Re: HTTP response version, again

At 10:14 AM 12/30/96 -0500, Abigail wrote:
>You, Brian Behlendorf, wrote:
>++ On Fri, 20 Dec 1996, M. Hedlund wrote:
>++ > On Fri, 20 Dec 1996, Dave Kristol wrote:
>++ > > I still consider the question unresolved as to what version an
>++ > > server should return for an HTTP/1.0 request.
>++ > [...]
>++ > > Case 1 (return HTTP/1.0 to HTTP/1.0 request):
>++ > > Case 2 (return HTTP/1.1 to HTTP/1.0 request):
>++ > 
>++ > I agree with Dave that Case 1 is preferable.  AOL's proxies apparently
>++ > started giving users errors this week when a new version of Apache was
>++ > released, which responded to 1.0 requests with 1.1 responses (Case 2).  
>++ > While this instance will likely be fixed next week, it does indicate
how an
>++ > HTTP/1.0 client can be confused by an HTTP/1.1 response.
>++ No, it indicates how a company with little concern for standards can
>++ implementations in other products through technological inertia.  There's
>++ nothing in the 1.1 response which should cause problems with the 1.0
proxy or
>++ 1.0 client - section 3.1 of both the 1.0 and 1.1 specs promise this,
and (as
>++ best this group can tell) 1.1 fulfills this promise.  
>But that wasn't known when HTTP/1.0 was made. It also isn't known
>whether HTTP/1.2 response headers won't contain anything that causes
>problems for HTTP/1.1 clients.
>++ If it becomes common acceptance that 1.0 and 1.1 are incompatible, then
no one
>++ will ever upgrade to 1.1.  This is exactly the perception this wg
labored long
>++ and hard to prevent.
>I don't follow this logic. Even if HTTP/1.1 and HTTP/1.0 would be
>completely imcompatible, as long as HTTP/1.1 servers talk HTTP/1.0 to
>HTTP/1.0 clients, there won't be a problem.
>++ The big question is, what will happen first: will AOL fix their
proxies, or
>++ will Apache users "fix" [hack] their servers?  Client service dictates
>++ we at Organic hack our servers, but the Apache development group has no
>++ requirements.  
>I think the problem is more fundamental. If we force HTTP/1.0 clients
>to accept HTTP/1.1 reponses, they also have to accept HTTP/1.2,
>HTTP/1.7, etc responses. That of course means no HTTP/1.x header can
>ever contain something which causes problems with HTTP/1.0 clients.

Your conclusion doesn't follow.  An HTTP/1.x server should be aware of the
version number of the client that it is speaking to, and refrain from using
header information or mechanisms that are not "understood" by that client.
A response from an HTTP/1.1 server can be constructed to be acceptable
to an HTTP/1.0 client, and still be a "legal" HTTP/1.1 response.

Steve Wingard
Spyglass, Inc.

Received on Monday, 30 December 1996 08:48:20 UTC