RE: analysis of YANG vs. RELAX NG

["David Harrington" <[e-mail]>]

I agree with Juergen, Andy, and Randy.

Our goal is interoperability between multiple managed devices and
multiple management applications - we need system-wide
interoperability for M:N relationships. 

Netconf has different requirements than SNMP, and a Netconf DML will
have different requirements than SMI, but NM requirements seem
relatively stable on this point.

A MIB module is a type of contract between the involved parties, and
it is unacceptable to modify the syntax or semantics in ways that
would break contract-compliance/interoperability between the parties.
The SMI does not allow changes to existing datatypes, with very few
exceptions. SMI does permit some extensibility, such as extending
enumerations or BITs, but the change controls for the exceptions are
clearly defined in the SMI. Those restrictions exist to ensure
interoperability among NM entities within the system. 

David Harrington
[e-mail]
[e-mail]


> -----Original Message-----
> From: Randy Presuhn [mailto:[e-mail]] 
> Sent: Monday, December 03, 2007 12:02 PM
> To: [e-mail]
> Subject: Re: analysis of YANG vs. RELAX NG
> 
> Hi -
> 
> > From: "Ladislav Lhotka" <[e-mail]>
> > To: "Andy Bierman" <[e-mail]>
> > Cc: <[e-mail]>
> > Sent: Monday, December 03, 2007 10:02 AM
> > Subject: Re: analysis of YANG vs. RELAX NG
> >
> 
> >
> > Andy Bierman píše v Po 03. 12. 2007 v 09:21 -0800:
> >
> > > I strongly agree with Juergen.
> > > YANG is based (in part) on 18+ years experience
> > > with SNMP and SMI.
> >
> > And ignoring the additional flexibility XML can provide.
> 
> I strongly agree with Andy and Juergen.  "Mushy" definitions
> are disastrous for interoperability.
> 
> > > It s absolutely forbidden in NM to redefine the syntax 
> and semantics
> > > of a managed object in this way.
> 
> Agreed.
> 
> > Why? Even if I explicitly specify I am using another data model?
The
> > difference is that if the data model is easily extensible, 
> I can just
> > write
> >
> > import parent-model;
> >
> > and then 5 lines or so describing the differences, so that it is
> > immediately clear what I am doing (and hopefully why). If the
> > extensibility is weak, I have to take the parent model, 
> change few lines
> > and declare it as a new data model - but the differences 
> are not that
> > clear.
> 
> Re-use of definitions in the process of creating *new* definitions
is
> another matter.  The important point is that those new definitions
in
> no way affect the models on which they were based, and therefore
must
> not be understood as a redefinition of either the semantics or the
> syntax of a managed object.
> 
> Randy
> 
> 
> 
> 
>