Re: Working group last call for -reg-

[Al Gilman <asgilman@xxxxxxxxxxxxxxxx>]

[Off-topic warning; this message does not suggest that anything
in any of the current drafts needs to be changed AFAIK.]

I almost agree with you.  Not to decide is to decide.

But this level of exception handling is "response to certain
exceptions" which is not the same as "default values for
expressions of known type and unknown tag."  The latter is not
safe.

The two exceptions of interest are

	- message does not mention feature of tag expected by agent
	- message mentions feature with tag unknown to agent

Boolean features are the clearest example that if either agent
in the protocol doesn't comprehend that feature, any boolean
value is irrelevant and dangerous.  A fail-safe response not
requiring a value for that feature must be determined.

To safely apply any default _value_ for a tag that the message
doesn't mention, one needs trustworthy indication by some other
means that the sending agent did know about that tag and is
tacetly agreeing to the tag-defined default.  If the sending
agent didn't understand the question, don't assume that there is
any answer that they are agreeing to.

Al

to follow up on what Koen Holtman said:
> From owner-ietf-medfree@xxxxxxx  Fri Jul 24 12:05:58 1998
> From: koen@xxxxxxxxxx (Koen Holtman)
> Message-Id: <199807241557.RAA09871@xxxxxxxxxxxxxxxxxxx>
> Subject: Re: Working group last call for -reg-
> To: GK@xxxxxxxxxxxxxx (Graham Klyne)
> Date: Fri, 24 Jul 1998 17:57:46 +0200 (MET DST)
> Cc: koen@xxxxxxxxxx, ietf-medfree@xxxxxxx
> In-Reply-To: <3.0.32.19980722075831.007e7bb0@xxxxxxxxxxxxxxxxxx> from "Graham Klyne" at Jul 22, 98 12:11:54 pm
> X-Mailer: ELM [version 2.4 PL23]
> Sender: owner-ietf-medfree@xxxxxxx
> Precedence: bulk

> 
> [Not about the -reg- draft]
> 
> Graham Klyne:
> >
> [...]
> >Koen,
> >
> >I think where we differ on this point is that I take a view that in the
> >absence of any mention in a feature set description, *either* value (TRUE
> >or FALSE) can be matched.  Hence the idea of a default value is not required.
> 
> Again, I would call this specifying a default value: you speficy the
> default result if the boolean equality operator is applied to the
> value of an unknown tag.
> 
> I don't think you can ever avoid specifying some kind of default
> behavior for cases where a party in the negotiation process comes
> across tags which it does not understand.  Even 'abort with an error'
> is a kind of default behaviour.  The default may not be at the level
> of tag values, as it is in TCN for boolean type tags, but it has to be
> somewhere.
> 
> >One of the problems in choosing a default value is clarifying whether it
> >applies to (say) the sender or the recipient.  From the sender's PoV, it
> >could be "does this resource *use* the <blob> feature?.  For the recipient,
> >it may be "does this agent *support* the <blob> feature?".  Now, if we are
> >say that the "default" value used in absence of any explicit mention is a
> >value that works with any value (or none) that may specified by any other
> >party to the exchange, then I think that we are in agreement.
> 
> I don't understand what you mean with 'works with any value (or none)
> that may be specified', so I suspect we do not agree.
> 
> >#g
> 
> Koen.
>