설문조사
PostgreSQL/PPAS 관련 듣고 싶은 교육은


총 게시물 118건, 최근 0 건
   

Converting Oracle function to PostgreSQL 3

글쓴이 : 모델광 날짜 : 2022-08-26 (금) 21:10 조회 : 98

Recently while watching a YouTube video, it has just occurred to me that I do something stupid when I do not know the technology I work with.


Here is the URL of the YouTube video.

(Warning! If you are under 19 years old, do not watch the video.)


https://www.youtube.com/watch?v=8rh6qqsmxNs


At the 40 second part of the video, the comedian said "Think about how stupid the average person is, and then realize that half of them are stupider than that."


Last year when I worked on a project where  I had to port Oracle functions to PostgreSQL functions, there was a user defined function which pushed me to do something stupid.


Let me kick things off by introducing an Oracle function code that I had to port.


CREATE OR REPLACE FUNCTION F_CONNECT_BY(

  P_GROUP_CD VARCHAR2 )

RETURNVARCHAR2

IS

  v_string   VARCHAR2(2000);

BEGIN

  SELECTSUBSTR(SYS_CONNECT_BY_PATH(CD_NM,','),2)

    INTOv_string

   FROM(

         SELECTCD_NM

                   , ROW_NUMBER() OVER (ORDER BY CD_NM) RN

                   , COUNT(*) OVER () AS CNT

           FROMCOM_CODE

         WHEREGROUP_CD = p_group_cd

            )

 WHERE RN = CNT                          --check condition

 START WITH RN = 1                        --access condition

 CONNECT BY RN = PRIOR RN+1;     --join   condition


  RETURNv_string;

END;


The following is the DDL for the table COM_CODE.


create table com_code (

group_cd  varchar(10),

cd        varchar(10),

cd_nm     varchar(100));

insert into com_code values ('G1','11001','SEOUL')

                           ,('G1','11002','PUSAN')

                           ,('G1','11003','INCHEON')

                           ,('G1','11004','DAEGU')

                           ,('G1','11005','JAEJU')

                           ,('G1','11006','ULEUNG')

                           ,('G1','11007','ETC');

insert into com_code values ('G2','1','Infant')

                           ,('G2','2','Child')

                           ,('G2','3','Adolescent')

                           ,('G2','4','Adult')

                           ,('G2','5','Senior');

insert into com_code values ('G3','01','Jeonbuk')

                           ,('G3','02','Kangwon')

                           ,('G3','03','Chungnam');

alter table com_code add constraint com_code_pk

  primary key (group_cd, cd);


Before reading this article on, think for a minute, "how would you write the above function code in PostgreSQL?"

PostgreSQL does not support the START WITH, CONNECT BY, PRIOR, and SYS_CONNECT_BY_PATH things. As soon as I saw the Oracle function code, I started to write a PostgreSQL function code using WITH RECURSIVE CLAUSE like the following:


WITH recursive W1 AS (
  SELECT group_cd, cd, cd_nm::varchar
       , row_number() over (order by cd_nm) rn
       , count(*) over () as cnt
    FROM COM_CODE
   WHERE group_cd = 'G3'
)
, W AS (
  SELECT group_cd, cd, cd_nm, rn, cnt
    FROM W1
   WHERE group_cd = 'G3'
 UNION ALL
  SELECT a.group_cd, a.cd, a.cd_nm||','||w.cd_nm::varchar as cd_nm, w.rn+1 as rn, w.cnt
    FROM w1 a join lateral (select *
                              from w
                             where a.group_cd = w.group_cd
                               and a.rn <= w.rn
                              offset 0) w
      ON true
     AND w.rn  <= a.cnt
)
SELECT *
FROM w
order by cd_nm;


Even though I spent about 2 hours trying to come up with an alternative PostgreSQL code to the Oracle code, I failed. I did a lot of googling and refered to the following posts:

https://stackoverflow.com/questions/35248217/multiple-cte-in-single-query

https://heap.io/blog/postgresqls-powerful-new-join-type-lateral 


Desperately when I ran the SQL statement in the Oracle function allocating the value 'G1' to the p_group_cd variable I got the following result:


DAEGU,ETC,INCHEON,JAEJU,PUSAN,SEOUL,ULEUNG


As you can see the result of the Oracle user defined function was the output of a built-in LIST_AGG() function. That was the mement when I realized that the writer of the F_CONNECT_BY() was an idiot and I was also an idiot. The Oracle function writer and I were trying to rewrtie the database built-in function. And PostgreSQL implements the same functionality with STRING_AGG().


Conclusion

Database companies have spent a lot of money developing database features. I am semi-serious. In PostgreSQL the stupidest PostgreSQL core team member is probably smarter than you. You need to know the technology you are working with. Do not try to rewrite the database feature on your own.


Footnote

In the video I introduced at the beginning of this note, the comedian went on to say "It does not take long to spot one of them (stupid people), does it? It takes you about eight seconds." But it took me about 2 hours to realize that I was an idiot. I must be one of those who are stupider than the average stupid person.


Addendum

If you are a God-demn idiot and love codes of high complexity, you might have ported the Oracle function to PostgreSQL like the following:


CREATE OR REPLACE FUNCTION F_CONNECT_BY(
  P_GROUP_CD VARCHAR)
RETURNS VARCHAR
LANGUAGE SQL STABLE PARALLEL SAFE STRICT
AS
$$
WITH recursive W1 AS (
  SELECT group_cd, cd, cd_nm, rn, lag(rn) over(order by rn) as rn2, count(rn) over() as cnt
    FROM (
          SELECT group_cd, cd, cd_nm
               , row_number() over (order by cd_nm desc) rn
            FROM COM_CODE
           WHERE group_cd = P_GROUP_CD
         ) AS FOO
  )
, W AS (
  SELECT group_cd, cd, cd_nm::varchar, rn, rn2, cnt
    FROM W1
   WHERE 1=1
     AND rn2  IS NULL
 UNION ALL
  SELECT a.group_cd, a.cd, a.cd_nm||','||w.cd_nm::varchar as cd_nm, a.rn, a.rn2 as rn2, a.cnt
    FROM w1 a join lateral (select *    -- I forced a nested loop join to keep the ordering.
                              from w
                             where w.group_cd = a.group_cd
                               and w.rn = a.rn2
                              offset 0) w
      ON true
)
SELECT cd_nm
  FROM w
 WHERE w.cnt = w.rn;
$$


   

postgresdba.com