Re: [YANG] errors

Martin Bjorklund wrote:
> Andy Bierman <ietf@andybierman.com> wrote:
>>> The idea was (I think it came from you :) that the identifier for the
>>> error clause is error-app-tag.  So the above would be in XML:
>>>
>>>   <rpc-error>
>>>     <error-tag>operation-failed</error>
>>>     <error-app-tag>lock-denied</error-app-tag>
>>>   </rpc-error>
>>>
>>> Thus, no name clash.
>>>
>> huh?
>>
>> then how do you specify a 'raise-error' for the lock-denied in RFC 4741?
> 
> You don't.  Unless you also have an error-app-tag:
> 
>   error my-extra-lock-denied {
>     error-tag lock-denied;
>   }
> 
> The idea was that really do not want to list all std errors that can
> occur:
> 
>    rpc my-rpc {
>      ...
>      raise-error missing-element;
>      raise-error bad-namespace;
>      ...     
>    }
> 
> These std errors can always occur.  No need to specify them again.
> 

okay, that's fine.

> -----
> 
> We also disucssed an alternative way to do this, which is slightly
> more flexible.  In this variant, the error is identified with an
> identifier, which has no meaning on the wire.  It's just a local
> identifier like a typedef or grouping identifier.  Then within the
> error, you have the statement 'error-app-tag' in addition to the other
> stmts:
> 
>    error my-error {
>      error-tag invalid-value;
>      error-app-tag bad-mtu;
>    }
> 
> 

I like the current design better because the key into the
error dictionary has meaning, but there is a problem...

There is still at least one open issue, which is that error-app-tag
is defined in RFC 4741 as a string and it should really be a QName.
Each error definition will only be unique within its module, so
error-app-tag is not globally unique as it should be (according
to the RFC).

IMO, this should be solved by calling it a bug in the RFC
(along with the bad-namespace element, which is a QName and MUST
be an Xpath string, not a QName at all), and encoding the
error-app-tag as a QName. The module namespace URI identifies
the prefix in an XML instance document.

Your 'raise-error' examples always show a local identifier reference
(i.e., no prefix), but the real syntax is an identifier-ref-string
(called a node-identifier in -00), so the error does not have to
be in the same module as the raise-error statement.

> 
> /martin
> 
> 

Andy

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