Re: [YANG] default values

Ladislav Lhotka wrote:
>> Yes, changing defaults is incompatible.  If you do that, you have to
>> change the namespace.
> 
> BTW, that's exactly why I am against the default values in typedef: they
> make the data model too specialised. The defaults could be easily
> handled as follows:
> 
> 1. The standard that defines the type specifies in plain RFC-English
> that the default value MUST, SHOULD or MAY be something.
> 
> 2. An implementor reflects this in a device, but is also required to
> fill the default (or another) value in the initial startup
> configuration.
> 
> 3. The client needn't care about what the default is, it just queries
> the device to see the actual value, changes it if necessary etc.
> 

I disagree 100%.
The management of agent-supplied defaults has always been a problem
in NE devices.  More machine-readable support is needed, not less.

YANG has the "DEFVAL with teeth" that SMIv2 never had.
In YANG the default MUST be supported by the agent.
It is not just a suggestion.  This means the manager can
actually pay attention to the default defined in the data model,
and change it only if needed.

On large NE devices, it is common practice to keep the CLI config file
as small as possible by only including non-default values.  NETCONF
needs to support this feature as well, and YANG leaf/typedef default does that.

> This approach would make the same data model reusable with other default
> value and, in addition, allow for honouring the semantics of MUST,
> SHOULD and MAY. The default in typedefs is only usable in the MUST case.
> 
> So my suggestion is to remove the default statement form YANG.
> 
> Lada
> 
> 

Andy

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