Re: [netmod] yang next issue #46 binary encoding support

[Juergen Schoenwaelder <[e-mail]>]

On Fri, Mar 29, 2019 at 09:30:19AM -0700, Andy Bierman wrote:
> On Fri, Mar 29, 2019 at 9:17 AM Juergen Schoenwaelder <
> [e-mail]> wrote:
> 
> > On Fri, Mar 29, 2019 at 09:07:18AM -0700, Andy Bierman wrote:
> > > On Fri, Mar 29, 2019 at 4:19 AM Juergen Schoenwaelder <
> > > [e-mail]> wrote:
> > >
> > > > Hi,
> > > >
> > > > this is issue is closed but I wonder whether this is correct. I have
> > > > several questions looking at the issue on github:
> > > >
> > > > - Why is this not a YANG issue?
> > > > - Which workaround is better?
> > > > - Why is this tagged as a NETCONF issue?
> > > >
> > > >
> > > Did you mean this should be NETCONF issue?
> > > It is more of a protocol problem then a modeling problem.
> > > The goal is to use the model unaltered.
> >
> > I think it would be valuable if say the definition of ipv4-address
> > could state that a canonical binary representation is of type binary {
> > length 4; }. Doing this is only meaningful for some types but it would
> > allow to add more binary representations over time.
> >
> > > > If we want to support binary encodings, we need to allow modelers to
> > > > define which types map to a canonical binary representation in
> > > > addition to the canonical string representation. As stated in the
> > > > issue description, hard-wiring some types in the encoding
> > > > specifications is very limited.
> > > >
> > > > In terms of backwards compatibility, this issue should IMHO be tagged
> > > > high (adding binary encoding for some types does not cause any
> > > > backwards compatibility problem since so far we have only strings).
> > > >
> > > >
> > > Not so sure.
> > > The base64 encoding could look like a valid string.
> > > The receiver must know a binary type is being sent (XML and JSON both
> > fail
> > > here, but not CBOR).
> >
> > I am talking about CBOR, not about XML or JSON. I want to provide
> > hints to CBOR (or similar binary encodings) that values can be
> > represented in a different format. I do not expect these hints to be
> > used by XML or JSON. If you need binary encoding efficiency, use CBOR
> > instead of JSON.
> >
> > > > While I do not have a solution proposal, I think this issue is worth
> > > > to look at and we should not close it right now.
> > > >
> > > >
> > > I have a solution proposal, but I have not implemented it yet, so it it
> > not
> > > detailed...
> > >
> > > Both sender and receiver need to agree on the binary encoding and how the
> > > data is tagged as binary.
> > >
> > > This expired draft should address that problem:
> > > https://tools.ietf.org/html/draft-mahesh-netconf-binary-encoding-01
> > >
> > > For every type T that they agree on, there are standard T.b2y() and
> > T.y2b()
> > > conversion functions.
> > > There are also some extensions to define conversion templates so vendors
> > > can add their own types.
> > >
> > > The YANG modules do not need to actually be altered.  The peers will
> > > negotiate the
> > > set of types that will be sent as binary when the session starts.
> > > The receiver knows T and the SID for each object, and will accept either
> > > the YANG or binary encoding.
> >
> > Sounds complex for me to negotiate this. I rather say once that a
> > binary encoding can ship an IPv6 address as type binary { length 16; }
> > and then CBOR will simply do the right thing.
> >
> >
> OK, but this would require new type names.
> You cannot simply change some standard type to be a union with a binary
> type.
>
> This forces all implementations of that type to support the binary variant.
> That breaks old clients that worked with the version before the binary
> variant.
> 
> The ripple effect on the models changing types would be non-trivial.
> Using this union-type approach forces every protocol to support the binary
> encoding,
> yet base64 in a union with strings is very error-prone.
> 

I am not proposing do change the type definitions we have. My idea was
to have an optional additional definition for binary encodings. Here
is an ad-hoc example (I do not like the details of the syntax, but
perhaps this helps to understand the idea):

     typedef ipv4-address {
       type string {
         pattern
           '(([0-9]|[1-9][0-9]|1[0-9][0-9]|2[0-4][0-9]|25[0-5])\.){3}'
         +  '([0-9]|[1-9][0-9]|1[0-9][0-9]|2[0-4][0-9]|25[0-5])';
       }
       description
         "The ipv4-address type represents an IPv4 address in
          dotted-quad notation.";

       binary-representation {
         type binary {
	   length 4;
	 }
	 description
	   "The binary representation uses as 4-byte binary string
	    in network byte ordering.";
       }
     }

The CBOR encoder (or other binary encoders) would then encode the
value as a 4 byte binary value, the XML and JSON encoder would use the
canonical string representation.  If the binary-representation is not
specified, then the generic CBOR encoding rules apply. I assume that
additional binary representation definitions will only be needed for a
couple of types (and I might even be fine to restrict that to
typedefs). Anyway, details need work, but if we want to support more
efficient binary encodings, then I think we should keep the issue #46
open.

/js

-- 
Juergen Schoenwaelder           Jacobs University Bremen gGmbH
Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
Fax:   [phone]         <https://www.jacobs-university.de/>