Обсуждение: Multi-Column Constraints and Null Values

I have a problem in a number of my tables. I'd like to add uniqueness  
constraints across multiple columns, but some of the columns are  
nullable. This gives me trouble since when I create a constraint on  
columns A and B.. I'd like the constraint to be enforced such that  
you couldn't insert values ("value for A", null) twice. I understand  
why the constraints work this way, but I'm wondering if anyone knows  
of a workaround.

Feel free to spare me any "don't use nullable columns" responses. I  
of course am aware that is an option. It's just one I'm hoping to  
avoid as this is a legacy database and it would be a mammoth effort  
to accomplish.

Thank you in advance!

Phill

On Thu, 27 Apr 2006, Phillip Tornroth wrote:

> I have a problem in a number of my tables. I'd like to add uniqueness 
> constraints across multiple columns, but some of the columns are nullable. 
> This gives me trouble since when I create a constraint on columns A and B.. 
> I'd like the constraint to be enforced such that you couldn't insert values 
> ("value for A", null) twice. I understand why the constraints work this way, 
> but I'm wondering if anyone knows of a workaround.
>

Phill,

You likely want a multicolumn unique index created like so:

CREATE UNIQUE INDEX foo_A_B_unique_idx ON foo (A,B);

See the docs here: 
http://www.postgresql.org/docs/current/interactive/indexes-unique.html


-- 
Jeff Frost, Owner     <jeff@frostconsultingllc.com>
Frost Consulting, LLC     http://www.frostconsultingllc.com/
Phone: 650-780-7908    FAX: 650-649-1954

On Sat, Apr 29, 2006 at 13:14:36 -0700, Jeff Frost <jeff@frostconsultingllc.com> wrote:
> On Thu, 27 Apr 2006, Phillip Tornroth wrote:
> 
> >I have a problem in a number of my tables. I'd like to add uniqueness 
> >constraints across multiple columns, but some of the columns are nullable. 
> >This gives me trouble since when I create a constraint on columns A and 
> >B.. I'd like the constraint to be enforced such that you couldn't insert 
> >values ("value for A", null) twice. I understand why the constraints work 
> >this way, but I'm wondering if anyone knows of a workaround.
> >
> 
> Phill,
> 
> You likely want a multicolumn unique index created like so:
> 
> CREATE UNIQUE INDEX foo_A_B_unique_idx ON foo (A,B);
> 
> See the docs here: 
> http://www.postgresql.org/docs/current/interactive/indexes-unique.html

Creating a multicolumn key is a simpler way of doing the same thing.
However, either way of doing the above doesn't completely solve his problem.
To block multiple entries where one of the columns is null he needs to
add extra checks. One way to do this is to add a partial index for each column
with the condition that the other column IS NULL. This will cover everything
but the case where both columns are null. If this case isn't allowed, then
a row constraint can be used to block it. If it is allowed, the only thing that
comes to mind is making an SQL function that counts the number of rows where
both columns are null and calling that function in a row constraint and
checking that the result is <= 1.

Bruno Wolff III <bruno@wolff.to> writes:
> To block multiple entries where one of the columns is null he needs to
> add extra checks. One way to do this is to add a partial index for each column
> with the condition that the other column IS NULL. This will cover everything
> but the case where both columns are null. If this case isn't allowed, then
> a row constraint can be used to block it. If it is allowed, the only thing that
> comes to mind is making an SQL function that counts the number of rows where
> both columns are null and calling that function in a row constraint and
> checking that the result is <= 1.

Nah, it's easy:
CREATE UNIQUE INDEX fooi ON foo ((0))  WHERE col1 IS NULL AND col2 IS NULL;

So you'd need a total of four unique indexes (3 of 'em partial) to
enforce the condition for two columns.  The real problem with this
is that it doesn't scale well for more than two columns :-(
        regards, tom lane

Wow. This is great feedback. As for more than two columns.. I don't think I have more than two nullable columns. Often
Ihave a three or four column constraint where only one or two of them are nullable. I've never used constraints or
indexeswith where clauses. I'll have to play with this a bit to understand how it works. Thanks for the tips! 

Phill


-----Original Message-----
From: Tom Lane [mailto:tgl@sss.pgh.pa.us]
Sent: Sat 4/29/2006 3:02 PM
To: Bruno Wolff III
Cc: Jeff Frost; Tornroth, Phill; pgsql-sql@postgresql.org
Subject: Re: [SQL] Multi-Column Constraints and Null Values
Bruno Wolff III <bruno@wolff.to> writes:
> To block multiple entries where one of the columns is null he needs to
> add extra checks. One way to do this is to add a partial index for each column
> with the condition that the other column IS NULL. This will cover everything
> but the case where both columns are null. If this case isn't allowed, then
> a row constraint can be used to block it. If it is allowed, the only thing that
> comes to mind is making an SQL function that counts the number of rows where
> both columns are null and calling that function in a row constraint and
> checking that the result is <= 1.

Nah, it's easy:
CREATE UNIQUE INDEX fooi ON foo ((0))  WHERE col1 IS NULL AND col2 IS NULL;

So you'd need a total of four unique indexes (3 of 'em partial) to
enforce the condition for two columns.  The real problem with this
is that it doesn't scale well for more than two columns :-(
        regards, tom lane