addImageAsync fails to invoke function callback

[PeqNP]

This issue happens ONLY when the app is first ran on a fresh Android device. Every time thereafter it appears to work without issue. I received several errors from WebKit stating that certain tables did not exist. The result was that the function callback was never called.

I lost most of the debug messages which emitted the necessary tables to be created. I did happen to save one:

DELETE FROM CacheGroups

It appears that the table CacheGroups is being cleared before the table is created.

I believe what is happening is that these tables need to be created before the WebView; which I assume is used to load the images asynchronously on Android. The thing is, I don't know whose responsibility it is to do this! Is it Android's? Or is it Cocos2d-x's? I see absolutely no reference to CacheGroups within the code; so I suspect there is some initialization that is not being performed from within the Android API.

As a workaround I'm simply going to add a FAQ, on our support page, to let the player know that they may need to restart the app if the game appears to hang after certain events; as there is absolutely no way for me to check if something went awry from within the API... well, I could add a delay and check to see if the particular process took longer than N seconds... but I really do not want to do that. I can just see it now... "Hey! We couldn't load a texture, which should have only taken 0.3 seconds! Restart the app now to correct the issue!" Yuck.

[PeqNP]

The issue is now happening again.

Reproduce steps:

• Turn the device off
• Turn the device on
• Open game

In Game:

• Load image asynchronously

Expected: The image to load asynchronously.
Actual: The image is never loaded asynchronously and the callback is not fired.

Again, this appears to only happen the first time you open the game after the phone is turned on.

I also don't see the above errors being emitted from adb logcat. There doesn't appear to be any logs that would positively identify the issue. Is there anything I can do to help you guys determine the issue? Would a dump of the logs be useful?

Here is the device's information:

Nexus S 4G
Android version: 4.1.2
Kernel version: 3.0.31-g5894150

[walzer]

Very interesting bug. We would try to reproduce it.

[PeqNP]

I'm now seeing the same issue on iOS. Essentially, the callback never happens. So I'll get the: "SpritePlist.loadFrames() - Loading texture asynchronously (%s)" log message, but not the "SpritePlist.textureLoaded() - Texture load complete (%s)" at certain times. It doesn't always happen... but it does appear to happen more often on Android. I think what I'll do right now is simply stop using the async call until I'm able to update to the latest. I want to say I'm using v3.6.

Here is my Lua code that makes the request to load the image asynchronously:

    function self.loadFrames(callback)
        if #_frames == 0 then
            callback({})
            return
        end
        local texture = cc.Director:getInstance():getTextureCache():getTextureForKey(_textureName)
        -- Already in memory.
        if texture then
            callback(self.getFrames())
            return
        end
        -- Load texture into memory.
        local function textureLoaded(texture)
            Log.d("SpritePlist.textureLoaded() - Texture load complete (%s)", _textureName)
            loadPlist()
            callback(self.getFrames())
        end
        Log.d("SpritePlist.loadFrames() - Loading texture asynchronously (%s)", _textureName)
        cc.Director:getInstance():getTextureCache():addImageAsync(_textureName, textureLoaded)
    end

[kawaiSky]

i have the same issue ,both on iOS and android.
cocos2dx version : v3.10

the _responseQueue.size is empty:

the _requestQueue size is always the total we add by 'addImageAsync'

the _loadingThread is always here

[JoHuang]

Hi,
any update?
I have the same issue on v3.10.

[JoHuang]

in function TextureCache::addImageAsync

    // ......
    // lazy init
    if (_loadingThread == nullptr)
    {
        // create a new thread to load images
        _loadingThread = new std::thread(&TextureCache::loadImage, this);
        _needQuit = false;
    }

    if (0 == _asyncRefCount)
    {
        Director::getInstance()->getScheduler()->schedule(CC_SCHEDULE_SELECTOR(TextureCache::addImageAsyncCallBack), this, 0, false);
    }

    ++_asyncRefCount;

    // generate async struct
    AsyncStruct *data = new (std::nothrow) AsyncStruct(fullpath, callback);
    
    // add async struct into queue
    _asyncStructQueue.push_back(data);
    _requestMutex.lock();
    _requestQueue.push_back(data);
    _requestMutex.unlock();

    _sleepCondition.notify_one();

After you start _loadingThread, there is a _sleepCondition in another thread waiting for notification.
_sleepCondition.notify_one(); above is the one who notifies.

What if _sleepCondition.notify_one(); happened earlier than _sleepCondition started to wait?

normally:

1. addImageAsync -- main thread
2. start _loadingThread -- main thread
3. _sleepCondition.wait(signal) -- sub thread
4. _sleepCondition.notify_one();-- main thread

I'm guessing maybe 3. and 4. may switch because they are in different thread.
What do you guys think?

Thanks.