pepperflash causing AssetManager to report onProgress ratio as Infinity

apellerano-kong

Got another fun bug that I can only reproduce in browser using pepperflash. This time it's related to the AssetManager.

I'm using the boilerplate AssetManager progress bar from the tutorials:

s_assets.loadQueue( function onProgress( a_ratio : Number ) : void
{
	progressBar.ratio = a_ratio;

	// show 100% for a while with a short delay
	if( a_ratio >= 1 || isNaN( a_ratio ) )
	{
		Starling.juggler.delayCall( function() : void
		{
			progressBar.removeFromParent( true );
			onAssetLoadQueueComplete();
		}, 0.15 );
	}
} );

Suddenly my complete handler was getting called 50+ times. So I added some debug statements to get a look at what's going on and here a portion of what ratio is during each call to onProgress.

[-] 0.19830919309932665
[-] 0.19994417235938586
[-] 0.19994417235938586
[-] 0.20106382978723408
[-] Infinity
[-] Infinity
[-] Infinity
[-] Infinity
[-] Infinity
[-] 0.21063829787234045
[-] Infinity
[-] 0.22021276595744685
[-] 0.22605732948600063
[-] 0.22605732948600063
[-] 0.22787234042553195
[-] 0.22978723404255322
[-] Infinity
[-] 0.23936170212765961
[-] 0.24702127659574474
[-] 0.24893617021276598
[-] Infinity
[-] Infinity
[-] Infinity
[-] Infinity
[-] Infinity
[-] 0.25851063829787235

It carries on like that. I guess since Infinity >= 1, each time it happens it causes the onComplete process to start.

I'll look into why ratio is coming out as Infinity in a bit, have to run to a dentist appt.

apellerano-kong

So far I've tracked this to ProgressEvent.bytesTotal sometimes being zero, but I have no conclusions yet.

function onLoadProgress(event:ProgressEvent):void
{
   if (onProgress != null)
      onProgress(event.bytesLoaded / event.bytesTotal);
}

apellerano-kong

This always seems to happen on my .json files, but not all of them. I did a little reading about URLLoader's bytesTotal and it says sometimes the bytesTotal can't be determined. (Some mixture of pepperflash, kongregate's servers, and my json files prevent flash from knowing the size.)

My final theory is that this bug, the potential for bytesTotal=0, has always been in AssetManager but was not likely to occur until the recent change which parallelized loaders. It's an async bug that you can luckily avoid by loading everything one at a time.

apellerano-kong

The fix/workaround was to change my loadQueue condition to check for ==1 instead of >=1

s_assets.loadQueue( function onProgress( a_ratio : Number ) : void
{
	progressBar.ratio = a_ratio;

	// show 100% for a while with a short delay
	if( a_ratio == 1 || isNaN( a_ratio ) )
	{
		Starling.juggler.delayCall( function() : void
		{
			progressBar.removeFromParent( true );
			onAssetLoadQueueComplete();
		}, 0.15 );
	}
} );

I know I pasted the original code from some Starling tutorial (I never use isNaN personally) but I have no idea which one. I think now is a good time to compare what I have to the official mobile_scaffold project, which doesn't have this problem!

edit: And it is worthwhile for Starling to improve its bytes loaded calculation to avoid passing a division by zero through to the user. I think in this case if bytesTotal=0 you could just report the loader's progress as 0 instead of Infinity. The progress bar will hang but it's more graceful.

Daniel

Sorry for that, I should have checked for division through zero here — what you're reporting never happened in any of my tests, so I didn't notice.

I just committed a fix — please try out if that helps!