"Joshua D. Drake" <jd@commandprompt.com> writes:
> > That's basically how the existing patch approached the problem. It invents a
> > new type of join and a new type of tuplestore that behaves this way. This has
> > the advantage of working the way Oracle users expect and being relatively
> > simple conceptually. It has the disadvantage of locking us into what's
> > basically a nested loop join and not reusing existing join code so it's quite
> > a large patch.
>
> I believe our Syntax should be whatever the standard dictates,
> regardless of Oracle.
Well the issue here isn't one of syntax. The syntax is really an orthogonal
issue. The basic question is whether to treat this as a new type of plan node
with its behaviour hard coded or whether to try to reuse existing join types
executing them recursively on their output. I can see advantages either way.
As far as the syntax goes, now that I've actually read up on both, I have to
say: I'm not entirely sure I'm happy IBM won this battle. The Oracle syntax is
simple easy to use. The IBM/ANSI syntax is, well, baroque. There's a certain
logical beauty to it but I can't see users being happy trying to figure out
how to use it.
-- Gregory Stark EnterpriseDB http://www.enterprisedb.com