Re: Data Model Advice

On Wed, Sep 15, 2010 at 12:04 PM, Gary Chambers <gwchamb@gmail.com> wrote:
> All,
>
> My data modeling and SQL are oftentimes woefully inadequate, and I am
> seeking advice on how to implement a particular database design.  My
> database (so far) is comprised of the following types of tables:
>
> The parts table contains parts of a particular type primary key is an
> alphanumeric part number, some with dashes.
>
>                       Table "public.parts"
>   Column   |         Type          |            Modifiers
> ------------+-----------------------+---------------------------------
>  partno     | character varying(64) | not null
>  partno_raw | character varying(64) | default NULL::character varying
>  boxno      | integer               | not null
>  slotno     | integer               | not null
> Indexes:
>    "pk_parts" PRIMARY KEY, btree (partno)
>
> The parts_subs table contains part numbers different manufacturers
> that are identical to something that already exists in the parts table
>
>              Table "public.parts_subs"
>   Column    |         Type          |     Modifiers
> -------------+-----------------------+--------------------
>  partno      | character varying(64) | not null
>  partsub     | character varying(64) | not null
>  partsub_raw | character varying(64) | not null
>  boxno       | integer               | not null
>  slotno      | integer               | not null
> Indexes:
>    "pk_parts_subs" PRIMARY KEY, btree (partno, partsub)
>
> I'm trying to determine, and I'm seeking advice on:
>
>  How to maintain the original part number format but create the
> primary key with only alphanumeric.  I currently have a 'before insert
> or update' trigger to strip the non-alphanumeric characters, but it
> doesn't permit the insertion of a record because it is null on insert
> (for some reason).  The trigger function, which I'm trying to make as
> flexible as possible, is:
>
> CREATE OR REPLACE FUNCTION fixup_partnumbers() RETURNS TRIGGER AS
> $fixup_partnumbers$
> BEGIN
>    IF (STRPOS(TG_TABLE_NAME, '_subs') > 0) THEN
>        NEW.partsub := REGEXP_REPLACE(NEW.partsub_raw, E'(\\W|_)', '', 'g');
>    ELSE
>        NEW.partno := REGEXP_REPLACE(NEW.partno_raw, E'(\\W|_)', '', 'g');
>    END IF;
>    RETURN NEW;
> END;
> $fixup_partnumbers$ LANGUAGE plpgsql;
>
> Thank you in advance for any advice and assistance you can provide.

hm.  let me make a general rule of thumb suggestion:  generic
behaviors, generic trigger function, specific behaviors, specific
trigger function.  setting a field that is only on a particular table
is specific and I'd prefer to make two trigger functions, which both
call something like:

create or replace function fixup_partno(text) returns text as
$$
  SELECT REGEXP_REPLACE(NEW.partno_raw, E'(\\W|_)', '', 'g');
$$ language sql immutable;

You should in theory be able to write a before trigger to intercept
and modify any field before it hits the table, including the primary
key.  The following works for me:
create table foo(t text primary key);

create or replace function fixup_foo() returns trigger as
$$
begin
  new.t := random()::text;
  return new;
end;
$$ language plpgsql;

create trigger on_foo_insert before insert on foo for each row
  execute procedure fixup_foo();

insert into foo values (null);
insert into foo values (null);
insert into foo values (null);