17

Using Oracle, if a column value can be 'YES' or 'NO' is it possible to constrain a table so that only one row can have a 'YES' value?

I would rather redesign the table structure but this is not possible.

[UDPATE] Sadly, null values are not allowed in this table.

Trigger
  • 285
  • 1
  • 4
  • 9
  • Oh dear - nulls not permitted - that changes things a little - now you have to go with the function-based index (@Tony Andrews). Still avoid triggers and autonomous transactions. – Nick Pierpoint Oct 09 '08 at 20:01

7 Answers7

21

Use a function-based index:

create unique index only_one_yes on mytable
(case when col='YES' then 'YES' end);

Oracle only indexes keys that are not completely null, and the CASE expression here ensures that all the 'NO' values are changed to nulls and so not indexed.

Tony Andrews
  • 129,880
  • 21
  • 220
  • 259
6

This is a kludgy hack, but if the column allows NULLs, then you could use NULL in place of "NO" and use "YES" just as before. Apply a unique key constraint to that column, and you'll never get two "YES" values, but still have many NOs.

Update: @Nick Pierpoint: suggested adding a check constraint so that the column values are restricted to just "YES" and NULL. The syntax is all worked out in his answer.

yukondude
  • 24,013
  • 13
  • 49
  • 58
4

You will want to check a Tom Kyte article with exactly this question being asked and his answer:

http://tkyte.blogspot.com/2008/05/another-of-day.html

Summary: don't use triggers, don't use autonomous transactions, use two tables.

If you use an Oracle database, then you MUST get to know AskTom and get his books.

Guy
  • 9,720
  • 7
  • 38
  • 42
  • The question asked there is slightly different to this, in Tom's question, the table could have multiple "Y", but only 1 per country. As I read it, in this example, the table can have only 1 Yes, in which case the index solution seems to work. But yes, AskTom is a must for Oracle database. – Matthew Watson Oct 09 '08 at 06:42
  • I agree. If Tom was asked this question I'd guess he'd definitely go for the index solution. – Nick Pierpoint Oct 09 '08 at 19:56
3

It doesn't work on the table definition.

However, if you update the table using a trigger calling a stored procedure, you could make sure that only one row contains "YES".

  1. Set all rows to "NO"
  2. Set the row you want to YES
poezn
  • 4,009
  • 4
  • 25
  • 27
2

Following on from my comment to a previous answer by yukondude, I'd add a unique index and a check constraint:

create table mytest (
    yesorno varchar2(3 char)
);

create unique index uk_mytest_yesorno on mytest(yesorno);

alter table mytest add constraint ck_mytest_yesorno check (yesorno is null or yesorno = 'YES');
Nick Pierpoint
  • 17,641
  • 9
  • 46
  • 74
  • I think that as long as NULL is suitable as the "not yes" value this method has the advantage over the FBI method because the constraints can at least be leveraged by the optimizer. I think that queries with predicates on yesorno='YES' and yesorno is null will get better cardniality estimates. – David Aldridge Oct 08 '08 at 13:36
  • Since there is only ever going to be 1 row with "YES" and every other row with a null, then you'll get the index used to find the "YES" row and a full scan (quite rightly) to get everything else (the "NO"s). – Nick Pierpoint Oct 08 '08 at 14:10
  • Sure -- it's just a question of whether replacing "No" with Null is a worthwhile compromise to achieve this aim. – David Aldridge Oct 08 '08 at 14:33
1

Does Oracle support something like filtered indices (last week I heard that e.g. MSSQL2008 does)? Maybe you can define a unique key which applies only to rows with the value "Yes" in your column.

Matthias Meid
  • 12,455
  • 7
  • 45
  • 79
  • No filtered indexes, but FBIs offer a more flexible (if arguably less concise) way to do the same thing. – orbfish Oct 25 '11 at 16:27
-2

I guess I'd use a second table to point to the appropriate row in your current table. That other table could be used to store values of other variables too too.

Rimas Kudelis
  • 555
  • 3
  • 12