This is a fundamental problem. Backwards compatibility (without introducing translation schemes and middleboxes) is literally impossible.
But no, someone said we must redo whole stack and we need every piece of sand to have public routable address..so now we are stuck between rock (old fossilized IPv4) and hard place (completely incompatible IPv6).
No it is not:
* You also have to deploy new DNS code to handle a new record type to handle longer "IPv4+" addresses.
* You also have to deploy new OS and library code with new socket, etc, APIs because all in_addr_t definitions and data structures are 32-bit-only.
If a public service has a "IPv4+" address, how does a not-IPv4+ host, or not-IPv4+ compliant code handle it? If you want >4B addresses you have to tweak all the code that touches address structures. You have to (re)deploy code on all the network elements that touch the packet bits: all the end-user applications (browsers, chat clients, etc), all the end-user operating systems, all the middle-boxes, all the routers. If you have network devices and segments between the public service and the client that are not IPv4+ compliant, you have to configure the clients to send the IPv4+ traffic to translation boxes that are IPv4+ compliant.
Basically all the stuff that is happening with IPv6.
You're re-inventing IPv6.
You're using different words, but you've got a separate addressing scheme, and dependence on proxies to enable everyone to talk to each other. This is exactly where we are with IPv6.
> You could instead move this "port" directly into IP header in backward-compatible way
If you're changing the meaning of the headers, it is by definition not backwards compatible.
