Does 'select 1 from mytable where 1 = 0' do a table scan?

[Joe Weinstein]

Hi. Some DBMSes are clever enough not to go to data pages if a
knowably constant search criterion is false. Is DB2 among them?

thanks,
Joe Weinstein at BEA

[Troels Arvin]

On Mon, 15 Nov 2004 08:50:46 -0800, Joe Weinstein wrote:

Hi. Some DBMSes are clever enough not to go to data pages if a
knowably constant search criterion is false. Is DB2 among them?

DB2 is smart enough to see that (1=0) and (4>5) will never be true, and
will do neither a table scan nor an index scan for such a simple query.

I don't know the limits of smartness, though (i.e. when predicated become
too tricky to decide on).

--
Greetings from Troels Arvin, Copenhagen, Denmark

[Philip Sherman]

The clever approach to answering this question is to do an "explain" on
the query, using a simple table (ie. the sample tables that come with
UDB) and examining the results.

If your predicate stated "where 1 = 1" then an output row would be
generated for every row in the table. A scan would be needed to generate
the correct number of output rows. If the developers of a retrieval
engine are foolish enough to look for this type of code and write
special logic to save the user from his/her own bad coding, then they
deserve the problems it can easily cause. Handling predicates in a
consistant manner, without special case code, is the way to long term
stability, consistancy, and overall performance in the retrieval engine.

Philip Sherman
Joe Weinstein wrote:

Hi. Some DBMSes are clever enough not to go to data pages if a
knowably constant search criterion is false. Is DB2 among them?

thanks,
Joe Weinstein at BEA

[Joe Weinstein]

Troels Arvin wrote:

On Mon, 15 Nov 2004 08:50:46 -0800, Joe Weinstein wrote:

Hi. Some DBMSes are clever enough not to go to data pages if a
knowably constant search criterion is false. Is DB2 among them?

DB2 is smart enough to see that (1=0) and (4>5) will never be true, and
will do neither a table scan nor an index scan for such a simple query.

I don't know the limits of smartness, though (i.e. when predicated become
too tricky to decide on).

Thanks!

[Joe Weinstein]

Philip Sherman wrote:

The clever approach to answering this question is to do an "explain" on
the query, using a simple table (ie. the sample tables that come with
UDB) and examining the results.
Thanks!
As long as the assumption that a DB2 instance is available to the asker is valid.
If your predicate stated "where 1 = 1" then an output row would be
generated for every row in the table. A scan would be needed to generate
the correct number of output rows.
That would be assumed. I'm just interested in the obvious 1 = 0 case.
I hope my assumption, that you are circumlocuting the answer,
"DB2 will not go to an index or to data for such a query" is correct.

If the developers of a retrieval engine are foolish enough to look for this type of code and write
special logic to save the user from his/her own bad coding, then they
deserve the problems it can easily cause. Handling predicates in a
consistant manner, without special case code, is the way to long term
stability, consistancy, and overall performance in the retrieval engine.
Thanks again. I'm not sure who you're referring to, but I've seen some
applications do that (adding a where 1 = 0 clause) even onto queries they
didn't generate themselves, in order to get the metadata about the query,
which will be sent to the client, even for a zero-row return. Not me, but
I get paid to deal with other's problems, even self-inflicted...
Philip Sherman
Joe Weinstein wrote:

Hi. Some DBMSes are clever enough not to go to data pages if a
knowably constant search criterion is false. Is DB2 among them?

thanks,
Joe Weinstein at BEA

[Serge Rielau]

Joe Weinstein wrote:

Hi. Some DBMSes are clever enough not to go to data pages if a
knowably constant search criterion is false. Is DB2 among them?

thanks,
Joe Weinstein at BEA

Yes.

[Troels Arvin]

On Mon, 15 Nov 2004 15:58:53 -0500, Serge Rielau wrote:

Hi. Some DBMSes are clever enough not to go to data pages if a
knowably constant search criterion is false. Is DB2 among them?

[...] Yes.

Is there any good documentation on what kinds of semantic query
optimizations like this DB2 tries to perform?

--
Greetings from Troels Arvin, Copenhagen, Denmark

[Serge Rielau]

Troels Arvin wrote:

On Mon, 15 Nov 2004 15:58:53 -0500, Serge Rielau wrote:

Hi. Some DBMSes are clever enough not to go to data pages if a
knowably constant search criterion is false. Is DB2 among them?

[...]

Yes.

Is there any good documentation on what kinds of semantic query
optimizations like this DB2 tries to perform?

Not that I'm aware of. DB2 UDB for LUW knows some 100 semantic query
rewrite rules. Some of which have patented algorithms which you can look
up, but in general this is all part of the secret mix ;-)
Here is a link to the "theorem prover" (note that it's not fully
exploited in rewrite to keep the compiler snappy :) It was originally
added in DB2 V5.2 for typed view hierarchy optimization.
http://patft.uspto.gov/netacgi/nph-P...&RS=PN/6728952

Cheers
Serge