SQL challenge: avoid this self-join!!!

Question

Here's something of a challenging SQL problem. I'm trying to persist an arbitrary number of attributes for an object. I am trying to do this in a regular relational table both for performance and to make future upgrades easier.
The problem is that I don't know what SQL cleverness I can use to only scan the ATTR table once.
Does Oracle (or for that matter the SQL standard) have some way to help me? Here's a simplified example:
Consider a table ATTR with columns OID, ATTR_ID, ATTR_VAL. Unique key is OID, ATTR_ID. Assume any other indexes that you want, but be aware that ATTR_VAL is modestly dynamic.
I can easily look for a OID for any one ATTR_ID, ATTR_VAL pair:
SELECT oid FROM attr
WHERE attr_id = 1 AND attr_val = :b1
I can also easily do this looking at multiple attributes when I only need one condition to be met with an OR, as:
SELECT DISTINCT oid FROM attr
WHERE (attr_id = 1 AND attr_val = :b1)
OR (attr_id = 31 AND attr_val = :b2)
But how to handle the condition where I want to have the two ATTR_ID, ATTR_VAL pairs "and-ed" together? I know that I can do this:
SELECT oid FROM
(SELECT oid FROM attr WHERE attr_id = 1 AND attr_val = :b1)
UNION
(SELECT oid FROM attr WHERE attr_id = 31 AND attr_val = :b2)
But this will necessitate looking at ATTR twice. This is maybe okay if there are only two conditions, but what about when there might be 10 or even 50? At some point this technique becomes unacceptable.
Clearly:
SELECT DISTINCT oid FROM attr
WHERE (attr_id = 1 AND attr_val = :b1)
AND (attr_id = 31 AND attr_val = :b2)
won't work (each row has but one ATTR_ID).
The following will end up doing the same basic thing as the UNION (it avoids a sort so is preferable):
SELECT oid FROM attr a1, attr a2
WHERE a1.oid = a2.oid
AND (a1.attr_id = 1 AND a1.attr_val = :b1)
AND (a2.attr_id = 31 AND a2.attr_val = :b2)
but the fundamental problem of scanning ATTR twice remains.
What cleverness can I apply here to only scan ATTR once?
Thanks,
:-Phil

Answer 1

An other way of having a dynamic in list build from a singel string is show at asktom at this link http://asktom.oracle.com/pls/ask/f?p=4950:8:2019864::NO::F4950_P8_DISPLAYID,F4950_P8_CRITERIA:210612357425,%7Bvarying%7D%20and%20%7Belements%7D%20and%20%7Bin%7D%20and%20%7Bin%7D%20and%20%7Blist%7D
an modified version for two columns:
Create or replace type in_list as object (col1 varchar2(20), col2 varchar2(30));
Create or replace type in_list_tab as table of in_list;
Create or replace function fn_in_list( p_string in varchar2) return in_list_tab
as
l_string long default p_string || ',';
l_data in_list_tab := in_list_tab();
pos number;
begin
pos := 0;
loop
exit when l_string is null;
pos := instr( l_string, ',' );
l_data.extend;
l_data(l_data.count) := in_list('','');
l_data(l_data.count).col1 := ltrim(rtrim(substr(l_string, 1, pos - 1)));
l_string := substr( l_string, pos + 1 );
if l_string is null
then
     l_data.Trim;
     exit;
end if;
pos := instr( l_string, ',' );
l_data(l_data.count).col2 := ltrim(rtrim(substr(l_string, 1, pos - 1)));
l_string := substr( l_string, pos + 1 );
end loop;
return l_data;
end;
create table testII (cola varchar2(10), colb varchar2(30));
insert into testII values ('abc',1);
insert into testII values ('abc',2);
insert into testII values ('def',1);
insert into testII values ('def',2);
commit;
var b1 varchar2(200);
exec :b1:='abc,1,def,2';
select * from testII where (cola,colb) in
(select col1, col2 from THE ( select cast(fn_in_list(:b1) as in_list_tab) from dual));
to handle cases like
attr_id = 41 and attr_val > :b3 I would say dynmaic SQL