Can I rewrite this as semi join ?

Question

Hi,
I hate queries like this:
select distinct t1.col1 from t1 , t2
where
t1.pk = t2.fk
I'm always trying to rewrite that as semi-join like
select t1.col1 from t1
where
t1.pk in (select t2.fk from t2);
but now I've got something like this:
SELECT DISTINCT prd.prd_id, prd.*
           FROM prod1 prd,
                prod1_parametry_kd prn,
                prod1_param_k_dys pkd
          WHERE prn.prn_prd_id = prd.prd_id
            AND prn.prn_data_od <= SYSDATE
            AND (prn.prn_data_do IS NULL OR prn.prn_data_do >= SYSDATE)
            AND prn.prn_pkd_id = pkd.pkd_id
            AND pkd.pkd_psl_lpr_id = :1
            AND prd.prd_dostepnosc_od <= :2
            AND (prd.prd_dostepnosc_do IS NULL OR prd.prd_dostepnosc_do >= TRUNC (:3)
I'm little confused because there is no connection between pkd and prd.
Is that query still rewritable as semi-join ?Regards.
Greg
Edited by: user10388717 on May 14, 2010 6:47 AM
Edited by: user10388717 on May 14, 2010 6:59 AM

Answer 1

Hi, Greg,
user10388717 wrote:
Hi,
I hate queries like this:
select distinct t1.col1 from t1 , t2
where
t1.pk = t2.fk
I'm always trying to rewrite that as semi-join like
select t1.col1 from t1
where
t1.pk in (select t2.fk from t2);
but now I've got something like this:
SELECT DISTINCT prd.prd_id, prd.*
FROM prod1 prd,
prod1_parametry_kd pkd,
prod1_param_k_dys pkd
WHERE prn.prn_prd_id = prd.prd_id
AND prn.prn_data_od <= SYSDATE
AND (prn.prn_data_do IS NULL OR prn.prn_data_do >= SYSDATE)
AND prn.prn_pkd_id = pkd.pkd_id
AND pkd.pkd_psl_lpr_id = :1
AND prd.prd_dostepnosc_od <= :2
AND (prd.prd_dostepnosc_do IS NULL OR prd.prd_dostepnosc_do >= TRUNC (:3)
I'm little confused because there is no connection between pkd and prd.
Is that query still rewritable as semi-join ?Regards.
Greg
Edited by: user10388717 on May 14, 2010 6:47 AMThere are two tables with the same alias, pkd. Should one of them be prn?
Anyway, sure, you can do what you want. The IN sub-query can be a join of two other tables, or perhaps there could be a nested IN sub-query.
Perhaps something like this:
SELECT DISTINCT prd.prd_id, prd.*
           FROM prod1 prd,
          WHERE prd_id     IN (  SELECT  prn.prn_prd_id
                       FROM    prod1_parametry_kd     prn,
                                      prod1_param_k_dys      pkd
                            WHERE   prn.prn_data_od         <= SYSDATE
                            AND     (   prn.prn_data_do   IS NULL
                                OR  prn.prn_data_do   >= SYSDATE
                            AND     prn.prn_pkd_id         = pkd.pkd_id
                            AND     pkd.pkd_psl_lpr_id    = :1
            AND prd.prd_dostepnosc_od <= :2
            AND (prd.prd_dostepnosc_do IS NULL OR prd.prd_dostepnosc_do >= TRUNC (:3)
;Of course, I can't test it.
Conditions that involve prd are in the main WHERE clause.
Conditions that involve the other tables, but not prd, are in the sub-query.
I suspect that doing a join would be at least as efficient. Why don't you want to do a join?
Edited by: Frank Kulash on May 14, 2010 10:08 AM