- From: Jeremy Carroll <jjc@hpl.hp.com>
- Date: Tue, 24 Jan 2006 10:30:24 -0500
- To: Martin Duerst <duerst@it.aoyama.ac.jp>
- Cc: Bjoern Hoehrmann <derhoermi@gmx.net>, public-iri@w3.org
Martin Duerst wrote: > > At 06:37 06/01/24, Jeremy Carroll wrote: > > >A further area of doubt in the first line is allowing the "-" in the > scheme name. RFC 2717 reserves "-" in scheme names for non-IANA trees, > and as far as I can tell none have been registered, and until there is > such a registration - should not be found in a scheme name. (No > provision is made in RFC 3986 for the use of schemes that do not conform > with RFC 2717). > > This is a very good example of the possible dangers of overly agressive > checking: "currently, there are no scheme names with '-'" -> > "let's check for that". This is very dangerous, because once > deployed, it is very difficult to upgrade such implementations. > Yes but ... Currently a "-" in a scheme name is an error. If noone checks for that, ever, then the text that defines that is a waste of time, and shouldn't have been written or reviewed. It is possible to distinguish warnings from errors. A related example is the allowUnassigned flag on IDNAs. For some uses of an IRI library it is pertinent to make the best check possible given the current state of the art. Specifically something that checks IRIs that someone has just minted, all possible checks should be provided. For other uses, it is not. However, the once deployed thing, suggests that having a machine readable database, on the Web, of some of the relevant facts seems good. Jeremy
Received on Tuesday, 24 January 2006 15:30:38 UTC