Re: Reject access to unlogged tables in contrib modules

Hi Jim,

I tested v5 on top of 510512876f9 with a primary/standby pair. Both
patches apply cleanly, build without warnings, and the pageinspect and
pgstattuple regression tests pass. On the standby, every pageinspect
and pgstattuple entry point I tried reports the recovery error for
unlogged relations, the init fork of an unlogged index is still
readable, and the detoast cases behave as in v4. On the primary,
other-session temp relations are still rejected, and pg_relpages() still
works for them.

One behavior change worth noting is with the relkind check in
check_relation_accessible(), pgstatindex() and pgstattuple() on a
temporary index of another session now report "cannot access temporary
indexes of other sessions" instead of "... tables ...", which seems more
accurate to me.

On pg_freespacemap, I agree with Michael's point, without an FSM fork, 0 is
also what the owning session sees, and with one, the other-session case
already fails in bufmgr. Dropping 0006 is fine.

Overall v5 looks good to me.

Regards,
Surya Poondla