Multithreaded issues

[bjkarasek]

I want to be to modify the World state on one thread (logic), and read it on another (view).
However I'm often getting errors at this line when trying to do it.
Should it even be possible to do?

[bjkarasek]

Would it be a good workaround to just copy the world to another every tick and use only the other world in view?

[nilpunch]

Would it be a good workaround to just copy the world to another every tick and use only the other world in view?

Yeah, you can. However, copying the world is not thread-safe and can be costly, you might also lose the performance gains from multithreading. You'll need to profile this.
I’ll look into the issue more closely. I think there is a way to implement a read-only form of thread safety.

[bjkarasek]

Yeah, you can. However, copying the world is not thread-safe and can be costly, you might also lose the performance gains from multithreading

Is there allocation when copying worlds?

I’ll look into the issue more closely. I think there is a way to implement a read-only form of thread safety.

Thank you.
I implemented the mentioned approach with copying worlds as part of the simulation, and I alternate between two copies for view. This allows me to do the copying in non-blocking thread when the tick simulation is completed.
However now I'm getting a similar error but in a different line:

And in a similar way, debugger tells me that the size of an array is 2, but the index is 0, which shouldn't cause an error.
I'm not sure if it's on my side or not. I'll try debugging it, but do you have any suspicions of what might be the case?

[bjkarasek]

Adding lock (_lock) around each method in BitSetBase seems to fix the issue I had, so that confirms that the issue is related to multithreading.

[bjkarasek]

After debugging it some more I realized that I misunderstood the debugger. The index was -1 and after RemoveOnAddCount++ was called it was changed to 0, which is what was displayed in the debugger. So it seems like this might not be an issue related to threads.

[nilpunch]

Is there allocation when copying worlds?

No, there shouldn't be. That said, memory copying is still costly regardless, depending on how large the world is.

[nilpunch]

@bjkarasek Also, based on your other comments, your overall use case is pretty much what this library was designed for. I just haven't had the time to test how it performs in multi-threaded scenarios.
Is there a better place to contact you so we can coordinate more easily?

[bjkarasek]

@bjkarasek Also, based on your other comments, your overall use case is pretty much what this library was designed for. I just haven't had the time to test how it performs in multi-threaded scenarios. Is there a better place to contact you so we can coordinate more easily?

Apologies for the late reply. You can add me on discord and we can communicate there.

After debugging it some more I realized that I misunderstood the debugger. The index was -1 and after RemoveOnAddCount++ was called it was changed to 0, which is what was displayed in the debugger. So it seems like this might not be an issue related to threads.

With regards to this bug. in the end it was an issue with using the same world from multiple threads. I had one more place where I was using multiple threads. After fixing it I didn't have any such issues in the past few days.