Re: EXISTS clauses not being optimized in the face of 'one time pass' optimizable expressions

On Fri, Jul 1, 2016 at 10:11 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>> On Fri, Jul 1, 2016 at 9:52 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> Maybe, but neither UNION nor UNION ALL would duplicate the semantics
>>> of OR, so there's some handwaving here that I missed.
>
>> SELECT * FROM foo WHERE a = 5 OR a = 4
>> isn't equivalent to
>> SELECT * FROM foo WHERE a = 5
>> UNION
>> SELECT * FROM foo WHERE a = 4
>> ?
>
> It probably is, but you're assuming that "a" appears in the list of
> columns being unioned. If you make that just "SELECT b FROM ..."
> then the latter form gets rid of duplicate b values where the first
> doesn't. On the other hand, UNION ALL might introduce duplicates
> not present in the OR query's result.

Right, so, significant query transformations are non-trivial. But the
point is that with the upper planification stuff, I think it is
possible, at least in some cases, that we could consider reordering
set operations with scan/join planning, just as we've previously
talked about reordering grouping stages relative to scan/join
planning.

The details are undeniably hard to get right.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company