db-derby-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Knut Anders Hatlen (JIRA)" <j...@apache.org>
Subject [jira] Commented: (DERBY-3634) Cannot use row_number() in ORDER BY clause
Date Thu, 08 Oct 2009 08:53:31 GMT

    [ https://issues.apache.org/jira/browse/DERBY-3634?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12763419#action_12763419
] 

Knut Anders Hatlen commented on DERBY-3634:
-------------------------------------------

Another thing I noticed, was a change in behaviour. Given this sequence of statements

create table t (x int);
insert into t values 1,2,3;
select * from (select row_number() over (), x from t) s(r,x) where x > 2;

the old implementation would return (3,3) whereas the new implementation returns (1,3), probably
because the restriction (x>2) is now pushed down. This is not a bug, as the order of the
rows in the inner select is undefined, but it might have some consequences that need to be
handled when ORDER BY in sub-selects is implemented (DERBY-4397).

Similarly, some may perhaps find it unintuitive that this query now returns 1 for all parameter
values in (1,2,3), whereas it previously returned different values for different parameter
values:

    select r from (select row_number() over (), x from t) s(r,x) where x = ?

But again, it's not a bug.

> Cannot use row_number() in ORDER BY clause
> ------------------------------------------
>
>                 Key: DERBY-3634
>                 URL: https://issues.apache.org/jira/browse/DERBY-3634
>             Project: Derby
>          Issue Type: Bug
>          Components: SQL
>    Affects Versions: 10.4.1.3
>            Reporter: Rick Hillegas
>            Assignee: Dag H. Wanvik
>         Attachments: derby-3634-a.diff, derby-3634-a.stat, derby-3634-a.txt, derby-3634-b.diff,
derby-3634-b.stat, derby-3634-c.diff, derby-3634-c.stat, derby-3634-newimpl-1.diff, derby-3634-newimpl-1.stat,
derby-3634-newimpl-1.txt, derby-3634-newimpl-2.diff, derby-3634-newimpl-2.stat, derby-3634-remove-2.diff,
derby-3634-remove-2.stat, derby-3634-remove.diff, derby-3634-remove.stat
>
>
> The following query works correctly:
> select abs(a), row_number() over ()
> from t
> where a > 100 and a < 111
> order by abs(a)
> I expected the following query to also work, but it raised an exception:
> select abs(a), row_number() over ()
> from t
> where a > 100 and a < 111
> order by row_number() over ()
> This is the error I saw: "ERROR 42X01: Syntax error: Encountered "over" at line 5, column
23".
> Here are the reasons why I think that this syntax is supposed to be supported:
> According to my reading of the 2003 SQL spec, the ORDER BY clause should be able to sort
on any expression in the SELECT list. That includes OLAP expressions. I believe this is so
because, according to part 2, section 10.10 (<sort specification>), a <sort key>
can be any <value expression> and if you follow the grammar for <value expression>,
it can resolve to be a <value expression primary> (see section 6.3), which can in turn
resolve to be a <window function>. This reasoning is supported by tracing the hotlinks
on the following page which lays out the SQL 2003 BNF: http://savage.net.au/SQL/sql-2003-2.bnf.html
This interpretation is further supported by the example of an ORDER BY clause referencing
an OLAP expression which is provided on page 23 of the introduction to OLAP written by Fred
Zemke, Krishna Kulkarni, Andy Witkowski, and Bob Lyle: www.cse.iitb.ac.in/dbms/Data/Papers-Other/SQL1999/OLAP-99-154r2.pdf

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message