Re: [Gen-art] Gen-ART Last Call review of draft-josefsson-scrypt-kdf-03

[Paul Kyzivat <[e-mail]>]

I did want to comment on one other thing:


 >>> (Issue-4): Identifiers reused for different meanings
>>>
>>> In scrypt, "B" is an array of "p" vectors, each of which is 128*r
>>> octets. In scryptROMix, "B" is a single vector of 128*r octets. In
>>> scryptBlockMix, "B" is a vector of 2*r 64-octet blocks.
>>
>> I'm not sure that changing variable names here is a good idea.  For any
>> performant implementation, these variables will refer to the same memory
>> area and thus a consistent variable name helps to signal that.  It is
>> just the interpretation of that memory area that is different in these
>> functions.

>>> The document would be clearer if distinct identifiers were used for
>>> each unique concept.
>>
>> I believe the identifier refers to unique concepts.  For B, what differs
>> is how that memory area is interpreted in each algorithm description.

I don't think it works for a *spec* to assume that a "memory area" can 
be interpreted differently, sometimes as a number, sometimes as an array 
of bytes, sometimes as an array of arrays, etc.

Making such assumptions depends on the language being used mapping these 
things to the same memory in a consistent way. Some languages don't 
allow this at all, and those that do often lay out memory with padding 
for efficiency, according to little/big endianness, etc.

So I think it is important to clearly define any type coercions that are 
needed.

	Thanks,
	Paul