britram / draft-trammell-wire-image Goto Github PK

While I do like the new text on invariants, based on our discussion, I wasn't sure if we wanted to be so explicit about the concept of invariants here in the wire-image document. Technically, an "invariant" seems like a contract about how the wire-image will evolve over time, rather than a property of a single wire-image itself. What is or isn't an invariant is very much up to the development of the protocol.

The idea that I thought would be useful to define in the wire image is the fact that certain properties of the wire image lend themselves more or less easily to being invariants—that is, whether they are static (bits at fixed positions) or fluid (timing or ordering of packets that can't be totally controlled by the endpoint anyway); or somewhere in the spectrum in between. These concepts could help advise future definitions of invariants by suggesting which elements are better candidates to be invariant.

The wire images of protocols are mixed on the wire, due to application mechanics and interactions with the substrate. DNS is the biggest example of this. There should be a paragraph pointing this out / acknowledging it.