ietf-mailsig
[Top] [All Lists]

Re: Why we really don't require requirements

2004-10-02 20:25:33

The problem here appears to be that by requirement, I understand
something that has to be accomplished by the final product, while
other people mean other things.

My understanding pretty much agrees with yours. A requirement is, well,
a requirement. My dictionary defines this using terms like "absolute
need" and "mandatory".

And in the present case, I think there's only two requirements here that
matter: The scheme has to be deployable and provide some minimum level of
effectiveness. (I note that even these simple requirements are somewhat in
conflict, since it seems that the more effective  and general a signature
scheme is the less deployable it is.)

If by requirement we mean something
that we'd like to have but we're willing to drop if we agree that it
would be better to do something else, then I think we can easily agree
on a list of them.  But if that's what we mean, I'd appreciate it if
we could use a phrase whose common meaning is closer to what we mean,
e.g., "desirable characteristic".

Agreed.

Requirements definition is found in more than just the waterfall
software development methodology.  Are you seriously trying to
suggest that requirements for software development is just a
35-year-old, faded and debunked fad?

Trying to develop fixed requirements in advance of development has
failed over and over and over again execpt in an extremely tightly
constrained environment, which e-mail sure isn't.  The processes that
work are more like spirals, going between design, implementation, and
evaluation.  But since we're trying to create a standard rather than
write a piece of software, we won't know for quite a while what the
results of future implementation and evaluation are.  (I guess that's
why it works better to start with stuff that's been implemented, so we
have at a few turns of the spiral to look at.)

An approach that used to be common in the IETF was to build alongside
the specification process. Sadly, it seems that few have the resources
to do this any more.

                                Ned


<Prev in Thread] Current Thread [Next in Thread>