Could you define "endpoint" in this context? Is it equivalent to an IPP "target"? Or are you using the term in the TCP sense?
>The
> assumptions we make about these URIs (there actual syntax is irrelevant)
> are:-
> a) an agent knows how to form them
> b) the only thing an agent may do with a URI it receives to it is to
> pass it to its underlying transport. This means that the creator of the URI
> MUST use the same URI forming convention as that which will be used by the
> receivers transport stack (ie. this is not a private issue for a given
> implementation). It also means that the receiver may not look at the URI to
> infer any deeper meaning (because that is a private issue for the sending
> implementation).
> This last restriction made us invent job-id. We moved to explicitly
> stating in IPP the way of identifying an endpoint.
> The real problem is that we end up with leakage from the transport
> up into the IPP layer. I cannot blindly forward requests from
> IPP-on-protocolX to IPP-on-protocolY. I have to find all the URIs and change
> them on the fly.
> There is another problem that assumpion b causes. It assumes that a
> printer knows how to form an address (URI) that makes sense in the clients
> transport stack. This is true for HTTP but not true (or certainly
> non-trivial) for other transports.
>
> I would propose that we use an adressing scheme that corresponds to
> the transport endpoint only. We then specifiy in IPP the ways of identifying
> the logical object that we wish to talk to (printer-ID, Job-ID,...).
>
Or you could invert this, and put the target addressing outside the IPP payload. Then you can forward requests and/or rewrite target addresses without ever opening the "envelop", to use Scott's postal analogy. For this to work, any internal target identifiers would have to be relative (like job-id).
It seems to me that your scheme would require the transport endpoint to be some kind of IPP Server that could route requests to Printers based on embedded printer-ID. Then you've added another layer of indirection to the IPP model.