Re: Allow virtual columns in index expressions or predicate

On Sep 30, 2025, at 17:13, Richard Guo <guofenglinux@gmail.com> wrote:

I'm looking for a solution to support virtual generated columns in
index expressions or predicate.  The comment in DefineIndex() seems to
suggest that virtual generated columns there can be expanded in
RelationGetIndexExpressions() and RelationGetIndexPredicate().

  /*
   * XXX Virtual generated columns in index expressions or predicates
   * could be supported, but it needs support in
   * RelationGetIndexExpressions() and RelationGetIndexPredicate().
   */

I'm not sure this is the appropriate approach, especially since we
don't have the heap relation available in these functions.  I think
we're good as long as we expand the virtual columns before the index
build process; otherwise, evaluating these expressions during the
build will cause executor errors due to unexpanded virtual columns.

I wonder if we can do the expansion in DefineIndex(), before the call
to index_create(), like the attached.

- Richard
<v1-0001-Allow-virtual-columns-in-index-expressions-or-pre.patch>

Hi Richard. 

This is a good usability improvement, making it easier to create an index when a virtual column is in an expression or a WHERE clause.

It might also be nice if this could be extended to allow creating an index directly on the virtual column, like CREATE INDEX ON tbl (virtual_col).

If we view this feature as a "syntactic sugar" to simplify creating an expression index, then I think this patch is very easy to accept.

However, if we consider more complex situations, like a user running ALTER TABLE ... SET EXPRESSION, it raises more questions. Whether we need to add the kind of full support for these cases that was discussed in [1] needs a more detailed discussion. But for the simpler goal of being a syntax improvement, this patch works well.

Thanks,

Chengpeng Yan

[1] https://www.postgresql.org/message-id/flat/CACJufxGao-cypdNhifHAdt8jHfK6-HX%3DtRBovBkgRuxw063GaA%40mail.gmail.com