W3C home > Mailing lists > Public > ietf-http-wg-old@w3.org > January to April 1995

Re: v10-spec-00 comments

From: Owen Rees <rtor@ansa.co.uk>
Date: Fri, 24 Mar 1995 13:48:18 +0000
Message-Id: <9503241348.AA06845@plato.ansa.co.uk>
To: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
"Roy T. Fielding" <fielding@avron.ICS.UCI.EDU> writes:
> > > 4.3.3 Message-ID
> > 
> >   The first paragraph specifies that this is a unique identifier
> > for the message, presumably the HTTP request or response
> > message. The final paragraph says that an HTTP response should
> > only include a Message-ID header if the entity has one. Since it
> > is possible to retrieve a Mail/News message more than once, the
> > HTTP Message-ID cannot be the Message-ID of the enclosed entity as
> > this would violate the unique identification property for the HTTP
> > response message.
> 
> No it would not.  There is only one message in that case.  This situation
> is identical to the handling of messages by proxies.

Here is the problem scenario:

Charles sends one piece of e-mail to both Alice and Bob, Charles' MUA assigns 
Message-ID M to this message.
Alice retrieves her e-mail via proxy AP1 to gateway AG.
Bob retrieves his e-mail via gateway BG.

HTTP traffic (without authentication):

1  Alice->AP1 request  GET uri-for-the-mail(via-AG)
2  AP1->AG    request  GET uri-for-the-mail; Forwarded: AP1 for Alice
3  AG->AP1    response Message-ID: M3 <the mail>
4  AP1->Alice response Forwarded: AP1 for AG;Message-ID: M4 <the mail>
5  Bob->BG    request  GET uri-for-the-mail
6  BG->Bob    response Message-ID: M6 <the mail>

Note that M3=M4=M6=M and "is a unique identifier which can be used for 
identifying the message (not its contents)" if the implied copying of M out of 
the mail headers is done.

AP1 is supposed to add a Forwarded header so message 4 is not the same 
"structured sequence of octets" (i.e. message as defined in 1.3) as message 3.

Since the email arriving at AG and BG will have different Received headers (if 
transmitted by SMTP), messages 3 and 6 cannot have the same "structured 
sequence of octets" unless these headers are stripped out, along with anything 
else that may have been added (what if Bob's copy was Resent-To him by Diana, 
those headers had better go too!).

Even if AG and BG are the same host and the mail transport notices this and 
sends one copy so that the Received headers have the same timestamps, the 
sameness requirement prohibits the use of the Digest-MessageDigest header in 
the response if digest access autentication is being used in the interaction 
with the gateway.

Tracing messages through chains of proxies has been suggested as a reason for 
having Message-ID, but consider what happens if Bob also uses AP1, but as a 
proxy to BG. We now have a situation where AP1 handles two response chains, 
AG->AP1->Alice and BG->AP1->Bob, but both are labelled with the same 
Message-ID.

> > 7.1.1 Allow
> > 
> > I would have classified this as a response-header rather than an 
> > entity-header; it is required with 405 Method Not Allowed and will not be 
> > meta-information about the entity containing an explanation of the error.
> 
> Unless it is included in a PUT or POST of a new resource.

PUT perhaps, but not POST. In a POST, the entity body is to be made 
subordinate to the resource identified by request-URI. It may be convenient to 
simultaneously attach a subordinate and specify methods that are to be allowed 
on the resource to which it is to be subordinate, but if this is the intent it 
needs to be made clearer. If the Allow is intended to apply to the entity-body 
in a POST, then the URI-header rather than the Request-URI identifies the 
resource.

With PUT, Allow would be an entity-header specifying the methods to be 
supported. If this is a current use, then it should be explained.

In a 405 response, Allow is meta-information about the resource identified by 
the Request-URI, it is not meta-information about the entity-body in the 
response.

There seem to be two distinct uses of Allow: information from a server about 
an existing resource, and request parameter from a client when sending a 
resource. The union of these usages does not fit the header taxonomy, so I 
assume that using the same header name for both purposes is a consequence of 
current usage.

Regards,
  Owen Rees <rtor@ansa.co.uk>
Information about ANSA is at <URL:http://www.ansa.co.uk/>.
Received on Friday, 24 March 1995 05:55:16 EST

This archive was generated by hypermail pre-2.1.9 : Wednesday, 24 September 2003 06:31:14 EDT