Bruce Momjian wrote:
> > Bruce Momjian wrote:
> >
> > > > The second option (your earlier suggestion) seems to be necessary and sufficient. The junk filter (and
> > > > jf_cleanTupType) will always exist, for SELECT statements, as long as the following is not a legal statement:
> > > >
> > > > SELECT FROM foo GROUP BY bar;
> > > >
> > > > Currently the parser will not accept it. Sufficient.
> > > >
> > > > The first option will set tupType, for non-SELECT statements, to something it otherwise may not have been.
> > > > I would rather not risk effecting those calling routines which are not executing a SELECT command. At this
> > > > time, I do not understand them enough, and I see no benefit. Necessary?
> > >
> > > OK, I will leave it alone. Is there a way to use junk filters only in
> > > cases where we need them?
> >
> > I have not YET come up with a clean method for detection of the a resjunk flag being set, on some resdom in the
> > tatget list, by a GROUP/ORDER BY. I will give it another look. It does seem a bit heavy handed to construct the
> > filter unconditionally on all SELECTS.
>
> David, attached is a patch to conditionally use the junk filter only
> when their is a Resdom that has the resjunk field set. Please review it
> and let me know if there are any problems with it.
>
> I am committing the patch to the development tree.
I did not get any attached patch. ??? I can check it out at home where I have cvsup.
Where there any confirmed problems cause by the aggressive use of the junkfilter? I ask because, adding this extra
check probably will not resolve them. It may only reduce the problem.
I was planning on including an additional check for resjunk as part of another patch I am working on. (GROUP/ORDER
BY
func(x) where func(x) is not in the targetlist) Graciously accepted.