Re: [codec] Codec proposal: draft-valin-celt-codec-00.txt

[Gregory Maxwell <[e-mail]>]

Stephan Wenger [[e-mail]] wrote:
[snip]
> number of lines of code.  At least during the I-D phase, a reference to an
> URL where one can retrieve a stable version of the code, plus a promise to
> copy-paste that code into a final RFC if that were the consensus, seems to
> suffice.
[snip]
> Third, it would be helpful if you would explicitly state what part of your
> I-D actually is the normative specification: source code, or textual
> description.  
[snip]

I'm strongly in favour of leaning heavily on the source code as the normative specification: Some parts, specifically any parts involving precise integer arithmetic, are very difficult to describe in english with the necessary precision without creating a difficult to understand convoluted mess.

But at the same time, if the source is the normative specification I think it would be inappropriate to leave it out of the draft even temporarily. 

There is also the issue that not all of the source is expected to have the same degree of mandatory compliance.  A convention for distinguishing these parts will probably need to be adopted. Currently the text denotes operations which must be bit-exact, so perhaps that is sufficient.