RE: NEW ISSUE: 13.1.2's Definition of 1xx Warn-Codes

If a client implements a cache, then it's a cache, and it's the server
side of the cache that adds the warn-code to the response.

-----Original Message-----
From: ietf-http-wg-request@w3.org [mailto:ietf-http-wg-request@w3.org]
On Behalf Of Travis Snoozy (Volt)
Sent: Tuesday, January 02, 2007 12:27 PM
To: William A. Rowe, Jr.
Cc: Mark Nottingham; ietf-http-wg@w3.org
Subject: RE: NEW ISSUE: 13.1.2's Definition of 1xx Warn-Codes


William A. Rowe, Jr. said:
> Travis Snoozy (Volt) wrote:
> > Mark Nottingham said:
> >> Added as i54;
> >>
> http://www.w3.org/Protocols/HTTP/1.1/rfc2616bis/issues/index.html#i54
> >>
> > <snip>
> >
> > The modified proposal (after discussion) is as follows (should fix
all
> three problems mentioned):
> >
> > "A cache MUST NOT generate 1xx warn-codes for any messages except
cache
> > entries, and MUST NOT generate 1xx warn-codes for a cache entry
except
> > in response to a validation attempt for that entry. 1xx warn-codes
MUST
> > NOT be generated in Request messages."
>
> Explain again how any code is passed as a Request message?  The
original
> language, "MUST NOT be generated by Clients" was correct.

To answer your question:

2xx warn-codes could be added to a Request, e.g., by a proxy that is
transforming requests between the user-agent and the origin server. This
is
perfectly valid, and desirable, though likely uncommon.

To the original language is incorrect:

Clients should be allowed to generate 1xx warn-codes in Responses -- all
the
client has to do is implement a cache, and the client will need the
ability
to add 1xx warn-codes to its cached Responses. Thus, the original
language
("1XX warn-codes MAY be generated by a cache only when validating a
cached
entry. It MUST NOT be generated by clients.") is incorrect, unless the
intent is to effectively prohibit clients (including proxies) from
implementing a cache. Such a prohibition, given the context of the rest
of
the spec, is so silly as to be laughable; thus, the wording here needs
to be
corrected.

The proposed wording means that clients (and everything else) MUST NOT
generate 1xx warn-codes in Request messages (because it's nonsense).
Clients
(and everything else) can generate 1xx warn-codes in Response messages,
though cache subsystems (in clients or otherwise) have some extra rules
to
follow. There's no reason to have a blanket prohibition on clients
generating 1xx warn-codes in Responses.

On further reflection, there is an issue that I have with the above
proposed
wording: the context might imply that "1xx warn-codes MUST NOT be
generated
in Request messages" _by caches_, instead of by anything.


Thanks,

-- Travis

Received on Tuesday, 2 January 2007 20:51:51 UTC