Re: Prototype: In-place upgrade v02

Zdenek Kotala <Zdenek(dot)Kotala(at)Sun(dot)COM> writes:
> Another idea is to create backward compatible AM and put them into separate
> library. If these AM will work also with old page structure then there should
> not be reason for reindexing or index page conversion after upgrade.

I don't think that'd be real workable. It would require duplicating all
the entries for that AM in pg_opfamily, pg_amop, etc. Which we could do
for the built-in entries, I suppose, but what happens to user-defined
operator classes?

At least for the index changes proposed so far for 8.4, it seems to me
that the best solution is to mark affected indexes as not "indisvalid"
and require a post-conversion REINDEX to fix 'em. Obviously a better
solution would be nice later, but we have to avoid putting huge amounts
of work into noncritical problems, else the whole feature is just not
going to get finished.

regards, tom lane