On 19/01/11 8:07 AM, Daniel Kahn Gillmor wrote:
On 01/18/2011 12:48 PM, Jon Callas wrote:
If we combine it with a hash-independent fingerprint -- e.g., first byte is an
algorithm ID, others are the actual hash -- then we can put it in now and then
run with it.
Daniel Nagy suggests that we also change the material being hashed in
the fingerprint (he wants to leave out the creation date). If that
proposal ends up achieving consensus then your proposal here will need
further modification.
Does anyone feel strongly about Nagy's proposal, by the way? i'm not
sure what the tradeoffs are.
Even without that concern, if we encourage super-flexible use like this,
user agents who wanted to use it to test for the presence of a given key
in an indexed keystore would need to index their keys with every
possible digest that might be used -- that seems excessive somehow.
(and unlikely that keyserver implementations would want a half-dozen
indexes, for that matter) Wouldn't it be better to just implement it
for today's fingerprint, and then when a new fingerprint is agreed upon,
determine (by subpacket length maybe?) whether it's the new fingerprint
or the old one. Compliant user agents would keep the two indexes around
until the v4 fingerprint goes away, and then drop the old one.
Alternately, we could embed the algorithm ID as you suggest, and SHOULD
people into generating them using only the consensus fingerprint
algorithms so that reasonable user agents only need to create indexes
over SHA1 (now) and SHA3 (whenever that happens).
The style of OpenPGP v5++ has been algorithm agility, and getting away
from that is likely to take time, and/or bravery.
Which leads me to agree that the fingerprint should include the leading
algorithm Id, because that's how we do message digests. And in text, we
browbeat the developers to stick to the "the one."
(In which, the one will be two, SHA1 then SHA3, until we can ditch the
other "one".)
On the other hand, there is that hanging question: do we want algorithm
agility in the future? What did we ever win by it? Did it ever meet
its promise?
If the answer is in the negative, I'd be inclined to ignore the whole
packet question and start thinking instead about a completely new layout
generation that simplified the coding requirements back to something ...
PGP 2.3 like.
iang
PS: at least, that's my question & hobby horse. Others have other
questions and hobby horses.