If the user wants to pass in and receive bytestreams, we should allow that. This means that each function would have to be tagged with the metadata (that it does its own serialization), and at a first cut, the user would have to specify at call time not to do any of our own serialization. Ideally, when you create or retrieve the function, you would actually get some metadata inside the CloudburstFunction object that would automatically tell the client to serialize or not.
If the user wants to pass in and receive bytestreams, we should allow that. This means that each function would have to be tagged with the metadata (that it does its own serialization), and at a first cut, the user would have to specify at call time not to do any of our own serialization. Ideally, when you create or retrieve the function, you would actually get some metadata inside the
CloudburstFunctionobject that would automatically tell the client to serialize or not.