What if I implement a self-contained Turing-machine, whose programming is such that it is always reading ahead (maybe as the result of a perfect compilation from a perfectly-abstract language). I will call this computer System U. Now I have a computer that is certain to terminate just because it is certain to run out of tape, not because it is not perfectly Turing-complete. This language, as a string, would represent the very limits of Kolmogorov complexity. It can only proceed from mind, of course, and I could say that the chromosome and DNA represent something like the ideal System U, save for the effects of sin and entropy, for which we are solely responsible, and which God never had a part in. "He does all things well."
In such a system, addresses can be of arbitrary complexity, since they are just programs in this language. They will be short and uncompressable. They will then be executed with confidence, since they will run out of "string". They terminate at the desired object.
I am excited about the recent JSON technologies, most-recently JSON API, but also JSON Patch and JSON Pointer. I think they are still exploratory, as nobody has shown my the maths. But perhaps they achieve something close to a totally abstract encoding of arbitrary operations on a JSON document. Now, the problem, as I have said, is the non-mathematical approach. They haven't, for instance, sought what the guys of the relational calculus sought. The problem is that with data across a network, we probably want to do things to a document, as the JSON specs seek, not just store and link. But though we are doing something ugly, we need to, say, seek the basic set of operations out of which all others can be built. I think I will do exactly that to the JSON specs.
And then, we can know that we have attained it when we can ecode a call-with-current-continuation in just one HTTP request. Then we would have attained a full computer with URLs and the (JSON-based) abstract document language.
But to do it properly, it should have, not HTTP, but AQMP. And probably not such a language, but a combinatorial expression, probably written in the map-reduce of the JSON databases. Or an equivalent.