There is one possibly minor technical problem with making the change.
Presently some of the pgp clients display only 32 bits of the 64
bit keyid. This might be done for example if they receive a message
encrypted to or signed by a key they don't know. In that case they have
the 64 bit keyid but they sometimes only display 32 bits because that
is a more manageable size for humans. Manual key server searches have
also supported 32 bit keyids.
With keyids as the right substring of both V3 RSA moduli and V4
fingerprints, it is most logical to use the right 32 bits of the 64 bits
for this purpose, and that is how it has been done.
If some keys start using left substrings, then it would be awkward
to continue to use the right 32 bits of the keyid as the short form.
This would end up being bits 32-63 of the fingerprint. For keys that
use left substrings, it would be more appropriate to use the left 32
bits of the 64 as the short form of the keyid.
However, in the circumstances above where this arises, we don't know
the key's version number. We have only the algorithm id and the keyid.
So we can't base the decision on anything other than the algorithm id.
There are a few possible solutions.
We could eliminate the 32-bit keyid display entirely, and only show
and input 64 bit keyids. This is really a UI issue which is outside
the scope of this group, and while we have no obligation to support any
particular shortening of keyids, I think we should be aware of the impact
our decisions make on other parts of the system.
We could make the decision about which way to shorten it depend entirely
on the key algorithm, so that we'd need a new algorithm identifier for
RSA keys which used the new convention.
We could make new signature and pkesk packets which include more
information about the keyid, enough to allow it to be shortened
unambiguously (and perhaps supporting variable length keyids).
Or of course we could leave it as it is now. I am still not convinced
that the universe is constructed such that left substrings are inherently
more natural than right ones. Endian wars, anyone?
Hal