Bruce Momjian <bruce@momjian.us> writes:
> Agreed, but my guess is that we are going to introduce shorter varlena
> headers for 8.3. It will hard to reject an optimization like that, and
> that will probably change the disk format for most columns.
Well, several of the proposals that have been made would not cause
existing disk images to become broken --- in particular, the idea of
introducing separate "short" datatypes without touching the existing
ones would have that merit. So we might want to factor that point into
our choices about what to do.
This is all pretty pointless unless someone actually writes a pg_upgrade
tool, but maybe we'll see one for 8.3.
regards, tom lane