On Wed, Sep 7, 2011 at 03:40:19PM +0200, Pavel Stehule wrote:
> 2011/9/7 Josh Kupershmidt <schmiddy@gmail.com>:
> > On Tue, Sep 6, 2011 at 10:54 PM, Bruce Momjian <bruce@momjian.us> wrote:
> >> Josh Kupershmidt wrote:
> >>> How about a doc tweak like the attached?
> >>
> >> Perfect. Applied to 9.0, 9.1, and head. Thanks. Sorry for the delay.
> >
> > Err, as Tom's first comment in this thread explains, Pavel and I were
> > both wrong: the variables in question are indeed NULL, not undefined.
> > I think the docs were fine the way they were.
>
> There is maybe bug - these variables are defined, but they has not
> assigned tupledesc, so there is not possible do any test
>
> postgres=# create table omega (a int, b int);
> CREATE TABLE
> postgres=# create or replace function foo_trig()
> postgres-# returns trigger as $$
> postgres$# begin
> postgres$# raise notice '%', new;
> postgres$# return null;
> postgres$# end;
> postgres$# $$ language plpgsql;
> CREATE FUNCTION
> postgres=# create trigger xxx after delete on omega for each row
> execute procedure foo_trig();
> CREATE TRIGGER
> postgres=# insert into omega values(20);
> INSERT 0 1
> postgres=# delete from omega;
> ERROR: record "new" is not assigned yet
> DETAIL: The tuple structure of a not-yet-assigned record is indeterminate.
> CONTEXT: PL/pgSQL function "foo_trig" line 3 at RAISE
>
> so current text in documentation is not correct too.
I used your queries to test NEW/OLD on DELETE/INSERT, respectively, and
for statement-level triggers, and you are right that they are
unassigned, not NULL.
The attached patch fixes our documentation for PG 9.3.
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ It's impossible for everything to be true. +