Re: [GENERAL] different results based solely on existence of index (no, seriously)

On Mon, Aug 11, 2008 at 10:35:26PM -0500, Matthew Dennis wrote:
> In reference to the script below (I know it can be rewritten, that's
> not the point), I get 3 rows if the referenced index exists but only
> two rows if it does not. This is observable and repeatable just by
> dropping/creating the index. Drop the index and two rows are
> returned. Create the index, three rows are returned. Drop the
> index, two rows again. In addition, in no case does the selected
> column t2.c2 actually contain a value (it's always null). Since in
> the 3 row case, it returns a row with t1.c1=2, I would have expected
> a value from t2 (if you add t2.c1 to select clause you can see that
> is null as well).
>
> It's probably worth mentioning (since it actually took me a while to
> notice) that the plans are subtlety different. Neither plan (with
> or without index existing) actually uses the index, but in one case
> there is an extra filter node.
>
> version string is PostgreSQL 8.3.1 on i686-redhat-linux-gnu,
> compiled by GCC gcc (GCC) 4.1.2 20070925 (Red Hat 4.1.2-33)

I have reproduced it on 8.3.3.

Just FYI, a bug isn't a bug unless you can reproduce it on the latest
minor version, in this case 8.3.3, of the major version, in this case
8.3, that the bug appears in.

Cheers,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778 AIM: dfetter666 Yahoo!: dfetter
Skype: davidfetter XMPP: david(dot)fetter(at)gmail(dot)com

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate