Re: possible DOMAIN implementation

=?iso-8859-2?Q?Hor=E1k_Daniel?= <horak(at)sit(dot)plzen-city(dot)cz> writes:
> When a new domain is created it will:
> - put a record into pg_type with typnam = domain name, new code for
> typtype = 'd' and typrelid = oid of a new record in pg_class (next line)
> - put a record into pg_class to create a fictional table with a new
> relkind ('d'?), relnatts = 1, relname can be system generated
> (pg_d_<domainname>)

Ugh. Don't overload pg_class with things that are not tables. I see no
reason that either pg_class or pg_attribute should be involved in the
definition of a domain. Make new system tables if you need to, but
don't confuse the semantics of critical tables.

> - put a records into pg_attribute and pg_attrdef with "column
> (attribute) definition" - real type, size, default value etc., owner
> will the fictional table from the previous step

> Then it will be required to modify functions that works with types. When
> typtype of a retrieved type is 'd' then it will perform lookups into
> pg_class, pg_attribute and pg_attrdef to find the real definition of the
> domain. These additional lookups will also create a performace penalty
> of using domains.

Why shouldn't this info be directly available from the pg_type row?

regards, tom lane