Re[2]: to_number question

From: Jonathan Gennick <jonathan_at_gennick.com>
Date: Fri, 16 Jul 2004 09:31:46 -0400
Message-ID: <401070086843.20040716093146@gennick.com>

Well put Stephen. I agree with your post.

SLDC> The comments were an expression of disbelief (not lack
SLDC> of understanding) that the specifications concerning
SLDC> subqueries would be so loose and open ended as to
SLDC> allow this level of unpredictability.

Yes. And there's a lot more to this issue than just the specific Oracle behavior we've been talking about. I have many questions, most of which I'll just have to research off-list. For example:

• What would relational theory have to say about this subquery behavior? What would Codd say? What would Date say? Do any of their writings bear directly on subquery execution?
• What about the SQL language designers? What did they have in mind when they spec'd out subqueries? What's their conceptual vision for how subqueries execute? Did they write anything into the standard in detail that bears on the behavior we've been discussing?
• It would seem that the mental model of FROM-clause subquery results becoming the source of data for the main query is flawed. What then, is the correct mental model to use when thinking about and writing such subqueries? (this last is a *very* important question)
• Clearly we have a case where an optimization changes the behavior of a statement. Without optimization the statement succeeds. With optimization the statement fails. Even though I understand what is happening, this troubles me. It raises an interesting question: should an optimization be allowed to change the results? Why/why not?
• Related to the above, it would now seem that in order to write a statement containing a subquery, that you must be aware of, and think about, how that statement's execution is optimized. Yet databases are supposed to free us from worrying about optimization and physical access and the like.
• What about other database vendors? Is this an Oracle-only issue? How does this scenario play out under MySQL? under DB2? under SQL Server? Is there consensus among vendors? Or do those who design optimizers take different, fundamental approaches to the issue we've been discussing? (I'd really like to talk to the developers behind the various database optimizers about this issue)
• Oh, and this *is* important, I haven't had time yet to put together a test case that reproduces Stephen's error. It's on my to-do list though. My one attempt so far failed to reproduce the error. If I can't put together a test case, then I have nothing to research.

Now that I see what Oracle is doing in this one case, I want to go deeper, find out about the intent (of theorists, language designers, optimizer writers), come up with a mental model that encompasses the behavior we've seen, and so forth. There's a lot of drilling down to be done here, and probably a good article to be written afterward.

Best regards,

Jonathan Gennick --- Brighten the corner where you are http://Gennick.com * 906.387.1698 * mailto:jonathan@gennick.com

Join the Oracle-article list and receive one article on Oracle technologies per month by email. To join, visit http://five.pairlist.net/mailman/listinfo/oracle-article, or send email to Oracle-article-request_at_gennick.com and include the word "subscribe" in either the subject or body.

Friday, July 16, 2004, 3:59:11 AM, Stephen.Lee_at_DTAG.Com (Stephen.Lee_at_DTAG.Com) wrote:

>> -----Original Message-----
>> Actually the subquery gets converted to a sql that has two predicates
>> grouped by "AND" (similar to yours).

SLDC> Well OK.  I've kept my mouth shut so far, and because I was getting some

SLDC> good info, I did not argue with the accusations that I "don't understand". SLDC> But I think it's time to clarify some things.

SLDC> I DO understand 100%, and (if I may be so bold as to speak for others) those SLDC> who have questioned the so-called logic understood 100%.

SLDC> Yes.  I'm very well aware that my SQL was essentially getting broken into
SLDC> two predicates and THEN the un-guaranteed order in the evaluation of
SLDC> predicates gets applied.  The people who argued that this should not be the
SLDC> case understood it too.  The comments were an expression of disbelief (not
SLDC> lack of understanding) that the specifications concerning subqueries would
SLDC> be so loose and open ended as to allow this level of unpredictability.
SLDC> Clearly, that is the case.  But that doesn't mean we can't bitch about it.
SLDC> And bitching about it doesn't mean we don't understand it.  You bitch about
SLDC> taxes, don't you?

SLDC> ----------------------------------------------------------------
SLDC> Please see the official ORACLE-L FAQ: http://www.orafaq.com
SLDC> ----------------------------------------------------------------
SLDC> To unsubscribe send email to:  oracle-l-request_at_freelists.org
SLDC> put 'unsubscribe' in the subject line.
SLDC> --
SLDC> Archives are at http://www.freelists.org/archives/oracle-l/

SLDC> FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html

SLDC> -----------------------------------------------------------------

----------------------------------------------------------------

Please see the official ORACLE-L FAQ: http://www.orafaq.com

---

To unsubscribe send email to: oracle-l-request_at_freelists.org put 'unsubscribe' in the subject line.

--
Archives are at http://www.freelists.org/archives/oracle-l/
FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------

Received on Fri Jul 16 2004 - 08:30:28 CDT