Using case for indexes in oracle - oracle

Question

I am having the below index defined on one of the tables in Oracle.
CASE WHEN C IS NOT NULL THEN A||','||B||','||C END
Initially, I thought the combination of A,B,C separated by commas should be unique when C is not null, but seems like I'm wrong. I did a bit of a research on this as well but could not find a good explanation.
Could someone kindly help me with this?
Thanks in Advance!!

Answer 1

Yes, it can be unique.
But don't forget - uniqueness is an additional cost.
There will be little benefit from such an index, since there will be duplicates of rows for fields A & B when C is NULL.
I suppose it might be worth changing the index function as :
a||','||b||case when c is not null then ','||c end
This will control duplicates if only columns A and B are used.
Of course, unless your table has other unique indexes using columns A and B.
For demonstration - a short example, with a unique functional index.
N.B. Please, when describing the problem /task, provide a description of the relevant objects (tables,indexes, procedures...etc.).)
For a test,try:
--ddl
drop table test_tab;
create table test_tab
( a varchar2(3) not null,
b varchar2(3) not null,
c varchar2(3),
idx varchar2(12) generated always as (case when c is not null then a||','||b||','||c end) virtual
);
--
create unique index test_tab_i1 on test_tab(idx);
--dml
insert into test_tab(a,b,c) values('a1','b1','c1');
insert into test_tab(a,b,c) values('a1','b2','c2');
insert into test_tab(a,b,c) values('a1','b1',null);
commit;
select * from test_tab;
ANALYZE TABLE test_tab COMPUTE STATISTICS FOR ALL INDEXES;
--get data of idx
select INDEX_NAME,NUM_ROWS,DISTINCT_KEYS from user_indexes i where i.table_name=upper('test_tab');
--dubt
insert into test_tab(a,b,c) values('a1','b1',null);
commit;
--
ANALYZE TABLE test_tab COMPUTE STATISTICS FOR ALL INDEXES;
--pass2
select * from test_tab;
--get data of idx
select INDEX_NAME,NUM_ROWS,DISTINCT_KEYS from user_indexes i where i.table_name=upper('test_tab');
--pass3
--dubt
insert into test_tab(a,b,c) values('a1','b1','c1');
Result:
Table created.
Index created.
1 row(s) inserted.
1 row(s) inserted.
1 row(s) inserted.
Statement processed.
Result Set 31
A B C IDX
a1 b1 c1 a1,b1,c1
a1 b2 c2 a1,b2,c2
a1 b1 - -
Download CSV
3 rows selected.
Statement processed.
Result Set 32
INDEX_NAME NUM_ROWS DISTINCT_KEYS
TEST_TAB_I1 2 2
Download CSV
1 row(s) inserted.
Statement processed.
Statement processed.
Result Set 33
A B C IDX
a1 b1 c1 a1,b1,c1
a1 b2 c2 a1,b2,c2
a1 b1 - -
a1 b1 - -
Download CSV
4 rows selected.
Result Set 34
INDEX_NAME NUM_ROWS DISTINCT_KEYS
TEST_TAB_I1 2 2
Download CSV
ORA-00001: unique constraint (SQL_XKMHFTSRXNSKMGVEDOYSVXZGL.TEST_TAB_I1) violated ORA-06512: at "SYS.DBMS_SQL", line 1721

Answer 2

Even if your question is vague, it is interesting. I will try to provide you a possible explanation.
A||','||B is unique if you can guarantee that for 2 distinct values of (A,B), the function A||','||B is not returning the same value.
You need to guarantee that TO_CHAR(A) ||','|| TO_CHAR(B) is unique.
I am considering that A and B are NOT NULL.
A||','||B is unique if A and B are VARCHAR but it is not the case if you have DATE or TIMESTAMP datatypes. This is not TRUE because TO_CHAR function has default parameters that depend on nls_parameters from your session.
Testcase - executing the same queries but changing the nls_date_format and try to create a unique index
drop table test;
create table test (a DATE, b VARCHAR2(10), C VARCHAR2(10));
alter session set nls_date_format= 'dd/mm/yyyy hh24:mi:ss';
drop index test;
truncate table test;
insert into test(a,b,c) values(sysdate+(1/60),'1','c1');
insert into test(a,b,c) values(sysdate+(1/160),'1','c1');
insert into test(a,b,c) values(sysdate+(1/80),'2','c2');
insert into test(a,b,c) values(sysdate+(1/180),'2','c2');
commit;
create unique index test on test (case when c is not null then a||','||b||','||c end);
select (case when c is not null then a||','||b||','||c end) valfn from test order by B;
alter session set nls_date_format= 'dd/mm/yyyy';
truncate table test;
drop index test;
insert into test(a,b,c) values(sysdate+(1/60),'1','c1');
insert into test(a,b,c) values(sysdate+(1/160),'1','c1');
insert into test(a,b,c) values(sysdate+(1/80),'2','c2');
insert into test(a,b,c) values(sysdate+(1/180),'2','c2');
create unique index test on test (case when c is not null then a||','||b||','||c end);
select (case when c is not null then a||','||b||','||c end) valfn from test order by B;
output with nls_date_format= 'dd/mm/yyyy hh24:mi:ss'
Index TEST created.
VALFN
------------------------
15/05/2021 08:40:35,1,c1
15/05/2021 08:25:35,1,c1
15/05/2021 08:34:35,2,c2
15/05/2021 08:24:35,2,c2
output with nls_date_format= 'dd/mm/yyyy'
ORA-01452: cannot CREATE UNIQUE INDEX; duplicate keys found
VALFN
---------------
15/05/2021,1,c1
15/05/2021,1,c1
15/05/2021,2,c2
15/05/2021,2,c2
Then we can see what happened when the indexes were created. We will query the index definition to explain this behavior.
output with nls_date_format= 'dd/mm/yyyy hh24:mi:ss'
alter session set nls_date_format= 'dd/mm/yyyy hh24:mi:ss';
drop table test;
create table test (a DATE, b VARCHAR2(10), C VARCHAR2(10));
create unique index test on test (case when c is not null then a||','||b||','||c end);
select index_name, column_expression from all_ind_expressions where table_name = 'TEST'
INDEX_NAME COLUMN_EXPRESSION
---------- ---------------------------------------------------------------------------------------------
TEST CASE WHEN "C" IS NOT NULL THEN TO_CHAR("A",'dd/mm/yyyy hh24:mi:ss')||','||"B"||','||"C" END
output with nls_date_format= 'dd/mm/yyyy'
alter session set nls_date_format= 'dd/mm/yyyy';
drop table test;
create table test (a DATE, b VARCHAR2(10), C VARCHAR2(10));
create unique index test on test (case when c is not null then a||','||b||','||c end);
select index_name, column_expression from all_ind_expressions where table_name = 'TEST'
INDEX_NAME COLUMN_EXPRESSION
---------- ----------------------------------------------------------------------------------
TEST CASE WHEN "C" IS NOT NULL THEN TO_CHAR("A",'dd/mm/yyyy')||','||"B"||','||"C" END
As you can see, the index is not the same because it depends on your nls_date_format when the index was created. I always suggest to validate what was the expression used by Oracle when creating the object.
This is the same when creating virtual columns
alter session set nls_date_format= 'dd/mm/yyyy hh24:mi:ss';
drop table test;
create table test (a DATE, b VARCHAR2(10), C VARCHAR2(10)
,idx varchar2(50) generated always as (case when c is not null then a||','||b||','||c end) virtual
);
select DATA_DEFAULT from all_tab_cols where table_name = 'TEST' and column_name = 'IDX';
DATA_DEFAULT
---------------------------------------------------------------------------------------------
CASE WHEN "C" IS NOT NULL THEN TO_CHAR("A",'dd/mm/yyyy hh24:mi:ss')||','||"B"||','||"C" END
alter session set nls_date_format= 'dd/mm/yyyy';
drop table test;
create table test (a DATE, b VARCHAR2(10), C VARCHAR2(10)
,idx varchar2(50) generated always as (case when c is not null then a||','||b||','||c end) virtual
);
select DATA_DEFAULT from all_tab_cols where table_name = 'TEST' and column_name = 'IDX';
DATA_DEFAULT
----------------------------------------------------------------------------------
CASE WHEN "C" IS NOT NULL THEN TO_CHAR("A",'dd/mm/yyyy')||','||"B"||','||"C" END
Now, you can add C to this equation and validate why in your case : CASE WHEN C IS NOT NULL THEN A||','||B||','||C END is not unique
Most of the time, by default the nls_date_format is not including the time.