Re: 7.1 gets stricter on SQL -- MIDRANGE-L

Joe,

I wonder if this could be of help to you (from the i7.1 Memo to Users):

**********
SQL Query changes in collation of errors might result in a SQL0802
A change was made to not collate errors for SQL queries with ordering,
distinct, or aggregation. This was
done to avoid providing unpredictable or wrong results when a result in
error cannot be included in
ordering, aggregation, or distinct result sets. A selection error is now
signaled in this scenario, using the
SQL0802 message with SQLCODE=-802 and SQLSTATE=’22003’. The SQL0802 message
will be preceded
by CPF5035 and CPD4019 escape messages.
*To allow compatibility with previous behavior, a new QAQQINI option,
COLLATE_ERRORS, is available.*
*COLLATE_ERRORS - Specifies how data errors are handled on the GROUP BY and
ORDER BY expression*
*during hash or sort processing within queries.*
Supported values:
Chapter 2. IBM i operating system 15
v *DEFAULT - The default is *NO.
v *NO - A value of *NO causes the query to be ended with an error when a
grouping or ordering
expression results in an error.
v *YES - A value of *YES indicates that the grouping or sort continues.
This support was also available by PTF in V5R4 and V6R1.
**********

HTH,

Luis Rodriguez
IBM Certified Systems Expert — eServer i5 iSeries
--

On Mon, Jan 31, 2011 at 2:26 PM, Joe Pluta <joepluta@xxxxxxxxxxxxxxxxx>wrote:

WARNING: This entire post can be classified in the "Doctor, it hurts
when I do this", "So don't DO that!" category, but even so, it might
save someone some serious pain in the future.

Somewhere between V5R4 and V7.1 DB2 got much more picky about the data
in your files. This especially affects queries against multi-format
physical data - that is, legacy data files defined with DDS where one or
more records DON'T match the physical format of the file. Examples are
files with control records (you all remember those?) or records where
they have order data and comments in the same file, with the comments
having a special record ID.

Anyway, the gist of the problem is this: in V5R4 you could work with
those files just fine provided your select clause omitted the improperly
formatted record(s). So, if I had an ID field for my ORDDET file and
the ID was '01' for order detail lines and '02' for comments, I could
easily do something like this:

SELECT DISTINCT ODCUST FROM ORDDET WHERE ID = '01' ORDER BY ODCUST

This would give me a nice, succinct ordered list of customer numbers in
the ORDDET file. However, as of V7.1, that will fail with a "selection
error" for the first type '02' record, telling me that the data in
ODCUST was invalid.

Note that the WHERE ID = '01' clause omits the offending record. That's
of no matter to DB2; it fails anyway. The data is bad, and it's not
happy. Now, if I take the ORDER BY clause off, it works fine. So I
tried this, thinking I could force the select ahead of time:

WITH T1 AS (SELECT ODCUST FROM ORDDET WHERE ID = '01')
SELECT DISTINCT ODCUST FROM T1 ORDER BY ODCUST

Uh uh. No way, dude. I still get the same problem. Interesting point
is that it doesn't happen on every case I tried. I've whittled it down
to this: if the field already has an access path (logical file) over it,
then DB2 is fine and the statement runs. My guess is that this is
because it uses the existing access path. If there is none, then the
SQL engine needs to build one, and it builds it regardless of the
selection criteria, and so it includes the bad record and promptly falls
over.

If I create a logical by the field in question, the statements work. If
I delete the logical, they fail again.

So there ya go.

Joe
--
This is the Midrange Systems Technical Discussion (MIDRANGE-L) mailing list
To post a message email: MIDRANGE-L@xxxxxxxxxxxx
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/mailman/listinfo/midrange-l
or email: MIDRANGE-L-request@xxxxxxxxxxxx
Before posting, please take a moment to review the archives
at http://archive.midrange.com/midrange-l.