Re: [YANG] default values

Juergen Schoenwaelder wrote:
> On Thu, Jan 17, 2008 at 04:01:50PM +0100, Ladislav Lhotka wrote:
>  
>> At the start of a NETCONF session, the values in the
>> running/startup/whatever configuration may be different even if the
>> server's factory defaults follow the standards. The client can't IMO
>> safely avoid checking or setting all variables in server's
>> configuration.
> 
> Sure. But when I create something new without specifying all defaults,
> do I know what I get? In SNMP, you don't. In YANG, you do.
>  
>>> The alternative would be to have a mechanism allowing implementations
>>> to post the defaults they are going to use rather than hard wiring
>>> defaults in the specification. So a vendor can augment a standard with
>>> the defaults that the vendor's implementation is using. This surely
>>> adds some complexity. An even further reaching mechanism would be to
>>> make the defaults being used by an implementation itself writable,
>>> e.g. by introducing a template mechanism which adds defaults for stuff
>>> not defined explicitely by a NETCONF client.
>>>
>>> Our goal was to start with something simple, allowing us to add more
>>> powerful mechanism via extensions over time once we see that they are
>>> really needed (that is there is enough uptake of NETCONF / YANG).
>> I think this applies to the default statement, too. I am arguing it's
>> not needed for NETCONF operation and may complicate things.
> 
> So if I understand right, your proposal is to remove the default
> statement, not to change its semantics, correct?
> 

I strongly object to removing the default-stmt.
As Balazs pointed out, if a WG does not agree on a suitable
default value, they do not have to define one.  This clause
is needed to support NETCONF, not only to provide as much
machine-readable info about the data model characteristics,
but to simplify NETCONF tasks for operators and applications.

There are likely to be many knobs in a real config file that are optional
and have suitable agent-supplied default values.  It is very useful
for operators and applications to know this information.

There are some cases where the agent implementation varies from
the data model specification.  This does not mean we should throw
out 'default'.  It means we need to address agent conformance
and agent capabilities (Something YANG ignores completely right now.)

> /js
> 

Andy

_______________________________________________
YANG mailing list
YANG@ietf.org
https://www1.ietf.org/mailman/listinfo/yang