Specifically, how do the client and server know, a priori, that the cur=
rent
operation is the last operation in a sequence (and therefore MUST inclu=
de the Co
nnection: close header)?
-Carl
rturner@sharplabs.com on 09/09/98 02:17:29 PM
Please respond to rturner@sharplabs.com
To: Carl Kugler/Boulder/IBM@ibmus
cc: ipp@@pwg.org
Subject: Re: PWG> IPP> PRO> HTTP Connection: close
I tend to follow the saying "Be conservative in what you send, and libe=
ral in
what
you accept..."
Whether the text says MUST or not, IMHO we should be designing clients =
and
servers
to handle a "connection: close" header whenever it is received and stil=
l
function
normally, albeit with possibly less performance.
Since I am not working on a client, I cannot speak for what clients are=
or will
actually do, but I do think the client end should drive the connection =
status,
whevever possible.
Randy
Carl Kugler wrote:
> Section 4.1, General Headers, says
> General Header: Connection
> "close" only. Both client and server SHOULD keep a connection for the=
> duration of a sequence of operations. The client and server MUST incl=
ude this
> header for the last operation in such a sequence.
> Consider the (presumably typical) case of an interactive IPP client w=
ith a
user
> interface. How is the client to know when the current operation is t=
he last
of
> a sequence of user-initiated actions? Apparently, the only safe way =
to sat
isfy
> the MUST (an absolute requirement of the specification) would be to s=
end
> "close" after each user-initiated action. But this would violate the=
SHOULD,
> and is contrary to the recommendations of "HTTP Connection Management=
",
> http://www.ics.uci.edu/pub/ietf/http/draft-ietf-http-connection-00.t=
xt
>
> Quoting:
> "Authoritative knowledge that it is appropriate to close a connecti=
on
> can only come from the user. Unfortunately, that source is not to =
be
> trusted. First, most users don't know what a connection is, and
> having them indicate it is okay to close it is meaningless. Second=
, a
> user that does know what a connection is probably inherently greed=
y.
> Such a user would never surrender the attention that a connection =
to
> a server implies.
> But this is talking about the client actually closing the connection,=
not
> sending a header to the server to signal that the connection will be =
closed
> after completion of the response. Obviously, the header can only be s=
ent as
> part of a request. Also, "HTTP Connection Management" says clients S=
HOULD
> close connections before servers when possible.
> How are you implementing this requirement?
>
> -Carl,
=