On Mon, Jan 6, 2025 at 3:45 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> SELECT ... FROM (VALUES (...) ORDER BY a) v(a,b,c)
>
> If it'd been like that all along, nobody would blink at it I think,
> even though you could argue that it's action-at-a-distance to let
> an outer alias affect what happens inside the implicit subquery.
To be honest, that pushdown feels really uncomfortable to me. To me,
the natural syntax for associating an alias with a VALUES clause would
be something like "VALUES (...) myalias" or, if you also wanted column
aliasing, "VALUES (...) myalias(a,b,c)". That would feel just like
what we allow when aliasing a table or a function reference. But to
not allow that syntax but then allow the outer alias to propagate
inward in this one case seems quite strange to me. It doesn't really
seem evil in a vacuum, but it seems non-orthogonal with what we do
everywhere else, because aliases in SQL generally flow outward, not
inward.
--
Robert Haas
EDB: http://www.enterprisedb.com
