W3C home > Mailing lists > Public > ietf-http-wg-old@w3.org > January to April 1996

Re: Length - any takers?

From: Dave Long <dave@navisoft.com>
Date: Sun, 04 Feb 1996 20:19:39 -0800
Message-Id: <199602050419.UAA06130@geek.navisoft.com>
To: BearHeart / Bill Weinman <bearheart@bearnet.com>
Cc: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com

>>>> However, closing the connection will still retain an important role in
>>>> the signaling of HTTP.
>
>>errors yes, end-of-object no.
>
>   If it is always an error to end a connection in the middle of 
>an object, then how does one "correctly" end a mulitpart server-push?

By ending between the objects, not in the middle of one.

H = headers		< = start of object	! = connection end
X = object payload	> = end of object 

H<xxxx><xxx><xxxxxx>!	- "correct" end
H<xxxx><xx><xxx><xxxx	- "incorrect" end

If a server takes care to always serve up complete objects, a client
can distinguish between:
	- HTTP error:
		the server ensures that the client receives a complete
		error response.
	- extra-protocol error: 
		the client receives no or partial response, due to
		lack of resources on the server-side, network
		disconnect, janitor unplugs machine to plug in vacuum,
		etc.

In the first case, depending upon the error, the client should not retry
until conditions have changed (such as the user providing different
authentication)

In the second case, the client may wish to retry immediately, in case
it is a transient error, or a bounced persistent connection.

-Dave
Received on Sunday, 4 February 1996 20:21:34 EST

This archive was generated by hypermail pre-2.1.9 : Wednesday, 24 September 2003 06:31:44 EDT