Re: Oracle 8i (Version 8.1.5) bug

This was discussed a few days ago on the newsgroup.

This feature seems to like the following environment :- RBO and a sort merge (I think - still to check this) with index scan underneath it in the execution plan.

Possible ways to step-around the problem (which may have side-effects):-

Analyze the tables (ideally in all schemas including sys and system - maybe use DBMS_UTILITY.ANALYZE_SCHEMA stored procs.), use the CBO and also use the /*+ all_rows */ hint in SQL (if you can actually get at the SQL).

CD-Rom documentation has instructions how to set the server to use CBO.

Note that you can't touch recursive (internal to the server).

I make no comment on a) the feature or b) the quality of the workaround or c) which version one should use in the meantime. Moreover, no inferences should be drawn from any part of this post vis a vi any of these issues, or any other related issues.

David P. OCP (DBA) MCP (TCP/IP)

Glasgow, Scotland.

Martin Haltmayer wrote:
>
> I also would be very interested in this bug and your scripts.
>
> Martin
>
> ShenHN wrote:
> >
> > I like to report an extremely serious and dangerous bug in
> > Oracle8i (Version 8.1.5). This bug has been reported in TAR
> > 1093520.999 and has been verified by Oracle support.
> >
> > The error, simply stated, is that two equivalent SQL SELECT
> > queries can return different results. I have submitted a test
> > case consisting of four SQL scripts: (1) create schema of 5
> > tables, (2) populate tables (app. 20 rows of data) and the two
> > SQL SELECT statements that return correct and incorrect result
> > sets. Oracle support has run the test case against both 32 bit &
> > 64 bit Oracle8i version 8.1.5 for both the Solaris and HP
> > platforms. In all these cases, the results are consistent, i.e.,
> > the two SQL SELECT statements return different answers. For
> > Johnson Controls Inc., a leader in the building automation
> > industry, the consequence of such a bug can be disastrous. Oracle
> > has found the source of the bug and has issued a patch that at
> > least works against the test case I submitted on the HP 64 bit
> > platform.
> >
> > I have suggested that Oracle issue an advisory, warning Oracle
> > customers of the potential problem. Oracle has steadfastly
> > refused to issue such warning, claiming the condition under which
> > this error can occur is rare. IMHO, I believe it's more common
> > than Oracle thinks.
> >
> > I can post the 4 SQL scripts if there is enough interest in the
> > newsgroup.
> >
> > Hwa N. Shen
> > Johnson Controls, Inc.
> >
> > * Sent from RemarQ http://www.remarq.com The Internet's Discussion Network *
> > The fastest and easiest way to search and participate in Usenet - Free!
  Received on Tue May 23 2000 - 00:00:00 CDT