Re: SQL question - why doesn't WHERE clause know about renamed fields?

[Charles Wilt <charles.wilt@xxxxxxxxx>]

No, the use of the CTE in your example wouldn't change the fact that
an index can't be used.

In fact, I'm pretty sure just moving the WHERE criteria out of the CTE
could cause an index to not be used.  It depends of if the SQE can
optimize out the CTE or not.  In other words:

while this will use an index:
select * from ech
where hedte < datetonum(CURRENT_DATE - 30 days)


I wouldn't be surprise to see that this wouldn't (after all you told
it to build a temporary table, which isn't going to have an index)
with temp1 as
  (select * from ech )
select * from temp1
where hedte < datetonum(CURRENT_DATE - 30 days)

As to your second question, numtodate() certainly prevents the use of
an index and the + 30 days may do so.  (It does at v5r4 at least,
perhaps later versions are a bit smarter and can rewrite +30 on one
side to - 30 on the other)

Running some actual tests at v5r4
select * from hlthprdfil/finvhdp
where invdat = current_date
--->uses an index probe (fastest)

select * from hlthprdfil/finvhdp
where invdat = current_date - 7 days
--->uses an index probe (fastest)

select * from hlthprdfil/finvhdp
where invdat + 7 days = current_date
--->uses an index scan (slow, but not as slow as a table scan)


Rule of thumb, try not to do anything to the data in a column in the
WHERE clause.

HTH,
Charles



On Mon, Nov 8, 2010 at 2:47 PM, Tom E Stieger <testieger@xxxxxxxxxxxx> wrote:
> Charles-
>
> If you did
> With temp1 as
>        (select numtodate(hord)+30 days as testdate
>        from ...
>        )
> Select * from temp1 where testdate < current_date
>
> Would that make any difference to the index selection?  I'm pretty sure it wouldn't but I thought I would ask.  Also is it the numtodate UDF or the "+ 30 days" on the field that cause the full table scan, or both?
>
> Thanks
>
> -Tom Stieger
> Engineer
> California Fine Wire