Comparing Substring Against Multiple Columns - sql

Question

I have a table which has 20 similar text attribute columns, text1..text20. These columns are of type CLOB. I am looking to find rows where one of these text attribute columns contain a specific phrase, such as '%unemployed%'. I need to know 2 things, which rows match, and which column was matched on. I thought I could use ANY as a starting point, but I am having issues.
It appears the ANY statement does NOT work with '%'. For example,
select * from emp where 'BLAKE' = ANY(ename, job); -- Returns Data
but
select * from emp where '%BLAKE%' = ANY(ename, job) -- No Data Found
What would be the proper way to do this? Pseudo-code would be...
Select name, addr,
which_column_matched(%unemployed%, text1..text20),
text1..text20
from table
where %unemployed% = ANY(text1..text20);

Answer 1

In Oracle, you can use unpivot for this. It still requires you to enumerate all the columns, but the syntax is quite neat.
If you want one record for each column that matches:
select *
from emp unpivot (col for src in (text1, text2, text3))
where col like '%unemployed%'
If you wanted one additional column with the list of matching columns instead, you can aggregate the resultset:
select ename, listagg(src, ', ')
from emp unpivot (col for src in (text1, text2, text3))
where col like '%unemployed%'
group by ename

Answer 2

You can use a subquery to identify the first column that matches and then return that:
select t.*
from (select t.*,
(case when text1 like '%unemployed%' then 'text1'
when text2 like '%unemployed%' then 'text2'
. . .
when text20 like '%unemployed%' then 'text20'
end) as col_match
from t
) t
where col_match is not null;

Answer 3

I'm always worried about how Oracle treats CLOB data, so here's a test that shows that the Pivot solution should do the trick.
drop table emptest;
-- Assuming we are using the venerable EMP table
create table emptest as select * from emp;
alter table emptest add(
text1 CLOB,
text2 CLOB,
text3 CLOB
)
/
declare
v_text clob;
begin
-- set one column to a length well beyond 16k but below 32k, max VARCHAR2 for PL/SQL
v_text := lpad('X', 16000, 'X')||' unemployed ' || lpad('X', 10000, 'X');
update emptest set text2 = v_text where ename = 'SMITH';
-- set others to short values
v_text := 'an unemployed salesman in text 1';
update emptest set text1 = v_text where ename = 'TURNER';
v_text := 'an unemployed manager in text 3';
update emptest set text3 = v_text where ename = 'JONES';
commit;
end;
/
declare
v_clob clob;
begin
-- Set a field to an absurdly long value, with the match value way beyond 32k.
update emptest set text1 = empty_clob() where ename = 'SMITH' returning text1 into v_clob;
for i in 1..10000 loop
dbms_lob.writeappend(v_clob, 36, 'ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890');
end loop;
dbms_lob.writeappend(v_clob, 18, 'unemployed manager');
commit;
end;
/
select empno, ename, clob_name, clob_value, length(clob_value) clob_length
from emptest unpivot (clob_value for clob_name in (text1, text2, text3))
where clob_value like '%unemployed%'
/
The result of this will be:
EMPNO ENAME CLOB_NAME CLOB_VALUE CLOB_LENGTH
----- ------- --------- ----------- -----------
7566 JONES TEXT3 <excluded> 31
7369 SMITH TEXT1 <excluded> 360018
7369 SMITH TEXT2 <excluded> 26012
7844 TURNER TEXT1 <excluded> 32
Of key importance is how Oracle handles the LIKE keyword when dealing with TEXT1 for SMITH: notice that the column has a length of >360k characters.
Much of the standard syntax we try using with CLOB datatypes works only because Oracle coerces the CLOB into a VARCHAR2, but that has the inherent length limitations.
As this test shows, the LIKE comparison does work for fat CLOB values--at least in Oracle 12c, where I tested it.
Things will be somewhat different if you try to display the actual content that matches: you will need to familiarize yourself with the DBMS_LOB package and its subprograms such as DBMS_LOB.INSTR and DBMS_LOB.SUBSTR if you are dealing with long CLOB values.