also removed unneeded elses after returns.
(KJS::Lexer::scanRegExp): Ditto.
- kjs/lexer.h: Removed the count argument from shift.
- kjs/math_object.cpp:
(KJS::mathProtoFuncPow): Call jsNaN instead of jsNumber(NaN).
- kjs/nodes.cpp: Removed some of the things needed only for the pre-SquirrelFish
execution model.
(KJS::ForNode::emitCode): Handle cases where some expressions are missing by
not emitting any code at all. The old way was to emit code for "true", but
this is an unnecessary remnant of the old way of doing things.
- kjs/nodes.h: Removed some of the things needed only for the pre-SquirrelFish
execution model.
- kjs/object.cpp:
(KJS::JSObject::fillGetterPropertySlot): Changed to only pass in the getter
function. The old code passed in a base, but it was never used when
actually getting the property; the toThisObject call was pointless. Also
changed to not pass a base for setUndefined.
- kjs/object.h: Added the new JSCell operations to GetterSetterImp.
Never called.
(KJS::JSObject::get): Initialize the object in the PropertySlot and don't
pass it in getValue.
(KJS::JSObject::getOwnPropertySlotForWrite): Removed the base argument
in calls to setValueSlot.
(KJS::JSObject::getOwnPropertySlot): Ditto.
(KJS::JSValue::get): Added. Here because it calls through to JSObject.
A version of JSObject::get that also handles the other types of JSValue
by creating the appropriate wrapper. Saves the virtual call to toObject.
(KJS::JSValue::put): Ditto.
(KJS::JSValue::deleteProperty): Ditto.
- kjs/property_slot.cpp:
(KJS::PropertySlot::undefinedGetter): Removed the originalObject argument.
(KJS::PropertySlot::ungettableGetter): Ditto.
(KJS::PropertySlot::functionGetter): Ditto. Use the value in the base
as the "this" object, which will be set to the original object by the new
PropertySlot initialization code. Also call toThisObject. The old code did
not do this, but needed to so we can properly handle the activation object
like the other similar code paths.
- kjs/property_slot.h:
(KJS::PropertySlot::PropertySlot): Added a constructor that takes a base
object. In debug builds, set the base to 0 if you don't pass one.
(KJS::PropertySlot::getValue): Don't take or pass the originalObject.
(KJS::PropertySlot::setValueSlot): Don't take a base object, and clear the
base object in debug builds.
(KJS::PropertySlot::setGetterSlot): Ditto.
(KJS::PropertySlot::setUndefined): Ditto.
(KJS::PropertySlot::setUngettable): Ditto.
(KJS::PropertySlot::slotBase): Assert that a base object is present.
This will fire if someone actually calls the get function without having
passed in a base object and the getter needs it.
(KJS::PropertySlot::setBase): Added. Used by the code that implements
toObject so it can supply the original object after the fact.
(KJS::PropertySlot::clearBase): Added. Clears the base, but is debug-only
code because it's an error to fetch the base if you don't have a guarantee
it was set.
- API/JSCallbackObject.h:
- API/JSCallbackObjectFunctions.h:
(KJS::JSCallbackObject::cachedValueGetter):
(KJS::JSCallbackObject::staticValueGetter):
(KJS::JSCallbackObject::staticFunctionGetter):
(KJS::JSCallbackObject::callbackGetter):
- kjs/JSActivation.cpp:
(KJS::JSActivation::getOwnPropertySlot):
(KJS::JSActivation::argumentsGetter):
- kjs/JSActivation.h:
- kjs/JSVariableObject.h:
(KJS::JSVariableObject::symbolTableGet):
- kjs/array_instance.h:
- kjs/function.cpp:
(KJS::FunctionImp::argumentsGetter):
(KJS::FunctionImp::callerGetter):
(KJS::FunctionImp::lengthGetter):
(KJS::Arguments::mappedIndexGetter):
- kjs/function.h:
- kjs/lookup.h:
(KJS::staticFunctionGetter):
(KJS::staticValueGetter):
- kjs/string_object.cpp:
(KJS::StringInstance::lengthGetter):
(KJS::StringInstance::indexGetter):
(KJS::stringInstanceNumericPropertyGetter):
- kjs/string_object.h:
Removed originalObject arguments from getters. Don't pass base values to
the various PropertySlot functions that no longer take them.
- kjs/value.cpp:
(KJS::JSCell::getOwnPropertySlot): Added. Calls toObject and then sets the slot.
This function has to always return true, because the caller can't walk the prototype
chain. Because of that, we do a getPropertySlot, not getOwnPropertySlot, which works
for the caller. This is private, only called by getOwnPropertySlotInternal.
(KJS::JSCell::put): Added. Calls toObject and then put.
(KJS::JSCell::toThisObject): Added. Calls toObject.
- kjs/value.h: Added get, put, and toThisObject to both JSValue
and JSCell. These take care of the toObject operation without an additional virtual
function call, and so make the common "already an object" case faster.
- wtf/AlwaysInline.h: Moved the UNLIKELY macro here for now. Maybe we can find a
better place later, or rename this header.
JavaScriptGlue:
2008-06-03 Darin Adler <Darin Adler>
- UserObjectImp.cpp:
(UserObjectImp::userObjectGetter): Removed originalObject argument.
- UserObjectImp.h: Ditto.
WebCore:
2008-06-03 Justin Garcia <[email protected]>
Reviewed by John.
<rdar://problem/5763082> GMail: Hang when removing indent from nested list
<rdar://problem/5775449> In Gmail and GoogleDocs, a hang occurs when I attempt to apply a list style to a large selection of text
<rdar://problem/5937624> 9D32: Hang in Safari. Using 100% of processor
- editing/InsertListCommand.cpp:
(WebCore::InsertListCommand::modifyRange): doApply() may operate on and remove
the last paragraph of the selection from the document if it's in the same list
item as startOfCurrentParagraph. Return early to avoid an infinite loop and
because there is no more work to be done. Added a FIXME (<rdar://problem/5983974>)
about the incorrect endingSelection()s.
