Re: Weird behaviour with ALTER TABLE ... SET TABLESPACE ... statement

Tom Lane wrote:
> Heikki Linnakangas <heikki(dot)linnakangas(at)enterprisedb(dot)com> writes:
>> Yeah, seems like we need to allocate a new relfilenode in the new
>> tablespace.
>
> I looked into tablecmds.c and verified that ATExecSetTableSpace doesn't
> worry about selecting a new relfilenode. I'm also noticing a number of
> permissions-type checks that seem like they'd better be done in
> ATPrepSetTableSpace, because we don't go through ATExecSetTableSpace
> if the table requires rewriting for other reasons.

The same tests are performed in the rewriting code path in
ATRewriteTables() and in heap_create_with_catalog().

I fixed the relfilenode allocation in 8.1-HEAD. Doesn't seem worth
fixing in 8.0, because GetNewRelFileNode() didn't exist before 8.1, so
we couldn't check for collisions anyway.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com