Thread: Re: [GENERAL] pg_attribute.attnum - wrong column ordinal?

Greg Stark escribió:
> On Wed, Nov 25, 2009 at 1:03 AM, Konstantin Izmailov <pgfizm@gmail.com> wrote:
>
> > My question: can pg_attribute.attnum be used to determine the sequential
> > ordinal positions of columns in a table? What is a right way to get the
> > ordinal numbers?
>
> You could use something like:
>
> row_number() over (partition by T.schemaname,T.viewname order by
> attnum) as "ORDINAL_POSITION"

Should we recast the attributes and columns views in information_schema?
I notice they still use attnum.

--
Alvaro Herrera                                http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.

Alvaro Herrera <alvherre@commandprompt.com> writes:
> Should we recast the attributes and columns views in information_schema?
> I notice they still use attnum.

I'd vote against it, at least until we have something better than a
row_number solution.  That would create another huge performance penalty
on views that are already ungodly slow.

When and if we get around to separating physical from logical column
position, the issue might go away "for free".

            regards, tom lane

On tor, 2009-12-03 at 10:09 -0500, Tom Lane wrote:
> Alvaro Herrera <alvherre@commandprompt.com> writes:
> > Should we recast the attributes and columns views in information_schema?
> > I notice they still use attnum.
>
> I'd vote against it, at least until we have something better than a
> row_number solution.  That would create another huge performance penalty
> on views that are already ungodly slow.

Should be easy to test the performance impact of this, since the limit
for columns per table is 1600.