On 2011-12-13 18:34, Tom Lane wrote:
> Heikki Linnakangas<heikki.linnakangas@enterprisedb.com> writes:
>> Attached is a patch with those changes. I also I removed a few of the
>> syntax error regression tests, that seemed excessive, plus some general
>> naming and comment fiddling. I'll apply this tomorrow, if it still looks
>> good to me after sleeping on it.
> However, I'm still concerned about whether this approach gives
> reasonable error messages in cases where the error would be
> detected during parse analysis of the rearranged statement.
> The regression test examples don't cover such cases, and I'm
> too busy right now to apply the patch and check for myself.
> What happens for example if a named parameter's value contains
> a misspelled variable reference, or a type conflict?
I tested this and seems to be ok:
regression=# select namedparmcursor_test1(20000, 20000) as "Should be
false", namedparmcursor_test1(20, 20) as "Should be true";
ERROR: column "yy" does not exist
LINE 1: SELECT x AS param1, yy AS param12;
regression=# select namedparmcursor_test1(20000, 20000) as "Should be
false", namedparmcursor_test1(20, 20) as "Should be true";
ERROR: invalid input syntax for integer: "2011-11-29 19:26:10.029084"
CONTEXT: PL/pgSQL function "namedparmcursor_test1" line 8 at OPEN
regards,
Yeb Havinga
last error was created with
create or replace function namedparmcursor_test1(int, int) returns
boolean as $$
declare c1 cursor (param1 int, param12 int) for select * from rc_test where
a > param1 and b > param12; y int := 10; x timestamp := now(); nonsense record;
begin open c1(param12 := $1, param1 := x);
end
$$ language plpgsql;
