httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Dirk.vanGulik" <>
Subject Re: Additional comments on the Host: header.... (fwd)
Date Wed, 20 Mar 1996 09:51:02 GMT

> On Tue, 19 Mar 1996, Dirk.vanGulik wrote:
> > I'd be very much in favour of the following scheme, which fits well
> > with the URN/URC and proxy we have here:
> > 
> > 1. All 'new' servers should be able to understand both
> >
> > 	GET http://something/somewhere HTTP/1.x
> > 	GET somewhere HTTP/1.y
> > 
> >    Wether x and y can be 1 and 0 is open to debate, I would be in
> >    favour of having y=0 only and x=0 or 1.
> Yes, I agree.  'new' server = HTTP/1.1 server.

That is the deifinition of new; what I mean, at no cost we can make the
software understand both 1 *and* 0 without flagging an error. That makes
(for now) the transition smoothly. We could have a flag in http.conf
like 'HTTP_Trasition On/Off'. Which would rule wether an error is flagged
for the x=0 and y=1:

	GET http://something/somewhere HTTP/1.0
 	GET somewhere HTTP/1.1

But I would not flag that at all !
> > 2. All (current) servers should be able to understand 
> >    the Host: something line. 
> We should support this for cruftiness's sake, since it would buy us time 
> during the transition to 1.1.
Exactly !

> > 3. All 'new' clients do a 
> > 
> > 	GET somewhere HTTP/1.0
> > 	Host: something
> > 
> >    while we are still in http 1.0
> And, all 'new' clients do a 
> GET http://something/somewhere HTTP/1.1
> (no Host:)

No, IMHO they most certainly should be free to do a Host, and
our software should *not* break on that. That way we get around
proxies and other problems.
> When they support the minimal 1.1 set finally.
> I would like to add that there is something we can do to ease the 
> transition, which is to automatically generate a page with links to the 
> same resources as mounted on virtual hosts when presented with a 1.0 
> request.  Let me give a concrete example.
> ...cut...cut
> Is it worth it to pursue a semi-graceful solution for HTTP/1.0 clients?  
> It would certainly speed up the abandonment of IP-burning.  Perhaps we 
> can support a "virtual directory" for backwards compatibility, i.e. in 
> httpd.conf we would see
I think this is a MOST excelent idea !

> <VirtualHost>
> DocumentRoot /export/www/
> ErrorLog logs/saturn-error_log
> ServerName
> VHDirectory /
> VHAlias
> </VirtualHost>
> Where "VHDirectory" is a "virtual" directory which maps in VH-fashion to 
> "" when a HTTP/1.0 request is sent.  Also, I just 
> realized it would probably be good to provide a "VHAlias" directive so 
> that and point to the 
> same resource.  Finally, it would seem that "ServerName" would be 
> redundant here... :)

I see what you are after, but I would thought of a more gracefull system
where we also include the http:// prefix; that allows you in the future
to also negotiate shttp:// or http=ng://asdasdas as well.


DocumentRoot /export/www/
ErrorLog logs/saturn-error_log
DistinuisingPrefix: http://(.*)\.saturncars\.com

Or whatever.


View raw message