Changeset 35442 in webkit for trunk/JavaScriptCore/API/JSBase.cpp

Timestamp:

Jul 29, 2008, 11:05:11 PM (17 years ago)

Author:

[email protected]

Message:

Reviewed by Geoff Garen.

Implement JSContextGroup APIs to make concurrent execution possible for
JavaScriptCore clients.

This changes the behavior of JSGlobalContextCreate(), so that it now uses a private context
group for each context, making JSlock implicit locking unnecessary.

• API/JSContextRef.h:
• API/JSContextRef.cpp: (JSContextGroupCreate): (JSContextGroupRetain): (JSContextGroupRelease): (JSGlobalContextCreate): (JSGlobalContextCreateInGroup): (JSGlobalContextRelease): (JSContextGetGroup): Added new methods. JSGlobalContextCreate() calls JSGlobalContextCreateInGroup() now.

• API/APICast.h: (toJS): (toRef): Added converters for JSContextGroupRef.
• API/JSBase.cpp: (JSGarbageCollect): JSGarbageCollect(0) is now a no-op, and the passed in context is actually used.

• API/JSBase.h: Aded a typedef for JSContextGroupRef. Updated documentation for JSGarbageCollect().

• JavaScriptCore.exp: Removed JSGlobalData::sharedInstance().

• kjs/JSGlobalData.cpp:
• kjs/JSGlobalData.h: Removed support for JSGlobalData shared instance. JSGlobalData::isSharedInstance member variable still remains, to be deleted in a followup patch.

• kjs/JSLock.cpp: (KJS::JSLock::JSLock): Disabled JSLock, to be deleted in a follow-up patch.

• kjs/collector.cpp: (KJS::Heap::markOtherThreadConservatively): Removed an assertion that referenced JSGlobalData::sharedInstance.

• kjs/collector.h: Made Heap destructor public, so that JSContextRelease can use it.

JavaScriptGlue:

• JSRun.cpp: (JSRun::JSRun):
• JSUtils.cpp: (getThreadGlobalExecState): Changed JavaScriptGlue to use a JSGlobalData of its own, now that there is no shared instance.

File:

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

trunk/JavaScriptCore/API/JSBase.cpp

@@ -86 +86 @@
 void JSGarbageCollect(JSContextRef ctx)
 {
-    // Unlikely, but it is legal to call JSGarbageCollect(0) before actually doing anything that would implicitly call initializeThreading().
+    // We used to recommend passing NULL as an argument here, which caused the only heap to be collected.
+    // As there is no longer a shared heap, the previously recommended usage became a no-op (but the GC
+    // will happen when the context group is destroyed).
+    // Because the function argument was originally ignored, some clients may pass their released context here,
+    // in which case there is a risk of crashing if another thread performs GC on the same heap in between.
     if (!ctx)
-        initializeThreading();
+        return;
 
-    // When using a shared heap, clients need to call JSGarbageCollect(0) after releasing the last reference to the context to avoid
-    // leaking protected objects. Because the function arguments were originally ignored, some clients may pass their released context here,
-    // in which case there is a risk of crashing if another thread performs GC on the same heap in between.
-    if (ctx) {
-        ExecState* exec = toJS(ctx);
-        JSGlobalData& globalData = exec->globalData();
-        Heap* heap = globalData.heap;
+    ExecState* exec = toJS(ctx);
+    JSGlobalData& globalData = exec->globalData();
+    Heap* heap = globalData.heap;
 
-        JSLock lock(globalData.isSharedInstance);
+    JSLock lock(globalData.isSharedInstance);
 
-        if (!heap->isBusy())
-            heap->collect();
-    } else {
-        JSLock lock(true);
-
-        if (JSGlobalData::sharedInstanceExists()) {
-            Heap* heap = JSGlobalData::sharedInstance().heap;
-            if (!heap->isBusy())
-                heap->collect();
-        }
-    }
+    if (!heap->isBusy())
+        heap->collect();
 
     // FIXME: Perhaps we should trigger a second mark and sweep