Re: Alias collision in `refresh materialized view concurrently`

On Fri, Aug 06, 2021 at 10:48:40AM -0400, Tom Lane wrote:
> AFAICT that works and generates the identical parse tree to the original
> coding. The only place touched by the patch where it's a bit difficult to
> make the syntax unambiguous this way is
>
> "CREATE TEMP TABLE %s AS "
> "SELECT _$mv.ctid AS tid, _$newdata "
> "FROM %s _$mv FULL JOIN %s _$newdata ON (",
>
> because newdata.* would ordinarily get expanded to multiple columns
> if it's at the top level of a SELECT list, and that's not what we want.
> However, that's easily fixed using the same hack as in ruleutils.c's
> get_variable: add a no-op cast to the table's rowtype. So this
> would look like
>
> appendStringInfo(&querybuf,
> "CREATE TEMP TABLE %s AS "
> "SELECT mv.ctid AS tid, newdata.*::%s "
> "FROM %s mv FULL JOIN %s newdata ON (",
> diffname, matviewname, matviewname, tempname);

Smart piece. I haven't thought of that.

> Given that it took this long to notice the problem at all, maybe
> this is not a fix to cram in on the weekend before the release wrap.
> But I don't see why we need to settle for "mostly works" when
> "always works" is barely any harder.

Yes, I would vote to delay that for a couple of days. That's not
worth taking a risk for.
--
Michael