RFC on issue 2701, default expiration time for new keys

[Robert J. Hansen]

> I'd not say "THE best practices document", but rather "A RANDOM best
> practices document someone wrote". There are lots of those, and can freely
> be ignored, IMNSHO.

I'd go one step further: this is not even a best-practices document.  Any
document which claims to be a best-practices document which does not have,
as a high-priority item, "Figure out your threat model," is frankly just
somebody pretending to be competent.

[puts on FAQ maintainer hat]
Every now and again I get someone asking me why there aren't best practices
listed in the FAQ.  The answer is always the same: because GnuPG doesn't
really have them.  GnuPG is a toolkit you use to implement part of your
custom solution to address your particular threat model.  As such, the idea
of "best practices" that are one-size-fits-all is really kind of silly.
There might be some merit in a "Things To Think About" document, but the
idea of a single best-practices document that applies to everyone everywhere
borders on the absurd.
[takes off hat]

> This document also recommends creating a 4K RSA key. And it complicates
> matters by strongly recommending installing parcimonie and Tor over just
> using --refresh-keys. That's one more hurdle for users to overcome in an
> already very complicated matter, and as such, IMNSHO, it is actually
> hindering user adoption.

It also recommends ignoring things like the keyserver-url field on a
certificate.  Which is ... interesting.  If Alice works for a company that's
rolled out GnuPG, the company may have its own LDAP server with
up-to-the-minute revocations.  And the company may wish you to fetch Alice's
certificate from it, in order to get up-to-the-minute details, as opposed to
getting it from the keyserver network, which the company doesn't sync with.

So if you follow these "best practices", you'll actually never get to update
Alice's certificate, even when it's revoked after she leaves the company...

Also, don't get me started on "Primary keys should be DSA2 or RSA (RSA
preferred)".  Right, like there's some inherent problem with DSA2 that makes
RSA such a superior choice...