As many protocols are currently used for information retrieval, the address must be capable of encompassing many protocols, access methods or, indeed, naming schemes.
The WWW scheme uses a prefix to give the addressing sub-scheme, and then a syntax dependent on the prefix used, in order to be open to any new naming systems.
With wide-area distributed systems, this distinction blurs. Locally, things which at first look like physical addresses develop more and more levels of translation, so that they cease to give the actual location of the object. At the same time, a logical name or a unique identifier must contain some information which allows the name server to know where to start looking. In a global context, for example "1237159242346244234232342342423468762342368" might well be unique, but it contains insufficient (apparent) structure for a name server to look it up. The name "info.cern.ch" has a structure which allows a search to be made in several stages. In fact, practical systems using unique identifiers generally hide within them some clues for the name server, such as a node name.
A hypertext link to a document ought to be specified using the most logical name as opposed to a physical address. This is (almost) the only way of getting over the problem of documents being physically moved. As the naming scheme becomes more abstract, resolving the name becomes less of a simple look-up and more of a search.
One expects in practice the translation of a document name taking several stages as the name becomes less abstract and more physical.
If this direction is chosen for naming, it still leaves open the question of the format of the address into which a document name will be translated. This must also be left as open-ended as the set of protocols.
Tim BL 1991