Re: Composite Datums containing toasted fields are a bad idea(?)

Merlin Moncure wrote:
> On Mon, Apr 21, 2014 at 10:40 AM, Andres Freund <andres(at)2ndquadrant(dot)com> wrote:

> > And too bad that infomask bits are so scarce :(. We really need to
> > reclaim HEAP_MOVED_OFF and HEAP_MOVED_IN.
>
> The only consequence of that is losing support for in-place update for
> pre-9.0 (of which the only supported version is 8.4). I figure it's
> also pretty reasonable to drop support for IPU for out of support
> versions for new versions going forward. That would recover the bits
> and yield some nice cleanups in tqual.c.

There's no way to be sure that the bits have been removed from tables
that were upgraded from a 8.4 server to a 9.0 server and from there to a
newer server. We'd need some way to catalogue which tables have been
cleaned prior to an upgrade to a release that no longer accepts those
bits; pg_upgrade would then say "table xyz needs to be vacuumed before
the upgrade" in check mode, or something like that.

--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services