Received: from sss.sss.pgh.pa.us (sss.pgh.pa.us [209.114.166.2]) by hub.org (8.9.3/8.9.3) with ESMTP id KAA38163 for ; Mon, 6 Dec 1999 10:15:08 -0500 (EST) (envelope-from tgl@sss.pgh.pa.us) Received: from sss.sss.pgh.pa.us (localhost [127.0.0.1]) by sss.sss.pgh.pa.us (8.9.1/8.9.1) with ESMTP id KAA18400; Mon, 6 Dec 1999 10:14:02 -0500 (EST) To: mjnf@uevora.pt cc: pgsql-sql@postgreSQL.org Subject: Re: [SQL] Wierd stuff In-reply-to: Your message of Mon, 06 Dec 1999 13:49:38 +0000 <384BBEF2.B6CB6CDD@neptuno.uevora.pt> Date: Mon, 06 Dec 1999 10:14:02 -0500 Message-ID: <18398.944493242@sss.pgh.pa.us> From: Tom Lane Mario Jorge Nunes Filipe writes: > Tom Lane wrote: >> Actually, I think that this may indeed be the root of the problem. >> I killed several bugs in cross-data-type joins a while back ... for >> example, IIRC pre-6.5 Postgres thought it could use hash joins for >> cross-data-type equality operators, but that doesn't work properly in >> general. Mario, what does EXPLAIN show about how the join query is >> being done? > To awnser your question... EXPLAIN shows this: > sac=> explain select c.nome from aluno a, concelho c where > sac-> a.concelho_naturalidade=c.codigo and a.numero_aluno=13792; > NOTICE: QUERY PLAN: > Hash Join (cost=27.44 size=164 width=18) > -> Seq Scan on c (cost=13.16 size=308 width=16) > -> Hash (cost=0.00 size=0 width=0) > -> Index Scan on a (cost=2.05 size=2 width=2) Yup, that's what I suspected. Hashing 2-byte and 4-byte integers doesn't deliver the same result --- at least not all the time, there are special cases such as zero. So this explains all your observed misbehaviors. You can work around it for now by remaking the tables with consistent data types. regards, tom lane