- From: Ross Patterson <Ross_Patterson@ns.reston.vmd.sterling.com>
- Date: Tue, 18 Feb 97 14:39:00 EST
- To: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
Ben Laurie <ben@gonzo.ben.algroup.co.uk> writes: >I was about to point this out. But it goes further. The point of a URL is that >it provides _complete_ information on how to retrieve the specified resource. >TCP is therefore mandated by http: in order to make the information complete >(we then know that we resolve the hostname with DNS and connect with TCP). Actually, TCP and port 80 are mandated by the fact that you've searched the DNS for an IP address that matched the specified name. People tend to forget that the DNS can and does support more than just IP. If you'd searched using a different record class (e.g. CHAOS), you'd need to use the appropriate native stream-transport protocol. It is perfectly reasonable for a given name to have both IP and non-IP data in the DNS, and which you ask for would depend on which you wanted. I'm certainly not going to propose a new DNS record class for, say, SNA network information. But if there was one it would have it's own set of defaults for all the various protocols. For example, HTTP would probably have a "well known program name" to which you would direct an APPC conversation. Whether or not TCP is mandated by the "http:" URL scheme depends at least in part on whether or not you're living on an IP network. It is not outside the bounds of reason to have "http:" URLs served over TP4 if the client and server are both on OSI-IP networks, as long as that's what is appropriate to the environment. Of course, the market for an OSI-IP capable version of Netscape Navigator might be vanishingly small, but that doesn't change the facts. Ross Patterson Sterling Software, Inc. VM Software Division
Received on Tuesday, 18 February 1997 14:10:44 UTC