On Fri, Jan 25, 2013 at 02:22:40PM +0100, Andres Freund wrote:
> On 2013-01-24 20:45:20 -0500, Bruce Momjian wrote:
> > On Sun, Sep 2, 2012 at 05:40:54PM +0000, james@illusorystudios.com wrote:
> > > The following bug has been logged on the website:
> > >
> > > Bug reference: 7515
> > > Logged by: James Bellinger
> > > Email address: james@illusorystudios.com
> > > PostgreSQL version: 9.1.5
> > > Operating system: Ubuntu Linux 12.04 Server
> > > Description:
> > >
> > > If the table being referenced has a schema in its name, and the schema does
> > > not exist, DROP TABLE IF EXISTS will have an ERROR instead of a NOTICE.
> > >
> > > So for instance,
> > > DROP TABLE IF EXISTS bar;
> > > This is a NOTICE if bar does not exist.
> > >
> > > DROP TABLE IF EXISTS foo.bar;
> > > This is an ERROR if foo does not exist, even though that implies bar does
> > > not exist which means it should be a NOTICE.
> > >
> > > Saw this because it was making a drop/recreate transaction fail on me, after
> > > I changed some code to use a schema.
> >
> > I looked at this bug report from September. The problem is that
> > LookupExplicitNamespace() doesn't have a missing_ok parameter, even
> > though get_namespace_oid(), which it calls, does. By adding a
> > missing_ok parameter and passing it cleanly, I fixed the problem:
> >
> > test=> DROP TABLE IF EXISTS foo.bar;
> > NOTICE: table "bar" does not exist, skipping
> > DROP TABLE
> >
> > Patch attached.
>
> That makes it a HEAD only fix though. I personally don't have a problem
> with that, but others might...
This turns an error into a notice/success. I don't think we would ever
backpatch such a behavioral change. I realize that nothing changes
either way in the database, but applications checking for return status
might care about it.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ It's impossible for everything to be true. +