Postgres Pro
Downloads
Home   >   mailing lists

Thread: Re: Lag and lead window functions order by weirdness

Re: Lag and lead window functions order by weirdness

From
Thom Brown
Date:
2009/10/10 Thom Brown <thombrown@gmail.com>
I've had a look at examples of lag and lead window functions with order by inside the OVER clause, and I'm confused as to why it influences the overall order in the output.

For example (for a table called category with incrementing id numbers up to 26):

SELECT id, lag(id) OVER (ORDER BY id DESC) FROM category;

This would yield:

id   lag
26   NULL
25   26
24   25

This is the equivalent of what we would get with:
SELECT id, lead(id) OVER (ORDER BY id ASC) FROM category ORDER BY id DESC;

I would expect the row order in the results not to be influenced by the OVER clause's ORDER BY as I thought that was just to determine how the aggregate window function's values would be output.  It's almost as if the main part of the query has inherited it's order from the OVER clause.

Am I missing something here?

Thanks

Thom

Having a look around, it looks as if Postgres might be misbehaving.  According to this page, http://my.safaribooksonline.com/0596004818/sqlnut2-CHP-4-SECT-3, the ORDER BY in the window function's clause shouldn't be having this ordering effect:

"Furthermore, the order within these groups is defined by an ordering clause, but that order only affects function evaluation, and has no effect on the order in which rows are returned by the query."

The behaviour is unexpected from my perspective, but obviously there are workarounds.  Is anyone able to confirm any of this?

Thom

Re: Lag and lead window functions order by weirdness

From
Tim Landscheidt
Date:
Thom Brown <thombrown@gmail.com> wrote:

> [...]
> Having a look around, it looks as if Postgres might be misbehaving.
>  According to this page,
> http://my.safaribooksonline.com/0596004818/sqlnut2-CHP-4-SECT-3, the ORDER
> BY in the window function's clause shouldn't be having this ordering effect:

> "Furthermore, the order within these groups is defined by an ordering
> clause, but that order only affects function evaluation, and has no effect
> on the order in which rows are returned by the query."

> The behaviour is unexpected from my perspective, but obviously there are
> workarounds.  Is anyone able to confirm any of this?

AFAIR, others have already pointed out that without an "OR-
DER BY" clause PostgreSQL can return the result set in *any*
"order" it deems fit. So why don't you use one?

Tim