Introduction
Many programming languages have supported User Defined Functions for years, but they
are new to SQL Server 2000. In this article we will look at some of the
ways functions can be used within SQL Server 2000.
This document was developed against SQL Server 2000 Beta 2, so some points may be subject to change in the release version.
What are functions?
Functions are pre-prepared pieces of code that may accept parameters, and
always return a value. Older versions of SQL Server supply some built-in
functions, such as getdate() and object_name() but to be able to
define your own functions is a new feature in SQL Server 2000.
Throughout the text I will use the
abbreviation UDF for User Defined Function.
All functions (not just UDFs) are split into two groups, deterministic and non-deterministic.
Deterministic functions will always return the save value if provided with the
same parameters. Non-deterministic functions may produce different results each
time they are called, even if any supplied parameters are unchanged.
SQL Server Books Online also uses the term "scalar functions" -
these are simply functions that return one value, as opposed to functions that
return a table.
getdate() is a good example of both a scalar and a non-deterministic function - call it
twice in a row and you will get two different date-time results, but only one
value from each call.
A simple UDF
It's not uncommon to want to know the number of days in particular month, so
let's look at a simple function to do this. Here is the code:
create function DaysInMonth (@when datetime)
returns int
as
BEGIN
declare@rv int
if datepart(month, @when) = 2
begin
if datepart(year, @when) % 400 = 0
select @rv = 29
else
if datepart(year, @when) % 100 = 0
select @rv = 28
else
if datepart(year, @when) % 4 = 0
select @rv = 29
else
select @rv = 28
end -- if
else
begin
select @rv = case (datepart(month, @when))
when 4 then 30
when 6 then 30
when 9 then 30
when 11 then 30
else 31
end -- case
end -- else
return @rv
END -- fn def
To execute the function,
select dbo.DaysInMonth('1 may 2000')
select dbo.DaysInMonth(getdate())
As you can see, most of the code could have come from any SQL Server stored
procedure. The new features are:
create function DaysInMonth (@when
datetime)
returns int
Clearly this is a function definition header - the returns int states
that the function returns a single integer value.
return @rv
The return @rv command must be the final line of the function, and
it's this instruction that sends the output value back to the call. Almost any
SQL Server data type can be returned by a UDF.
Using function calls
You can code a function call anywhere you could code a variable or literal of
the same data type as the function returns. For example, to calculate the number
of days between now and the end of the month:
select dbo.DaysInMonth(getdate()) - datepart(day,
getdate())
Note the repeated use of the owner qualification before the UDF name, (dbo.DaysInMonth()
instead of DaysInMonth() - This qualification is a SQL Server requirement
in Beta 2, and will probably go forward into the release version.
Functions within functions
You can nest a function within another of you so wish.
create function daysTillMonthEnd (@when Datetime)
returns int
as
begin
return dbo.DaysInMonth(@when) - datepart(day, @when)
end
Restrictions on using functions within functions
There is a restriction on this which I ran straight into when I first started to play with UDFs. You cannot use a built-in non-deterministic
function within a user defined function. When I made my first stab at coding
the DaysInMonth function shown above, I tried to use the getdate()
function as a default in case no parameters were supplied in the function call.
SQL Server will not allow this.
Using other data types
You can return almost any data type in a function, including User Defined Data
Types. The following simple example returns a UDDT based on the varchar data
type.
sp_addtype varstring, 'varchar(32)'
go
create function capitalise (@string varstring)
returns varstring
as
BEGIN
return UPPER(left(@string, 1)) +
LOWER(substring (@string, 2, datalength(@string)))
END
go
select dbo.capitalise ('abcdef')
SQL Server 2000 will not allow you to return text, image, cursor or timestamp
data types from a function.
Recursive functions
Functions have the ability to call themselves, or "recurse". The
example below works out the factorial of the integer input.
create function factorial (@i bigint)
returns bigint
as
BEGIN
declare @internalI bigint
if @i > 20 OR @i IS NULL
select @internalI = NULL
else
if @i < 2
select @internalI = @i
else
select @internalI = @i * dbo.factorial(@i - 1)
return @internalI
END -- fn def
go
select dbo.factorial (3)
select dbo.factorial (8)
For the non-mathematically inclined, factorials are what you get when you
multiply together all the whole numbers between 1 and the number you thought of.
For example:
Functions can recurse up to level 32 deep, after which SQL Server will generate an
error. The example above is restricted to 20 times because Factorial
21 is too big for the bigint data type.
Passing tables out of functions
Create a function that passes back a table and what you end up with is a kind
of "parameterized view" - something you can treat as a view, but which
will return different values depending on the parameters passed in. The
following example for the PUBS database takes an author_id as it's parameter and
returns a table containing all the author IDs that have collaborated with them,
and the title_id they worked together on.
create function collaborator (@au_id char(11))
returns @collaborator table (
target_authorchar(11),
title_idchar(6),
partner_authorchar(11)
)
as
BEGIN
insert @collaborator
select l.au_id, l.title_id, r.au_id
from titleauthor l inner join titleauthor r
on l.title_id = r.title_idAND l.au_id <> r.au_id
where l.au_id = COALESCE(@au_id, l.au_id)
orderby l.title_id, l.au_id
RETURN
END -- fn def
select * from dbo.collaborator('472-27-2349')
select * from dbo.collaborator(NULL)
Note the use of the COALESCE function. We use this to return data for all
authors that have collaborated on a book if no specific author_id is given.
You could code the line create function collaborator (@au_id char(11) =
NULL) to define a default value for the parameter, and SQL Server would
accept this. However, if you then try calling the function without parameters (a
perfectly valid thing to do with stored procedure parameters) then SQL Server
will complain. Instead you have to code dbo.collaborator(DEFAULT) for the
function call.
In-line table functions
SQL Server differentiates between "user defined functions that return a
table" and "in-line table functions". The previous example is a
"user defined functions that return a table". We can re-code this as an
in-line function as follows:
create function collaborator_inline (@au_id char(11))
returns table as
RETURN (
select l.au_id as "author",
l.title_id as "title",
r.au_id as "co-author"
from titleauthor l inner join titleauthor r
on l.title_id = r.title_id AND l.au_id <> r.au_id
where l.au_id = COALESCE(@au_id, l.au_id)
-- order by l.title_id, l.au_id
)
go
select * from dbo.collaborator_inline('724-80-9391')
select * from dbo.collaborator_inline(NULL)
This syntax is more compact than the previous example, but also more
restrictive. While the "user defined functions that return a table"
can include any amount of code, the in-line version is restricted to a single
select statement. Even the "order by" clause needs to be removed from
the in-line version.
Table function in joins
Parameterized functions that return tables may be part of a standard SQL join
statement.
select *
from dbo.collaborator('472-27-2349') inner join authors as a
on partner_author = a.au_id
Functions as columns in views
UDFs can be used as columns in a view.
Building on the daysInMonth() function we defined earlier, and
a new function startOfMonth() function, we can create this
NORTHWIND database view showing daily average sales figures per month without
going to excessive lengths to cater for months with different numbers of days.
create function startOfMonth (@when datetime)
returns smalldatetime as
BEGIN
return dateadd (day, 1 + datepart(day, @when) * -1, @when)
END -- fn def
go
create view monthlyAverages as
selectmonth,
sum(quantity) /
dbo.daysInMonth(month)
as dailyAvgSale
from (selectdbo.startofMonth (orderDate)
as month,
quantity,
o.OrderID
fromorders o
inner join "order details" od
on o.OrderID = od.OrderID
) as derived
group by month
go
select * from monthlyAverages order by month
Functions and constraints
You can use functions as to define constraints, provided that both the table
and the function share the same owner. This example defines a function called
" midnight" that truncates a date-time to midnight, then uses this in
a table's check constraint to ensure that the prevShutdown column in test_table
is always at least a day earlier then the value in the startup column.
create function midnight (@when smalldatetime)
returns smalldatetime
as
BEGIN
return convert(varchar(11), @when)
END -- fn def
go
create table test_table (
id int identity,
prevShutdown smalldatetime,
startup
smalldatetime,
CONSTRAINT chk_midnight CHECK (prevShutdown < dbo.midnight(startup))
)
go
You can also use a function as the basis for a default constraint, but
presently you are limited to using constants in the function parameters, which
restricts their usefulness.
Schema bound functions
Particularly with older versions of SQL Server, you could get into trouble if
you defined one object that was dependent on another, then dropped or
changed the subordinate object. (In plain English: if you created a view on a
table, then dropped the table, the view would still exist but it would obviously
not work)
Schema binding tries to put a stop to this. If you create a function and
specify it is Schema Bound, then the objects it depends on cannot be altered
without you first dropping the schema binding. There are limitations as to when
this will work (it will not work across multiple databases for example) but it
is still a very useful feature.
Schema binding is invoked at function creation time. We can change our collaborator()
function to a schema bound version as shown below. Note that in addition to
the with schemabinding
option, we need to specify a two part name (dbo.titleauthor
instead of just titleauthor) for all referenced objects.
create function collaborator_sb (@au_id char(11))
returns table
with schemabinding
as
RETURN (
select l.au_id as "author",
l.title_id as "title",
r.au_id as "co-author"
from dbo.titleauthor l inner join
dbo.titleauthor r
on l.title_id = r.title_id AND l.au_id <> r.au_id
where l.au_id = COALESCE(@au_id, l.au_id)
-- order by l.title_id, l.au_id
)
As long as schema binding is in place for this function, any attempt to
change the structure of the titleauthor table will generate an error.
Overloading functions
Overloading is the practice of having multiple functions with the name name,
each doing different things. The '+' operator is overloaded as it can both add
up numbers and concatenate strings.
select 1 + 2 + 3
select '1' + '2' + '3'
In Beta 2 at least, overloading functions is a non-starter due to the UNIQUE
constraint on the sysobjects columns name and uid. There is
some discussion in the Beta support newsgroups about implementing this in
future, but I would not hold my breath.
About the author
Neil Boyle is an independent SQL Server consultant working out of London,
England. Neil's free SQL Server guide is available on-line at http://www.impetus-sql.co.uk
