Re: That killer 3rd join... - Mailing list pgsql-general

From Tom Lane
Subject Re: That killer 3rd join...
Date
Msg-id 1683.968304770@sss.pgh.pa.us
Whole thread Raw
In response to Re: That killer 3rd join...  (Stephan Szabo <sszabo@megazone23.bigpanda.com>)
List pgsql-general
Stephan Szabo <sszabo@megazone23.bigpanda.com> writes:
> As a separate issue I can't even do the query listed below on my
> machine with reasonably current sources, as it fails after a while with
> an ExecRestrPos: node type 18 not supported.  Haven't looked yet...

Yes, there's a nasty little planning bug in 7.0.*: the planner can
try to use merge or nestloop joins as the inner input of another
mergejoin.  That doesn't work, and what's worse, 7.0.* will just
silently give wrong answers for such a plan (well, there's a DEBUG
message in the postmaster log, but that's little help :-().  Current
sources give an executor error instead.  I have the planner bug fixed
in my own files but it's intertwined with a ton of OUTER JOIN changes
I'm not ready to commit yet...

            regards, tom lane

pgsql-general by date:

Previous
From: Tatsuo Ishii
Date:
Subject: Re: client_encoding
Next
From: Matt Beauregard
Date:
Subject: back-end dies on select