- From: Paul Prescod <paul@prescod.net>
- Date: Mon, 18 Mar 2002 18:10:00 -0800
- To: www-tag@w3.org
Chris Lilley wrote: > > ... > > Okay, but with a flourish I can produce the URI > zip://atm.example.org/06902 or, worse, zip:06902 but just because > these now have URIs does not lessen in any way their proprietary > nature, and the second one is probably not dereferencable either. Okay, but 1. Why would anyone do such a thing? People know that non-HTTP URI schemes have a massive barrier to acceptance. 2. If you define your interface in terms zip: URIs, XML documents and a REST-y zip: protocol, I can easily compete with you by defining an alternate interface in terms of http: URIs, XML documents and the HTTP protocol. You will be forced to support HTTP also just to compete. Whereas, if you hide your data model behind a non-REST interface, building an HTTP bridge to it could be a huge challenge. 3. Even a proprietary URI scheme is better than an arbitrary RPC interface as long as it supports GET, because it is possible to add dereferencing for a new URI scheme to (let's say) the whole JVM or all .NET applications through well-documented component interfaces. So you could get both an XSLT engine and a browser written for the JVM to have access zip: data for the price of one component, not two. If these new schemes actually arose as often in practice as they do in theory I could imagine some slick APIs where an application would have a private cache of handlers, and then fall back to the runtime library, which would fall back to the operating system. So if you could find a handler for any of Mozilla, or XPCOM or Linux you would be able to access zip: data. I'm fairly confident that this stuff only works today for GET, not PUT/POST/DELETE. Paul Prescod
Received on Monday, 18 March 2002 21:13:39 UTC