- From: Balint Nagy Endre <bne@bne.ind.eunet.hu>
- Date: Sat, 7 Oct 1995 20:39:27 +0100 (MET)
- To: Paul Hoffman <ietf-lists@proper.com>
- Cc: http WG <http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com>
Paul Hoffman writes: > This is true for present software: the server software had to be listening > to a particular port in order to even get the "Host" header, or anything > for that matter. My thought was, in the future, one copy of a server might > be watching more than one port number for requests. As far as I know, none > do that now. In that case, it could be useful for the server to know which > port the request came in on. This seems a bit far-fetched, given that any > 1.0-speaking client would never tell the server, and thus it would be > assuming way too much. This is the pathological case I mentioned, but even this pathological case needs pathological programmers to not work without port numbers in Host header field value. In TCP/IP there is no way to listen simultaneously on two ports on the same socket. The getsockname API call will return the sockaddr_in structure (containing the port number) even if the programmer was too lousy to remember the port number to which the socket is bound. > I think it's probably not worth adding the port for this very small > potential in the future since it might lead to some ambiguities. Really not. > Which reminds me, we should remember in the "Host" wording of the spec that > there may be an IP address instead of a domain name in the URL. The URL > spec allows IP addresses in the host part, and some people use them when > their chosen DNS server is broken. And, I've seen a fair number of > hard-coded IP addresses in some of the big databases like Yahoo and Lycos. Agree. (Someone may think, that an IP address is a special case of host name, but an explicit statement on this is better.) Andrew. (Endre Balint Nagy) <bne@bne.ind.eunet.hu>
Received on Saturday, 7 October 1995 12:50:35 UTC