Re: Lies, damned lies, and statistics that don't lie.
it is therefore a simple logical fact that there can be no such thing as a backwards-compatible new version. You can't invent a way for a machine that believes all addresses are 32 bits long to receive or send packets with longer addresses.
That's not quite correct. There's certainly no way for a machine that believes all addresses are 32 bits long to correctly interpret longer addresses, but it can certainly process them to some extent if it thinks those longer addresses are just part of the data, say in a header extension.
The trick with any backwards-compatible protocol is to create packet formats that new-version systems can fully understand, and old-version systems can at least process enough to get them to some translator or gateway system from where they can be routed. It usually looks like a horrible kludge, which is why the IETF academics ruled it out, but it need not be impossible, at least as a transition measure.
The big problem has always been that if you plug an IPv6 device onto a network that is 100% IPv4, it's just a doorstop. There's no way, not even a kludgy, inefficient, way, that an unmodified IPv4 router can get an IPv6 packet to somewhere where it can be processed correctly.