Re: plpgsql - DECLARE - cannot to use %TYPE or %ROWTYPE for composite types

2016-03-17 0:39 GMT+01:00 Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>:

> Jim Nasby <Jim(dot)Nasby(at)BlueTreble(dot)com> writes:
> > On 3/3/16 4:51 AM, Pavel Stehule wrote:
> >> CREATE TABLE a(a int);
> >> CREATE TABLE b(a a.a%TYPE)
> >>
> >> And the people expecting the living relation between table a and table
> >> b. So when I do ALTER a.a, then b.a should be changed. What if I drop
> >> a.a or drop a?
> >>
> >> So this is reason, why I don't would this feature in SQL side.
>
> > I don't buy that. plpgsql doesn't work that way, so why would this?
> > *especially* with the %TYPE decorator.
>
> Yeah. The %TYPE decorator doesn't work like that in the core parser
> either: when you use it, the referenced type is determined immediately
> and then it's just as if you'd written that type name to begin with.
> I do not see a reason for any of these "type operators" to work
> differently.
>
> Another analogy that might help make the point is
>
> set search_path = a;
> create table myschema.tab(f1 mytype);
> set search_path = b;
>
> If there are types "mytype" in both schemas a and b, is myschema.tab.f1
> now of type b.mytype? No. The meaning of the type reference is
> determined when the command executes, and then you're done.
>

This is valid for PostgreSQL. I am not sure if it is true in Oracle, if
%TYPE means only reference to type, or %TYPE holds reference to original
object - and when you change the original object, then the function is
invalidated.

Using %TYPE with create time only semantic has not big practical benefit.
But when %TYPE enforce all life dependency, then I have guaranteed so
change on original object will be propagated to depend object. With all
advantages and disadvantages.

Postgres uses %TYPE in create time only semantic - but it is not big issue
in PLpgSQL, because the creation time there is often - every first
execution in session.

The usage of %TYPE outer PL/pgSQL is probably only in FK. But nothing
similar is in standard, and I don't see a reason, why we should to
implement it. In this moment I don't see any important use case.

Pavel

>
> regards, tom lane
>