[Incremental Delivery RFC] Requirement on Content-Length
ificator opened this issue · comments
I had a read through the proposal and it seems like Content-Length MUST be provided for each part. I'm curious as to why this is the case - is it simply because it makes client implementation easier? From a service perspective accomplishing this behavior requires buffering the result of serialization in memory before being written to the response stream, which can have a negative impact on performance in high-throughput services. It's much more efficient to serialize directly to the response stream, and let the clients rely on the boundary markers to determine if a part is complete.
This would necessitate a more complex boundary marker, as proposed by #135.