Re: Comments on draft-hollenbeck-ietf-xml-guidelines-02.txt

[Graham Klyne <GK-lists@xxxxxxxxxxxxxx>]

Protocols (in the sense that I understand the draft to be directed at) are 
about specifying specific exchanges of bits-on-the-wire to achieve some 
desired goal of information exchange.

Bits on the wire are the common ground between protocol 
participants.  Interoperability requires that a receiving application 
properly understands the bits-on-the-wire that are sent.  Everything 
eventually comes down to this.  By using a higher level specification that 
less directly constraints bits on the wire, I think there are two potential 
problems:
(1) greater opportunity to get it wrong (ASN.1 had 3 or 4 different 
possible wire formats, I think).
(2) the protocol implementer has to digest a greater volume of 
documentation in order to correctly implement the protocol.

Against this, I see no real advantage for using an abstract format 
specification for a protocol.  What is the benefit of having an abstraction 
here?

(Interestingly, and this is off topic so I won't expand, I'm not sure that 
XMLP falls into the same category, and I won't way that infoset based 
specifications would not be appropriate for XMLP applications.)

#g

At 03:10 PM 5/16/02 +0200, Jean-Jacques Moreau wrote:
> Thanks for your comments, and sorry for the delay, I have been away and am 
> catching
> up with email.
>
> I'd be curious as to the reasons for which you, Tim and Graham think it is 
> a bad
> idea to use the XML Infoset for a protocol. I'd be happy if you could 
> expand your
> reasonning further. I, for now, have seen this rather positively.
>
> Maybe we should cc: dist-app in the future. The folks there would likely be
> interested in that discussion.
>
> Jean-Jacques.
>
> Martin Duerst wrote:
>
> > I agree with Graham.     Regards,   Martin.
> >
> > At 16:12 02/05/07 +0100, Graham Klyne wrote:
> >
> > >At 04:25 PM 5/7/02 +0200, Jean-Jacques Moreau wrote:
> > >>What I am really saying is that we would be better off in the long run
> > >>recommending XML Infoset over XML 1.0. Amongst other things, this 
> gives you
> > >>serialization independence.
> > >
> > >I can imagine many situations in which that is a real advantage, but
> > >protocol specification isn't one of them ;-)
> > >
> > >(I'll hazard that one of the problems with ASN.1 in protocol design was
> > >the fact that the wire format  wasn't obvious from the specification 
> document.)
> > >
> > >#g
> > >
> > >
> > >-------------------
> > >Graham Klyne
> > ><GK@xxxxxxxxxxxxxx>
> > >

-------------------
Graham Klyne
<GK@xxxxxxxxxxxxxx>