Changeset 34947 in webkit for trunk/JavaScriptCore/API/JSStringRefCF.cpp

Timestamp:

Jul 2, 2008, 12:00:53 AM (17 years ago)

Author:

[email protected]

Message:

Reviewed by Darin.

Disable JSLock for per-thread contexts.

No change on SunSpider.

• kjs/JSGlobalData.h:
• kjs/JSGlobalData.cpp: (KJS::JSGlobalData::JSGlobalData): (KJS::JSGlobalData::sharedInstance): Added isSharedInstance as a better way to tell whether the instance is shared (legacy).

• kjs/JSLock.cpp: (KJS::createJSLockCount): (KJS::JSLock::lockCount): (KJS::setLockCount): (KJS::JSLock::JSLock): (KJS::JSLock::lock): (KJS::JSLock::unlock): (KJS::JSLock::currentThreadIsHoldingLock): (KJS::JSLock::DropAllLocks::DropAllLocks): (KJS::JSLock::DropAllLocks::~DropAllLocks):
• kjs/JSLock.h: (KJS::JSLock::JSLock): (KJS::JSLock::~JSLock): Made JSLock and JSLock::DropAllLocks constructors take a parameter to decide whether to actually lock a mutex, or only to increment recursion count. We cannot turn it into no-op if we want to keep existing assertions working. Made recursion count per-thread, now that locks may not lock.

• API/JSBase.cpp: (JSEvaluateScript): Take JSLock after casting JSContextRef to ExecState* (which doesn't need locking in any case), so that a decision whether to actually lock can be made. (JSCheckScriptSyntax): Ditto. (JSGarbageCollect): Only lock while collecting the shared heap, not the per-thread one.

• API/JSObjectRef.cpp: (JSClassCreate): Don't lock, as there is no reason to. (JSClassRetain): Ditto. (JSClassRelease): Ditto. (JSPropertyNameArrayRetain): Ditto. (JSPropertyNameArrayRelease): Only lock while deleting the array, as that may touch identifier table. (JSPropertyNameAccumulatorAddName): Adding a string also involves an identifier table lookup, and possibly modification.

• API/JSStringRef.cpp: (JSStringCreateWithCharacters): (JSStringCreateWithUTF8CString): (JSStringRetain): (JSStringRelease): (JSStringGetUTF8CString): (JSStringIsEqual):
• API/JSStringRefCF.cpp: (JSStringCreateWithCFString): JSStringRef operations other than releasing do not need locking.

• VM/Machine.cpp: Don't include unused JSLock.h.

• kjs/CollectorHeapIntrospector.cpp: (KJS::CollectorHeapIntrospector::statistics): Don't take the lock for real, as heap introspection pauses the process anyway. It seems that the existing code could cause deadlocks.

• kjs/Shell.cpp: (functionGC): (main): (jscmain): The test tool uses a per-thread context, so no real locking is required.

• kjs/collector.h: (KJS::Heap::setGCProtectNeedsLocking): Optionally protect m_protectedValues access with a per-heap mutex. This is only needed for WebCore Database code, which violates the "no data migration between threads" by using ProtectedPtr on a background thread. (KJS::Heap::isShared): Keep a shared flag here, as well.

• kjs/protect.h: (KJS::::ProtectedPtr): (KJS::::~ProtectedPtr): (KJS::::operator): (KJS::operator==): (KJS::operator!=): ProtectedPtr is ony used from WebCore, so it doesn't need to take JSLock. An assertion in Heap::protect/unprotect guards agains possible future unlocked uses of ProtectedPtr in JSC.

• kjs/collector.cpp: (KJS::Heap::Heap): Initialize m_isShared. (KJS::Heap::~Heap): No need to lock for real during destruction, but must keep assertions in sweep() working. (KJS::destroyRegisteredThread): Registered thread list is only accessed for shared heap, so locking is always needed here. (KJS::Heap::registerThread): Ditto. (KJS::Heap::markStackObjectsConservatively): Use m_isShared instead of comparing to a shared instance for a small speedup. (KJS::Heap::setGCProtectNeedsLocking): Create m_protectedValuesMutex. There is currently no way to undo this - and ideally, Database code will be fixed to lo longer require this quirk. (KJS::Heap::protect): Take m_protectedValuesMutex (if it exists) while accessing m_protectedValues. (KJS::Heap::unprotect): Ditto. (KJS::Heap::markProtectedObjects): Ditto. (KJS::Heap::protectedGlobalObjectCount): Ditto. (KJS::Heap::protectedObjectCount): Ditto. (KJS::Heap::protectedObjectTypeCounts): Ditto.

• kjs/ustring.cpp:
• kjs/ustring.h: Don't include JSLock.h, which is no longer used here. As a result, an explicit include had to be added to many files in JavaScriptGlue, WebCore and WebKit.

• kjs/JSGlobalObject.cpp: (KJS::JSGlobalObject::init):
• API/JSCallbackConstructor.cpp: (KJS::constructJSCallback):
• API/JSCallbackFunction.cpp: (KJS::JSCallbackFunction::call):
• API/JSCallbackObjectFunctions.h: (KJS::::init): (KJS::::getOwnPropertySlot): (KJS::::put): (KJS::::deleteProperty): (KJS::::construct): (KJS::::hasInstance): (KJS::::call): (KJS::::getPropertyNames): (KJS::::toNumber): (KJS::::toString): (KJS::::staticValueGetter): (KJS::::callbackGetter):
• API/JSContextRef.cpp: (JSGlobalContextCreate): (JSGlobalContextRetain): (JSGlobalContextRelease):
• API/JSValueRef.cpp: (JSValueIsEqual): (JSValueIsStrictEqual): (JSValueIsInstanceOfConstructor): (JSValueMakeNumber): (JSValueMakeString): (JSValueToNumber): (JSValueToStringCopy): (JSValueToObject): (JSValueProtect): (JSValueUnprotect):
• JavaScriptCore.exp:
• kjs/PropertyNameArray.h: (KJS::PropertyNameArray::globalData):
• kjs/interpreter.cpp: (KJS::Interpreter::checkSyntax): (KJS::Interpreter::evaluate): Pass a parameter to JSLock/JSLock::DropAllLocks to decide whether the lock needs to be taken.

File:

• trunk/JavaScriptCore/API/JSStringRefCF.cpp (modified) (1 diff)

trunk/JavaScriptCore/API/JSStringRefCF.cpp

@@ -38 +38 @@
 JSStringRef JSStringCreateWithCFString(CFStringRef string)
 {
-    JSLock lock;
     CFIndex length = CFStringGetLength(string);
     UString::Rep* rep;