Re: BUG #1277: plpgsql EXECUTE bug in beta3

"Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote in message
news:366(dot)1096909353(at)sss(dot)pgh(dot)pa(dot)us(dot)(dot)(dot)
> "PostgreSQL Bugs List" <pgsql-bugs(at)postgresql(dot)org> writes:
>> Under beta3, the following behaviour is observed:
>
>> test=# create or replace function execute_sql(text) returns void AS
>> $$begin
>> execute $1; return; end; $$ language plpgsql;
>> CREATE FUNCTION
>> test=#
>> test=# select execute_sql('create table a (i integer); insert into a(i)
>> values(1);');
>> ERROR: relation "a" does not exist
>> CONTEXT: SQL query "create table a (i integer); insert into a(i)
>> values(1);"
>> PL/pgSQL function "execute_sql" line 1 at execute statement
>
> This is happening because EXECUTE now parses and plans the whole string
> in one go, so that it tries to plan the INSERT before the CREATE has
> been carried out. You would see the same behavior if you tried for
> instance to execute those two commands as the body of an SQL function.
> I am inclined to regard this as "not a bug", and tell you to execute the
> two queries in separate EXECUTE commands. I'm not sure it's worth the
> substantial additional complexity in spi.c that would be needed to
> preserve the old behavior --- especially when the documentation does not
> suggest anywhere that you can use EXECUTE to execute more than one
> command in the first place.
>
> Anyone else have an opinion?
>
> regards, tom lane
>

I guessed it would be something like that - thanks for the explanation -
will update our code to perform each step individually.

--
Tom Hebbron
www.hebbron.com