Adding Jörg to review the audio changes.

Thanks for your input.

According to the Python API: Py_Initialize() is a 'no-op' when called for a second time without calling Py_Finalize() first. We don't have to worry about that creating a leak. If it is necessary Py_Finalize() can be called when the blenderplayer ends it's main loop. I am quite new to memory management and I thought the memory used by an application would automatically be released when you exit the application.

Is there any way I can update my diff/patch with Py_Finalize() at the end of the main loop or do I have to submit an entirely new one?

First of all: I don't like this kind of hacks and prefer nice solutions.

So the first approach would be: can we fix the Py_Initialize/Py_Finalize problems? It would probably be smartest to really just call them once, that should be possible to implement for blenderplayer? Shall we include Campbell in the discussion? I guess just not calling Py_Finalize but Py_Initialize multiple times leaves even more memory leaks than calling them multiple times.

The second question would be if we really want to do this Py_Finalize hack, why the sound gets distorted, I'd prefer to find that out as it's way more efficient to fix that bug than to unitialize and reinitialize the whole sound system!

And lastly: even if we do this sound reinitialization hack: why exactly are you adding sound_exit_ghost() and what is the name ghost for? The code does nothing different from sound_exit(), actually worse, you directly call the destructor of the device (something you should never do!) before setting the shared pointers to NULL (this calls the destructor!) and thus calling the destructor twice, which results in a pretty bad bug, that could potentially crash the application horribly.

Adding Campbell. Hopefully he can shine some light on a proper fix, or maybe explain why we think we need to keep restarting Python. I get the feeling there is not a good reason other than no one has bothered fixing it.

@Joerg Mueller (nexyon),
I created the sound_exit_ghost() function, because sound_exit() is used by the embedded player. Since the embedded player aint broken I did not want to change anything there. I named the function 'ghost' to the file name where the main loop of the blenderplayer is.

Next, I have to be honest to tell you that I do not have a great understanding of shared_ptr(). I just kept trying my luck to get the blenderplayer to work. sound_exit_ghost() was the only way I managed to get the sound work normal after loading in a new blendfile in the blenderplayer.

I agree that fixing the re-initialization of Python would be a better solution.