Re: Reducing NUMERIC size for 8.3

Bruce Momjian <bruce@momjian.us> writes:
> Would someone please explain why we are considering this so far past
> features freeze, and who suggtested that the 8.3->8.4 upgrade being a binary
> upgrade was anything more than a pipe dream?

Well, Greg had left further squeezing of numerics as an open item in his
varlena patch, so it's not totally unreasonable to consider a patch for
that now --- as long as it's pretty small and simple.

I agree that in-place upgrade is a pipe dream until we see someone
actually step forward and do the work to build a usable pg_upgrade
utility.

If nothing else, we should consider swapping the n_sign_dscale and
n_weight fields now, since that would enable upward-compatible
implementation of these ideas later.  Otherwise any such patch
would probably get rejected if pg_upgrade did happen to emerge out
of nowhere.
        regards, tom lane