- From: Roy T. Fielding <fielding@apache.org>
- Date: Mon, 28 Apr 2003 16:20:44 -0700
- To: Tim Bray <tbray@textuality.com>
- Cc: WWW-Tag <www-tag@w3.org>
> 1. Roy Fielding is concerned about the fact that the IRI spec isn't > finished, saying "it would be ridiculous to say we support IRIs" when > it isn't clear yet what they are. Actually, what I said is that it is absurd to recommend IRIs when the IRI definition is unknown, and that absurdity is demonstrated by the technical errors now present in the namespaces 1.1 CR section 9. > I share concerns about the wording in the namespaces draft, BTW. Roy > suggested that it should be reworded to say that no canonicalization > is required before namespace comparison rather than say that ~wilbur, > %7ewilbur, and %7Ewilbur "are different", because in fact per the RFCs > they're not different. But this wouldn't stop me saying that it's OK > to start writing in support for internationalized identifiers. More importantly, it is because the namespaces draft cannot declare them to be different because a normalizer has every right (and in some cases a responsibility) to normalize those URIs before the namespace processor even sees them. Namespaces would therefore be violating the definition of URIs by declaring them to be different in spite of their equivalence. Therefore, Namespaces shouldn't say that the namespaces (identified) are different, even for a limited purpose. What it should say is that the identifier is assumed to be in normal form (because consistency has its own rewards) and that no additional normalization is required prior to comparison (for efficiency reasons), noting that *because* of this decision, inconsistent use of equivalent URIs in the namespace attributes will result in a regrettable, but not fatal, false negative match when they are mixed within the same process. Authors are therefore encouraged to be consistent for the sake of efficiency. That applies to both IRIs and URIs. In fact, it applies to anything that might appear in those attributes: there is no technical need for the Namespace attribute syntax and processing to be dependent on IRIs and URIs. There is a social need, but that can be accomplished without introducing dependencies on nonexistent specifications. > In any case, at the moment, we're paralyzed on this issue because of > these unresolved differences. This is on the face of it at one level > ridiculous, because the first W in WWW stands for "World" and it's a > no-brainer that identifiers ought to include non-ASCII characters. I'll take issue with that. Using localized characters in a namespace name is an incredibly stupid idea that will result in systems that do not work as well as those that stick to ascii URI. It is a trade-off that the technology should allow in the hope that some day conditions will improve, but anyone who treats that decision as a no-brainer will be doomed to regret it as soon as their namespace becomes interesting outside their own locality. > I think we do generally agree that the IRI work is in a good and > useful direction, and that one thing that would be totally useful > would be to get behind the work on the IRI draft: > > http://www.w3.org/International/iri-edit/ > > And get that nailed down and blessed. Yes, please. ....Roy
Received on Monday, 28 April 2003 19:19:10 UTC