- From: Mike Meyer <mwm@contessa.phone.net>
- Date: Fri, 6 Oct 95 11:02:54 PST
- To: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
It strikes me that this problem has been dealt with before when mail went to domain-based addressing. I did a quick check, and RFC822 takes one of the approaches advocated here - that domain fragments can be used so long as the two hosts agree on how to resolve them. However, RFC1123 changes this, requiring fqdn in all cases (both SMTP commands and mail headers) so that forwarding can be done without the forwarder having to muck with the headers in question. Since mail forwarding seems to be similar to proxies, this would imply that the correct solution - as demonstrated by the experience with mail - is: I) The value of the Host header is the fully-qualified domain name. However, there might be better solutions, and servers will still have to deal with broken implementations, so I'd suggest relaxing this slightly: II.a) The value of the Host header should be the fully qualified domain name. II. b) A server receiving a domain fragment may handle it in any manner desired - as if Host were missing; as an error; or by offering an appropriate 300 alternative. On the other hand, I get the feeling that some people feel that Host should reflect what the user typed, and nothing else. This means we get: III) The value of the Host header may be an fqdn, IP address or domain fragment. And the onus of resolving it is on the server, and could well result in "misfetches" as user at foo.dom asks for "www", which is being served by dual-homed isp.dom, which is also serving www.isp.dom, so they get www.isp.dom instead of www.foo.dom. Personally, I don't like this one. If the client is allowed to send domain fragments, a decision about whether a proxy is allowed to rewrite them needs to be made. If option I or III is taken, this doesn't matter. <mike
Received on Friday, 6 October 1995 11:13:32 UTC