split column value of a table and skip some words - sql

Question

Hil All,
I have a table , count is about 200M. It has a column which contains data separated by '~'. I want to parse it.
e.g:
Column1
A~B~C~D~E~F
Result :
Column_new1
A~C~E
I just want to skip 2,4,6,n th. words. I don't want plsql. I need sql query. And table is very big,I also need performance.
I use substr,instr functions and I can parse. But it runs really slowly..
Thanks for help.

Answer 1

If you are after performance then use the INSTR and SUBSTR simple string functions:
SELECT SUBSTR(column1, 1, p1 - 1 ) || '~' ||
SUBSTR(column1, p2 + 1, p3 - p2 - 1) || '~' ||
SUBSTR(column1, p4 + 1, p5 - p4 - 1) AS column1_new
FROM (
SELECT column1,
INSTR(column1, '~', 1, 1) AS p1,
INSTR(column1, '~', 1, 2) AS p2,
INSTR(column1, '~', 1, 3) AS p3,
INSTR(column1, '~', 1, 4) AS p4,
INSTR(column1, '~', 1, 5) AS p5
FROM table_name
);
Which, for the sample data:
CREATE TABLE table_name (column1) AS
SELECT 'A~B~C~D~E~F' FROM DUAL;
Outputs:
COLUMN1_NEW
A~C~E
If you want a shorter query then you can use regular expressions:
SELECT REGEXP_REPLACE(column1, '([^~]+)~[^~]+~([^~]+)~[^~]+~([^~]+).*', '\1~\2~\3' )
AS column1_new
FROM table_name;
However, you will find that performance is likely to be an order of magnitude worse than simple string functions.
Another alternative would be to generate a materialized view.
db<>fiddle here

Answer 2

This is a regular expressions option. Looks nice, isn't PL/SQL, works OK (for 2 rows). I'm afraid that anything will run slow for 200 million rows.
SQL> with test (id, col) as
2 (select 1, 'A~B~C~D~E~F' from dual union all
3 select 2, 'M~N~O~P~Q-R' from dual
4 )
5 select id,
6 regexp_substr(col, '\w+', 1, 1) || '~' ||
7 regexp_substr(col, '\w+', 1, 3) || '~' ||
8 regexp_substr(col, '\w+', 1, 5) result
9 from test;
ID RESULT
---------- -----------------------------------
1 A~C~E
2 M~O~Q
SQL>