Currently, the library seems to only decode data once an object is complete, as far as I can tell from benchmarking.
Am I missing something, or is this indeed how the decode stream operates? If so, is the ability to decode during transit planned? Otherwise, what might I be doing wrong? For my benchmark, I simply stream in all the bytes for a large object (115kB if as JSON) except for the last, then measure the time it takes between sending the last byte and receiving the data.