You're probably right. Something like Urbit[1] is more a romantic notion than a way forward—people won't give up the many for the one.
Still, though, I'm heartened by the Unified Theory Of Gradual Protocol Ossification—this being the idea that, even though we've ratcheted up from computer networks speaking any kind of IP-based transport protocol to only being able to speak TCP, and then in some places only being able to speak HTTP over TCP—that what this really implies is that HTTP is commoditizing the transport layer, and that, as soon as the abstraction is airtight—which it seems increasingly to be with HTTP2—we'll be able to just sweep the redundancy under the rug by replacing the IP+TCP+HTTP2 mess with a single cleaned-up protocol that presents the same abstraction.
In this case, I imagine that the way forward is meta-languages that are effectively DSLs targeting multiple abstract machines, allowing you to mix-and-match semantics in isolated subsets of your programs, and resulting in mixed-target code generation. Once people are on that abstraction—where "language" is not 1:1 with "abstract machine", and "abstract machine" is not 1:1 with deployment platform—then we can come along and sweep the redundancy under the rug.
Huh? It sounds like you're engaging in the same kind of wishful thinking about HTTP2 now, rather than programming languages. What you are describing will never happen.
Still, though, I'm heartened by the Unified Theory Of Gradual Protocol Ossification—this being the idea that, even though we've ratcheted up from computer networks speaking any kind of IP-based transport protocol to only being able to speak TCP, and then in some places only being able to speak HTTP over TCP—that what this really implies is that HTTP is commoditizing the transport layer, and that, as soon as the abstraction is airtight—which it seems increasingly to be with HTTP2—we'll be able to just sweep the redundancy under the rug by replacing the IP+TCP+HTTP2 mess with a single cleaned-up protocol that presents the same abstraction.
In this case, I imagine that the way forward is meta-languages that are effectively DSLs targeting multiple abstract machines, allowing you to mix-and-match semantics in isolated subsets of your programs, and resulting in mixed-target code generation. Once people are on that abstraction—where "language" is not 1:1 with "abstract machine", and "abstract machine" is not 1:1 with deployment platform—then we can come along and sweep the redundancy under the rug.
In 20 years or so.
[1] http://doc.urbit.org/