Horton

Thread

Commits

  1. Add COSTS off to two EXPLAIN using tests.

  2. Change unknown-type literals to type text in SELECT and RETURNING lists.

  1. explain with costs in subselect.sql

    Andres Freund <andres@anarazel.de> — 2018-03-12T22:20:23Z

    Hi Tom, all,

When forcing JITing to be enabled for all queries, obviously only useful
for testing, I noticed that two explain outputs changed after I added
explain support.

The only differences come from:

-- Unspecified-type literals in output columns should resolve as text

SELECT *, pg_typeof(f1) FROM
  (SELECT 'foo' AS f1 FROM generate_series(1,3)) ss ORDER BY 1;

-- ... unless there's context to suggest differently

explain verbose select '42' union all select '43';
explain verbose select '42' union all select 43;

which don't use costs=off.  Is there a reason for that? I assume it was
just a harmless oversight?

Has been added in
http://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=1e7c4bb0049732ece651d993d03bb6772e5d281a

Greetings,

Andres Freund

  2. Re: explain with costs in subselect.sql

    Tom Lane <tgl@sss.pgh.pa.us> — 2018-03-12T22:25:42Z

    Andres Freund <andres@anarazel.de> writes:
> When forcing JITing to be enabled for all queries, obviously only useful
> for testing, I noticed that two explain outputs changed after I added
> explain support.

> The only differences come from:

> -- Unspecified-type literals in output columns should resolve as text

> SELECT *, pg_typeof(f1) FROM
>   (SELECT 'foo' AS f1 FROM generate_series(1,3)) ss ORDER BY 1;

> -- ... unless there's context to suggest differently

> explain verbose select '42' union all select '43';
> explain verbose select '42' union all select 43;

> which don't use costs=off.  Is there a reason for that? I assume it was
> just a harmless oversight?

Duh, yeah, explain (verbose, costs off) would do fine there.
Mea culpa.  Do you want to fix it?

			regards, tom lane

  3. Re: explain with costs in subselect.sql

    Andres Freund <andres@anarazel.de> — 2018-03-12T22:28:07Z

    On 2018-03-12 18:25:42 -0400, Tom Lane wrote:
> Andres Freund <andres@anarazel.de> writes:
> > When forcing JITing to be enabled for all queries, obviously only useful
> > for testing, I noticed that two explain outputs changed after I added
> > explain support.
> 
> > The only differences come from:
> 
> > -- Unspecified-type literals in output columns should resolve as text
> 
> > SELECT *, pg_typeof(f1) FROM
> >   (SELECT 'foo' AS f1 FROM generate_series(1,3)) ss ORDER BY 1;
> 
> > -- ... unless there's context to suggest differently
> 
> > explain verbose select '42' union all select '43';
> > explain verbose select '42' union all select 43;
> 
> > which don't use costs=off.  Is there a reason for that? I assume it was
> > just a harmless oversight?
> 
> Duh, yeah, explain (verbose, costs off) would do fine there.
> Mea culpa.  Do you want to fix it?

Yup, will do.

Random aside: Noticed that we have *no* proper coverage of AND / OR
behaviour of returning NULL if no element returns true. Will push
something to add coverage, I've now manually retested that twice for
JITing, which seems stupid.

Greetings,

Andres Freund