Re: Allow to collect statistics on virtual generated columns

On Tue, 24 Mar 2026 11:43:47 +0000
Dean Rasheed <dean(dot)a(dot)rasheed(at)gmail(dot)com> wrote:

> On Tue, 24 Mar 2026 at 10:57, Yugo Nagata <nagata(at)sraoss(dot)co(dot)jp> wrote:
> >
> > One concern is that CREATE STATISTICS does not allow creating extended
> > statistics on a single colmun. If we try this, it raises the following error:
> >
> > ERROR: extended statistics require at least 2 columns
> >
> > Therefore, if we take this approach, we would need to add an exception
> > to this rule for virtual columns.
>
> I think that would be fine. It makes sense because a virtual column
> expands to an expression, and we already allow CREATE STATISTICS on a
> single expression.

I've attached a patch updated along these lines.

This allows users to define extended statistics on virtual generated
columns. Expressions including such columns are stored in the catalog
as-is, and expanded at ANALYZE time. Extended statistics on a single
virtual generated column are also allowed, since it is treated as a
single expression.

To enable the optimizer to make use of these statistics, expressions
are also expanded at planning time.

Additional documentation and tests are needed, but are not yet included.

Is this approach better then the previous one?

Regards,
Yugo Nagata

--
Yugo Nagata <nagata(at)sraoss(dot)co(dot)jp>