RE: Returning zero or empty string for unsupported objects

[Andy Bierman <abierman@cisco.com>]

At 10:49 PM 11/6/2003, Wijnen, Bert (Bert) wrote:
>Maybe I was not clear enough
>
>The case I was wondering about is NOT if it is OK to sometimes
>design a special value in a MIB definition that indicates that
>an agent can return that value to say "not supported".
>
>What I am talking about here is that some MIB implementation
>of a random MIB is just deciding to return zero or empty strings
>because it does not feel like supporting one or more objects
>in that MIB module.

Clearly it is wrong for an agent to ignore the semantics of 
a MIB object and invent new semantics for any portion of the
object's valid values.  It's just as bad to return values
which are invalid according to the SYNTAX clause.


>Thanks,
>Bert 

Andy



>> -----Original Message-----
>> From: Randy Presuhn [mailto:randy_presuhn@mindspring.com]
>> Sent: vrijdag 7 november 2003 7:08
>> To: Mreview (E-mail)
>> Subject: Re: Returning zero or empty string for unsupported objects
>> 
>> 
>> Hi -
>> 
>> > From: "C. M. Heard" <heard@pobox.com>
>> > To: "Mreview (E-mail)" <mreview@ops.ietf.org>
>> > Sent: Thursday, November 06, 2003 5:54 PM
>> > Subject: Re: Returning zero or empty string for unsupported objects
>> >
>> 
>> > Greetings,
>> >
>> > I think Dave's analysis is right on ... it is not necessarily wrong
>> > for an object definition to specify that a special value be returned
>> > to indicate that a function isn't supported.  If we did make such a
>> > blanket rule I think we'd be guilty of creating an unnecessary rule.
>> > I say let this remain a design choice.  (Of course, it's always
>> > fair for a reviewer to ask a designer to consider alternatives ... I
>> > just don't think this one should be elevated to the status of a
>> > SHOULD or MUST.)
>> ...
>> 
>> I think the underlying principle is that the agent should not lie.
>> If the object definition provides a "not supported" value, then
>> an agent reporting that value isn't lying.  If the object definition
>> (or underlying data type) does not provide for such a sentinel value,
>> then instantiating the object will lead management applications to
>> draw erroneous conclusions.
>> 
>> Randy (who just saw a MIB where the WG wanted interface counters
>> to count or return constant zero depending on what chip set
>> was used in the implementation.)
>> 
>> 
>>