Re: [BUGS] documentation bug - behave of NEW a OLD in plpgsql's triggers

Josh Kupershmidt <schmiddy@gmail.com> writes:
> [Moving to -docs]
> On Mon, May 2, 2011 at 12:00 PM, Pavel Stehule <pavel.stehule@gmail.com> wrote:
>> It isn't correct. NEW is not declared in DELETE trigger, OLD isn't
>> declared in INSERT

That claim is flat out wrong.

> If I've understood you correctly, the problem is that the docs claim
> that the variables are defined with a value of NULL, when in fact they
> are undefined. For example, if you try to use variable NEW in a delete
> trigger, you'll get an error message like:
> |  ERROR:  record "new" is not assigned yet
> |  DETAIL:  The tuple structure of a not-yet-assigned record is indeterminate.

That is, in fact, exactly the behavior you get if you declare a RECORD
variable and set it to NULL.  If these variables were indeed not
declared, you'd get a complaint about "new" not being a known variable.
Observe:

regression=# create function foo(int) returns void as $$
regression$# begin
regression$#   new.x := $1;
regression$# end$$ language plpgsql;
ERROR:  "new.x" is not a known variable
LINE 3:   new.x := $1;
          ^

versus

regression=# create function foo(int) returns void as $$
regression$# declare new record;
regression$# begin
regression$#   new := null;
regression$#   new.x := $1;
regression$# end$$ language plpgsql;
CREATE FUNCTION
regression=# select foo(1);
ERROR:  record "new" is not assigned yet
DETAIL:  The tuple structure of a not-yet-assigned record is indeterminate.
CONTEXT:  PL/pgSQL function "foo" line 5 at assignment

            regards, tom lane