1) Changed cc.REPEAT_FOREVER = - 1 to cc.REPEAT_FOREVER = 0xffffffff
[Reason]: If cc.REPEAT_FOREVER = -1, it will be a very big double value after converting it to double by JS_ValueToNumber on android.
Then cast it to unsigned int, the value will be 0. The schedule will not be able to work.
I don't know why this occurs only on android.
[Solution]: Instead of passing -1 to it, I assign it with max value of unsigned int in c++.
2) Added two helper function, cc.ArrayGetIndexOfObject and cc.ArrayContainsObject.
3) Added JSScheduleWrapper::removeTargetForNativeNode to avoid memory leaks.
4) Improvments for JSTouchDelegate. Added four functions as follows:
// Set the touch delegate to map by using the key (pJSObj).
static void setDelegateForJSObject(JSObject* pJSObj, JSTouchDelegate* pDelegate);
// Get the touch delegate by the key (pJSObj).
static JSTouchDelegate* getDelegateForJSObject(JSObject* pJSObj);
// Remove the delegate by the key (pJSObj).
static void removeDelegateForJSObject(JSObject* pJSObj);
void unregisterTouchDelegate();
And exported cc.unregisterTouchDelegate(); to js.
Fix a memory leak for JSTouchDelegate by making it as an autorelease object.
5) Don't add js callback function to the reserved slot of object.
[Reason]: The target object may execute more than one schedule.
Therefore, previous js callback function will be replaced
by the current one. For example:
this.scheduleOnce(function() { temporary function 1 }, 0.5);
this.scheduleOnce(function() { temporary function 2 }, 0.5);
In this case, the temporary function 1 will be removed from reserved slot 0.
And temporary function 2 will be set to reserved slot 0 of this object.
If gc is triggered before the JSScheduleWrapper::scheduleFunc is invoked,
crash will happen. You could simply reproduce it by adding jsc.garbageCollect(); after scheduleOnce.
[Solution] Because one schedule corresponds to one JSScheduleWrapper, we root
the js callback function in JSScheduleWrapper::setJSCallbackFunc and unroot it
at the destructor of JSScheduleWrapper.
Some fixes of JSBinding codes:
[1] Check whether the proxy was already added in JS_NEW_PROXY
[2] In struct schedFunc_proxy_t, JSScheduleWrapper* --> CCArray*
Reason: One js function may correspond to many targets. To debug this, you could refer to JSScheduleWrapper::dump function. It will prove that i'm right. :)
[3] In ScriptingCore::cleanupSchedulesAndActions function, we must invoke unschedule for all targets and remove the proxy both in _schedFunc_target_ht and _schedTarget_native_ht, otherwise the hash tables will grow bigger and bigger, so I added a new function JSScheduleWrapper::removeAllTargetsForNatiaveNode to make this things easier.
[4] To easily find out the bugs of binding codes, I add JS_SetGCZeal in ScriptingCore::createGlobalContext, it only works in DEBUG mode.
[5] In js_cocos2dx_CCNode_getChildren, we should add the generated array to root to avoid gc happen when invoking JS_SetElement.
[6] The JSCallFuncWrapper isn't needed since an action will be run by a cc.Node and it will be released at the CCNode::cleanup.
[7] Some improvements of JSScheduleWrapper class.
[8] Added a new function JSScheduleWrapper::setTarget, it's for js_CCNode_unschedule to find out which target need to be unscheduled.
[9] Commented JS_SetReservedSlot in js_CCNode_scheduleOnce and js_CCNode_schedule.
Reason:
For js_CCNode_scheduleOnce: Don't add the callback function to the reserved slot of this js object.Since the class of js object may be inherited from cocos class(e.g. cc.Sprite). The subclass will not contain reserved slots. It will crash if invoking this.
For js_CCNode_schedule: Don't add js callback function to the reserved slot of scheduler js object. Since the scheduler is an object always rooted.
So the callback function might not be released when gc comes.I looked inside the implementation of cc.Node.schedule, and it doesn't use JS_SetReservedSlot there.
James Chen