[SQL] plpgsql: PERFORM vs. SELECT INTO (PERFORM not setting FOUND variable?)

[Marinos Yannikos]

Hi,

I'm probably misintepreting the documented equivalence of PERFORM and
SELECT INTO as far as the special variable FOUND is concerned, but the
following 2 definitions do not seem to produce the same result.

create table blup ( t1 text, t2 text );

create function blup_unique2 (text,text) returns boolean as 'begin
perform (select 1 from blup where t1=$1 or t1=$2 or t2=$1 or t2=$2 or
$1=$2 limit 1); return NOT FOUND; end' LANGUAGE plpgsql;

create function blup_unique3 (text,text) returns boolean as 'declare x
record; begin select into x 1 from blup where t1=$1 or t1=$2 or t2=$1 or
t2=$2 or $1=$2 limit 1; return NOT FOUND; end' LANGUAGE plpgsql;

The first will always produce false, i.e. apparently the subquery used
by PERFORM will not set the variable FOUND as expected. Is this correct?

(btw.: I'm trying a few ways to ensure that all values in both t1 and t2
are unique:
alter table blup add constraint check (blup_unique3(t1,t2));
- perhaps there are more elegant ways, any suggestions?)

Regards,
Marinos

---------------------------(end of broadcast)---------------------------
TIP 6: Have you searched our list archives?

http://archives.postgresql.org

[Tom Lane]

Marinos Yannikos <m...@geizhals.at> writes:
> create function blup_unique2 (text,text) returns boolean as 'begin
> perform (select 1 from blup where t1=$1 or t1=$2 or t2=$1 or t2=$2 or
> $1=$2 limit 1); return NOT FOUND; end' LANGUAGE plpgsql;

You've got a syntax problem. PERFORM is syntactically like SELECT,
so what you wrote is equivalent to
SELECT (SELECT 1 FROM blup ....)
In other words, you are evaluating a scalar subquery, which is going to
return either "1" or "NULL" depending on whether the WHERE matches,
or give an error if the WHERE matches multiple rows (a case you wouldn't
hit because of the LIMIT). So the outer SELECT produces exactly one row
containing the scalar result, and FOUND always ends up TRUE.

So what you want is just

PERFORM 1 FROM blup ...

and then check the FOUND result from that.

(The 8.0 docs hopefully explain this more clearly; PERFORM was
certainly not very well documented before.)

Note that I'm concerned that the performance of this will suck ...
in particular you really ought to test the $1=$2 case separately.

regards, tom lane