- From: Larry Masinter <masinter@parc.xerox.com>
- Date: Thu, 10 Jul 1997 08:17:05 PDT
- To: http-wg@cuckoo.hpl.hp.com
Whether something is "MUST" or "SHOULD" or "MAY" is not really a matter of deciding who wants things to be that way, or what the contractual expectations could be if an implementation is declared to be 'conformant'. The requirements we place on implementations should be justified as being necessary, either for interoperability, the proper functioning of the Internet, the avoidance of security or other threats, etc. In a few cases, we've tried to influence interoperability more indirectly (the requirement that a server MUST reject a 1.1 request without a Host header comes to mind) but those situations are unusual. In general, the transition from Proposed to Draft will loosen those requirements that are discovered, in practice, to not actually be required. We don't do ourselves or anyone else a service trying to impose anything other than sound advice on what MUST be done in order to create a reliable, functioning, non-disruptive implementation. We've tried, in HTTP/1.1, to be clearer about the requirements on proxies, but it may be that there are more categories of agents that act as servers to some HTTP clients and clients to other HTTP servers than simple caching proxies, and that we need to acknowledge this more explicitly. Henrik's rewording of the -- http://www.parc.xerox.com/masinter
Received on Thursday, 10 July 1997 08:52:36 UTC