ZF-5295: Failure on Oracle database

Issue Type:
Bug
Created:
2008-12-17T02:02:52.000+0000
Last Updated:
2009-06-18T19:41:13.000+0000
Status:
Resolved
Fix version(s):
  • 1.8.4 (23/Jun/09)
Reporter:
Bogdan Albei (bogdan.albei)
Assignee:
Jurrien Stutterheim (norm2782)
Tags:
  • Zend_Paginator
Related issues:
Attachments:

Description

The DbSelect adapter fails under Oracle, because Oracle does not support statements like:



Instead, in Oracle, the statement should be:

I made a quick fix by changing Zend/Paginator/Adapter/DbSelect.php, but then this would only support Oracle(which is no problem in my case) The fix is not perfect, there is a chance of failure, but I suppose the '###' string could be changed to something unique.


187,188d186
<           //ORACLE COUNT(DISTINCT col1||col2..) fix
<           $groupPart = str_replace(",", "||'###'||", $groupPart);

Sample code:


$select = new Zend_Db_Select($this->db);

$select->from("product_transfers", 
                array(
                    "quantity" => "SUM(quantity)",
                    "product_id",
                    "name",
                    "batch_number"));

$select->group(array(
                "product_id",
                "name",
                "batch_number"));

$paginator = Zend_Paginator::factory($select);

Comments

Posted by Mickael Perraud (mikaelkael) on 2008-12-17T02:07:22.000+0000

Please, could you provide a complete reproduce code?

Posted by Mickael Perraud (mikaelkael) on 2008-12-21T09:39:18.000+0000

I understood the problem but your fix could have problem:


create table foo(
    bar number(3) not null,
    baz varchar2(50)
);
insert into foo( 1, '0abc');
insert into foo( 10, 'abc');
select count(distinct bar || baz) from foo;
// this will return 1 because '||' will concat the 2 columns and won't apply a real distinct

Posted by Mickael Perraud (mikaelkael) on 2008-12-23T01:06:39.000+0000

If we want something independent of the DBMS, we must use a request like:


SELECT COUNT(*) FROM (SELECT DISTINCT column1, column2, column3 FROM table_name)

Note: Problem also occurs with DB2

Posted by Jurrien Stutterheim (norm2782) on 2008-12-23T01:14:25.000+0000

I'd prefer to avoid subqueries if possible because they're a lot slower than regular queries. Wouldn't a GROUP BY on the multiple columns provide the same result?



would become

Feel free to slap me if I'm saying stupid things atm... I need more coffee!

Posted by Mickael Perraud (mikaelkael) on 2008-12-23T01:25:02.000+0000

No, this wouldn't function ;) I f you take my example above, you will receive 2 lines of result. I understand though the problem of performance. Take a coffee :D

Posted by Jurrien Stutterheim (norm2782) on 2008-12-23T11:08:41.000+0000

Hehe, coffee taken ; ) We could probably detect these problematic queries and create subqueries if there are no other nice ways of supporting them. The solution would preferably work on every DB. If we go for subqueries, we should make an entry in the manual saying these kind of queries are better off with a custom COUNT query (also supported by the DbSelect adapter).

Posted by Jurrien Stutterheim (norm2782) on 2009-05-30T20:44:25.000+0000

This thread: http://dbasupport.com/forums/showthread.php/… might have a solution. The Paginator Adapter solution would check the no. of columns. If there's one column, we keep doing things the way they're done now. If there are multiple columns, we apply the suggestion from that forum thread. Currently haven't got time to fully investigate this, so feedback is welcome : )

Posted by Matthew Ratzloff (mratzloff) on 2009-06-08T14:23:35.000+0000

This isn't a blocking issue, so I'm lowering priority to critical. An alternative exists: passing a query to setRowCount().

Posted by Jurrien Stutterheim (norm2782) on 2009-06-18T19:41:01.000+0000

Resolved in r16149 and merged to release-1.8 in r16150