BufferedDecoder.get_frame() and BufferedDecoder.seek() have direct access to the active buffer in a seemingly non-locked context. This seems to me like a cause for potential race conditions and should be investigated further. If it's already save, we should consider making the safety more obvious.