Indexing with null or zero

I am adding an int field to a table that will be used to link it to another
table when it has a number in it. I plan to index it for faster access but
wondered if my default for that table should be NULL or zero or if it makes
a difference. Thanks.
DavidTo an index, NULL is just a value (as is 0). So use whatever makes most sens
e to you.
Tibor Karaszi, SQL Server MVP
http://www.karaszi.com/sqlserver/default.asp
http://www.solidqualitylearning.com/
"David C" <dlchase@.lifetimeinc.com> wrote in message news:uy%238XzdSFHA.252@.TK2MSFTNGP12.ph
x.gbl...
>I am adding an int field to a table that will be used to link it to another
table when it has a
>number in it. I plan to index it for faster access but wondered if my defa
ult for that table
>should be NULL or zero or if it makes a difference. Thanks.
> David
>|||I have to say that it would be nice if there was a built-in means to create
a
unique constraint on a column such that nulls are ignored. Going the route o
f
indexed views or triggers is a colossal pain.
Thomas
"Tibor Karaszi" <tibor_please.no.email_karaszi@.hotmail.nomail.com> wrote in
message news:OmrRbKeSFHA.580@.TK2MSFTNGP15.phx.gbl...
> To an index, NULL is just a value (as is 0). So use whatever makes most se
nse
> to you.
> --
> Tibor Karaszi, SQL Server MVP
> http://www.karaszi.com/sqlserver/default.asp
> http://www.solidqualitylearning.com/
>
> "David C" <dlchase@.lifetimeinc.com> wrote in message
> news:uy%238XzdSFHA.252@.TK2MSFTNGP12.phx.gbl...
>|||Tibor,
I don't think this is quite true. A column's nullability (assuming
that is part of the choice here) affects what query plans can be
considered. Also, indexes aside, there can be different and
unexpected results depending on whether NULL or 0 is used.
An example we often see in the newsgroups is
select this, that
from T
where T.parent not in (
select parent
from T_children
)
which returns no rows even when there are T.parent values
not present in the T_children table.
There are other differences less likely to be an issue, like
calculating percentages:
select
T.parent,
sum(case when parent = T_children.parent then 1.00 else 0.00 end) /
count(T_children.parent) as percentage
from T join T_children
on T.parent = T_children.parent
group by T.parent
Unfortunately, while using NULL is probably truer to the principles
of good modeling, using 0 may have some practical advantages, if
the referring column can then be declared as NOT NULL. To maintain
referential integrity via a foreign key constraint, NULL is a better choice,
since 0 requires putting a dummy entry into the referenced table.
Steve Kass
Drew University
Tibor Karaszi wrote:

>To an index, NULL is just a value (as is 0). So use whatever makes most sen
se to you.
>
>|||
Thomas wrote:

>I have to say that it would be nice if there was a built-in means to create
a
>unique constraint on a column such that nulls are ignored. Going the route
of
>indexed views or triggers is a colossal pain.
>
>
There is at least one solution using an indexed column instead of
an indexed view. Functionally, it is much the same, but you might
find it easier to maintain.
http://groups.google.co.uk/groups?q...B8-B7567063D1CC
SK

>Thomas
>
>"Tibor Karaszi" <tibor_please.no.email_karaszi@.hotmail.nomail.com> wrote in
>message news:OmrRbKeSFHA.580@.TK2MSFTNGP15.phx.gbl...
>
>
>|||Hmm..That's an clever solution. For the purposes of other people reading thi
s
thread, the idea is to create a computed column that either equals the colum
n on
which you really want a unique index or the primary key when that value is n
ull.
You would then put the unique index on this computed column.
Granted, it does mean dealing with computed columns which can be a persnicke
ty
but it does get around the issue.
Thomas
"Steve Kass" <skass@.drew.edu> wrote in message
news:uSJlWXeSFHA.3788@.tk2msftngp13.phx.gbl...
>
> Thomas wrote:
>
> There is at least one solution using an indexed column instead of
> an indexed view. Functionally, it is much the same, but you might
> find it easier to maintain.
> http://groups.google.co.uk/groups?q...B8-B7567063D1CC
> SK
>|||If a Foreign Key column allows null, then this is definitely NOT the same as
putting a zero in the column. A value of zero MUST exist in the reference
table, a null value does NOT have to (indeed it cannot exist) as PK in the
reference table. The distinction is critical, it is the dfference between a
[one]-to-[zero or many] relationship, and a [zero or one]-to-[zero or many]
relationship.
Example, In a [one]-to-[zero or many] Employees have zero o many timecard
punches, but for each time card punch there must be one employee - and only
one an employee.
In a [zero or one]-to-[zero or many] relatonship, Each Library patron can
have zero or many books checked out to them, and each book can be checked
out to zero or one library patron...
The only way to model this distinction is by allowing, (and using) null
values in the Foreign Key column on the many side of the DRI constraint.
"Tibor Karaszi" wrote:

> To an index, NULL is just a value (as is 0). So use whatever makes most se
nse to you.
> --
> Tibor Karaszi, SQL Server MVP
> http://www.karaszi.com/sqlserver/default.asp
> http://www.solidqualitylearning.com/
>
> "David C" <dlchase@.lifetimeinc.com> wrote in message news:uy%238XzdSFHA.25
2@.TK2MSFTNGP12.phx.gbl...
>
>|||When I want a Foreign Key Column to be unique, but Allow any number of Nulls
,
(that's a [One]-to-[Zero or One] Relationship, I use an extra table
CREATE TABLE [TabA] (
[AID] [int] NOT NULL ,
[Name] [varchar] (25) COLLATE SQL_Latin1_General_CP1_CI_AS NOT NULL ,
CONSTRAINT [PK_TabA] PRIMARY KEY CLUSTERED
([AID]) ON [PRIMARY]
) ON [PRIMARY]
GO
-- ****************************************
***
CREATE TABLE [TabB] (
[BID] [int] NOT NULL ,
[AID] [int] NULL ,
CONSTRAINT [PK_TabB] PRIMARY KEY CLUSTERED
( [BID]) ON [PRIMARY] ,
CONSTRAINT [FK_TabB_TabA] FOREIGN KEY
([AID]) REFERENCES [TabA] ([AID])
) ON [PRIMARY]
GO
-- ****************************************
***
CREATE TABLE [TabC] (
[BID] [int] NOT NULL ,
[AID] [int] NOT NULL ,
CONSTRAINT [PK_TabC] PRIMARY KEY CLUSTERED
([BID]) ON [PRIMARY] ,
CONSTRAINT [IX_TabCUniqueAID] UNIQUE NONCLUSTERED
([AID]) ON [PRIMARY] ,
CONSTRAINT [FK_TabC_TabB] FOREIGN KEY
([BID]) REFERENCES [TabB] ([BID])
) ON [PRIMARY]
GO
-- Then I add ALL the child records to TabB, Both those with null values of
AID, and Non-Null values of AID, but only add the Non-Null AID Records to
TabC... TabB is teh real table, and has all the other attriobutes in it,
TabC is only there to enforce uniqueness on the Non-Null Values of AID in Ta
bB
"Thomas" wrote:

> Hmm..That's an clever solution. For the purposes of other people reading t
his
> thread, the idea is to create a computed column that either equals the col
umn on
> which you really want a unique index or the primary key when that value is
null.
> You would then put the unique index on this computed column.
> Granted, it does mean dealing with computed columns which can be a persnic
kety
> but it does get around the issue.
>
> Thomas
>
>
> "Steve Kass" <skass@.drew.edu> wrote in message
> news:uSJlWXeSFHA.3788@.tk2msftngp13.phx.gbl...
>
>|||Perhaps I was in a bit too much hurry when responding. I was looking at it p
urely from the index'
perspective, how the data is stored in the index. You definitely want to thi
nk through semantics
carefully. :-)
Tibor Karaszi, SQL Server MVP
http://www.karaszi.com/sqlserver/default.asp
http://www.solidqualitylearning.com/
"CBretana" <cbretana@.areteIndNOSPAM.com> wrote in message
news:EC2328B5-9A67-4E1A-A096-9B071AE179EB@.microsoft.com...
> If a Foreign Key column allows null, then this is definitely NOT the same
as
> putting a zero in the column. A value of zero MUST exist in the referenc
e
> table, a null value does NOT have to (indeed it cannot exist) as PK in the
> reference table. The distinction is critical, it is the dfference between
a
> [one]-to-[zero or many] relationship, and a [zero or one]-to-[zero or many]
> relationship.
>
> Example, In a [one]-to-[zero or many] Employees have zero o many timecard
> punches, but for each time card punch there must be one employee - and on
ly
> one an employee.
> In a [zero or one]-to-[zero or many] relatonship, Each Library patron can
> have zero or many books checked out to them, and each book can be checked
> out to zero or one library patron...
> The only way to model this distinction is by allowing, (and using) null
> values in the Foreign Key column on the many side of the DRI constraint.
> "Tibor Karaszi" wrote:
>|||Tibor,
Neglected to make clear that details in my post,were, of course, not
directed at you... figured you just overlooked that. I slip into
pedanticLand way too easily...
Respectfully, Charly
"Tibor Karaszi" wrote:

> Perhaps I was in a bit too much hurry when responding. I was looking at it
purely from the index'
> perspective, how the data is stored in the index. You definitely want to t
hink through semantics
> carefully. :-)
> --
> Tibor Karaszi, SQL Server MVP
> http://www.karaszi.com/sqlserver/default.asp
> http://www.solidqualitylearning.com/
>
> "CBretana" <cbretana@.areteIndNOSPAM.com> wrote in message
> news:EC2328B5-9A67-4E1A-A096-9B071AE179EB@.microsoft.com...
>
>