Re: [GENERAL] Returning a RECORD, not SETOF RECORD

On Fri, Apr 29, 2005 at 10:36:05AM -0400, Tom Lane wrote:
>
> regression=# select (xyz(unique1,unique2)).* from tenk1 limit 5;

This is a little off topic, but I've noticed that the above invokes
the function once per output column:

CREATE FUNCTION xyz(INOUT x integer, INOUT y integer, OUT z integer) AS $$
BEGIN
RAISE INFO 'calling xyz';
z := x + y;
END;
$$ LANGUAGE plpgsql IMMUTABLE;

SELECT xyz(1,2);
INFO: calling xyz
xyz
---------
(1,2,3)
(1 row)

SELECT (xyz(1,2)).*;
INFO: calling xyz
INFO: calling xyz
INFO: calling xyz
x | y | z
---+---+---
1 | 2 | 3
(1 row)

Is that because the splat causes the query to be expanded into
"SELECT (xyz(1,2)).x, (xyz(1,2)).y, (xyz(1,2)).z"? Is it possible
or desirable to optimize that into a single call, at least if the
function were stable or immutable?

--
Michael Fuhr
http://www.fuhr.org/~mfuhr/