(Sorry for my last message which got away from me when I hit ^X instead
of ^C in PC-Pine.)
Gene,
You were right. I guess I didn't try letting postgres pick the opclass
after all, because when I did, the index was still used.
I'll take out the opclass from all my index creates.
Thanks for all your help.
Regards,
../fam
--
Frank A. Mandarino
fam@risca.com
---------- Forwarded message ----------
Date: Thu, 7 Oct 1999 07:53:10 -0400
From: Frank Mandarino <fam@dbsys.risca.com>
To: "Gene Selkov, Jr." <selkovjr@mcs.anl.gov>
Cc: pgsql-general@postgreSQL.org
Subject: Re: [GENERAL] btree index on a char(8) field (fwd)
Gene,
Thanks for your response.
I knew from the programming documentation that the opclass was optional.
I'm pretty sure, although I will check again tonight, that I tried
creating the index without specifying the opclass, but I found that the
index was still not used in my example query.
Do you know which opclass that Postgres should choose for char(8) types?
Regards,
../fam
--
Frank A. Mandarino
fam@risca.com
On Thu, 7 Oct 1999, Gene Selkov, Jr. wrote:
>
> My comment:
>
> The deficiency of the docs in regards to operator classes probably
> results from the fact that no one is asking about those. The opclass
> parameter in CREATE INDEX is no longer required (Herouth has been
> around long enough to recall the times when it was).
>
> As you have just witnessed, in a standard situation, you are better
> off without knowing about it -- postgres will pick the right opclass
> for you. That will not happen, however, when the values you want to
> index are of a custom type, or when a built-in type does not have an
> opclass of its own (as is the case with the point type). Also, you
> need this option to override the default opclass for those types that
> can work with multiple opclasses (which is what you attempted to
> achieve).
>
> Will anyone with a solid knowledge of the type system want to augment
> the existing docs?
>
> --Gene
>
> ************
>