Re: [Cellar] AD Review: draft-ietf-cellar-ffv1

[Michael Niedermayer <[e-mail]>]

On Thu, Jan 09, 2020 at 03:14:26PM -0600, Adam Roach wrote:
> On 1/9/20 12:23 PM, Michael Niedermayer wrote:
> >The implementation definedness can be avoided by simply refering to an implementation
> >and not just "C90" or other. not sure "twos complement" is strictly enough but it
> >may be the cleanest way to specify it.
> 
> 
> I don't think "two's complement" clearly indicates sign extension versus
> zero padding. I think you need to clearly specify precisely what happens to
> new high-order bits during a right-shift.

thats what i meant by "strictly enough".
here a gradient from clear to precisse to clarify more what i meant:
A. >> is defined as on common mainstream two's complement architectures
B. >> with signed values is defined with two's complement representation
   and sign extension
C. a >> b with signed a is defined so that the two's complement representation
   of a is shifted toward the less significant bit by b places. The newly
   introduced bits are identical to the previous sign bit of a. The bits
   shifted beyond the least significant bit are discarded. If b is negative
   the behavior is undefined.
   
when reading the text for A., i think its clear to everyone knowing ISO-C
while C. is more precisse it requires more effort to understand that this
is mostly what is done by ISO-C on mainstream architectures.
I guess we probably should combine something like A and C so its both
precisse and still can be immedeatly understood without the need to
disentagle the definition

Thanks
   

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Avoid a single point of failure, be that a person or equipment.