James M Snell wrote: > Brian Smith wrote: > > In other words, categories are (supposed to be) uniquely > > identified by @term, not by (@scheme, @term). So, terms must > > be globally unique to be unambiguous and so they SHOULD (IMO) > > be IRIs already. > > There was never any intent to make categories globally > unique; nor is there any justification to requiring them to > be so.
I didn't say that categories were required to be unique. I said that if you want to uniquely identify a category, it has to have a unique term. If you are happy with them being ambiguous, then uniqueness isn't necessary. > The scheme attribute tells the processor how to > interpret the value of term[...] The specification does not say anything about how to interpret terms, only that categories are identified by terms. > [...] which can be any string value, including an IRI. I know any string value can be used (even the empty string). But I still recommend that people use IRIs as for the term attribute to avoid any confusion. I can see why you might think that the authors of the specification intended to write the opposite of what they wrote. But, I doubt that all the people that have reviewed the specification overlooked such an obvious contradiction. It is better to assume that the specification says what it means. Regardless, It is prudent to apply the robustness principle here: when creating categories, ensure the category is uniquely identified by the term attribute (interpret the specification as it is written), and when accepting categories from others, copy them unmodified (if the producer doesn't take the spec literally, you should also tolerate that as much as possible). If you do that, you are unlikely to run into any problems. - Brian
