tsql very optimized UDF - sql

Question

What is the best code (UDF) to calculate the age given birthday? Very optimized code. I can write the date related code just to make sure what best answer I get from best minds in the industry.
Would like to calculate based on days.

Answer 1

The following function gives very reasonable answers for all the date ranges that I've commented in this thread. You will notice that some values may be off by 1 day, but this is an artifact of how months are counted due to them being varying length. February always gives the most trouble being the shortest. But it should never be off by more than 1 day from dead reckoning (and even that comes down to semantics of how you are supposed to count the months, and is still a valid answer). Sometimes, it may give a number of days when you might have expected 1 month, but again it is still a reasonable answer.
The function, using only date math instead of string operations until the very end, should also yield very good performance when stacked up against any other function that returns calculations that are similarly accurate.
CREATE FUNCTION dbo.AgeInYMDFromDates(
#FromDate datetime,
#ToDate datetime
)
RETURNS TABLE
WITH SCHEMABINDING
AS
RETURN (
SELECT
Convert(varchar(11), AgeYears) + 'y '
+ Convert(varchar(11), AgeMonths) + 'm '
+ Convert(varchar(11), DateDiff(day, DateAdd(month, AgeMonths,
DateAdd(year, AgeYears, #FromDate)), #ToDate)
) + 'd' Age
FROM (
SELECT
DateDiff(year, #FromDate, #ToDate)
- CASE WHEN Month(#FromDate) * 32 + Day(#FromDate)
> Month(#ToDate) * 32 + Day(#ToDate) THEN 1 ELSE 0 END AgeYears,
(DateDiff(month, #FromDate, #ToDate)
- CASE WHEN Day(#FromDate) > Day(#ToDate) THEN 1 ELSE 0 END) % 12 AgeMonths
) X
);
Use like so (SQL 2008 script):
SELECT
FromDate,
ToDate,
ExpectedYears,
ExpectedMonths,
ExpectedDays,
(SELECT TOP 1 Age FROM dbo.AgeInYMDFromDates(FromDate, ToDate)) Age
FROM
(
VALUES
(Convert(datetime, '20120201'), Convert(datetime, '20120301'), 0, 1, 0),
('20120228', '20120328', 0, 1, 0),
('20120228', '20120331', 0, 1, 3),
('20120228', '20120327', 0, 0, 27),
('20120801', '20120802', 0, 0, 1),
('20120131', '20120301', 0, 1, 2),
('19920507', '20120506', 19, 11, 29),
('19920507', '20120507', 20, 0, 0)
) X (FromDate, ToDate, ExpectedYears, ExpectedMonths, ExpectedDays)
Because it is an inline function, it can be inserted into the execution plan of the query and will perform the best possible. If you convert it to a scalar-returning function (so you don't have to use (SELECT Age FROM func) then you will get worse performance. The WITH SCHEMABINDING directive can help because it precalculates that the function makes no data access to any tables rather than having to check it at runtime.
Here are the results of the above execution:
FromDate ToDate ExpectedYears ExpectedMonths ExpectedDays Age
---------- ---------- ------------- -------------- ------------ -----------
2012-02-01 2012-03-01 0 1 0 0y 1m 0d
2012-02-28 2012-03-28 0 1 0 0y 1m 0d
2012-02-28 2012-03-31 0 1 3 0y 1m 3d
2012-02-28 2012-03-27 0 0 27 0y 0m 28d
2012-08-01 2012-08-02 0 0 1 0y 0m 1d
2012-01-31 2012-03-01 0 1 2 0y 1m 1d
1992-05-07 2012-05-06 19 11 29 19y 11m 29d
1992-05-07 2012-05-07 20 0 0 20y 0m 0d
Enjoy!

Answer 2

Try This.
CREATE TABLE patient(PatName varchar(100), DOB date, Age varchar(100))
INSERT INTO patient
VALUES ('d','06/02/2011',NULL)--,('b','07/10/1947',NULL),('c','12/21/1982',NULL)
;WITH CTE(PatName,DOB,years,months,days) AS
(SELECT PatName, DOB, DATEDIFF(yy,DOB,getdate()), DATEDIFF(mm,DOB,getdate()),
DATEDIFF(dd,DOB,getdate())
FROM patient)
SELECT PatName, DOB,
CAST(months/12 as varchar(5)) + ' Years' +
CAST((months % 12) as varchar(5)) + ' month/s ' +
CAST( CASE WHEN DATEADD(MM,(months % 12), DATEADD(YY,(months/12),DOB)) <= GETDATE()
THEN DATEDIFF(dd,DATEADD(MM,(months % 12), DATEADD(YY,(months/12),DOB)),GETDATE())
ELSE DAY(getdate())
END as varchar(5))+' days' AS Age
FROM CTE

Answer 3

Performance of scalar functions is usually not very good, to do what you ask I would write it as a table valued function. A table valued function has the benefit that it is inlined properly.
Other forms of the query will not make a huge difference as it is the calls to the function which eat up the time.
CREATE FUNCTION dbo.fn_age(#birthdate datetime, #current_date datetime)
RETURNS TABLE
WITH SCHEMABINDING
AS
return (
select CAST(DATEDIFF(year, #birthdate, #current_date) - CASE WHEN((MONTH(#birthdate) * 100 + DAY(#birthdate)) > (MONTH(#current_date) * 100 + DAY(#current_date))) THEN 1 ELSE 0 END AS varchar(3)) + ' y, '
+ CAST(DATEDIFF(MONTH, DATEADD(year
, /* the year calculation*/ DATEDIFF(year, #birthdate, #current_date) - CASE WHEN((MONTH(#birthdate) * 100 + DAY(#birthdate)) > (MONTH(#current_date) * 100 + DAY(#current_date))) THEN 1 ELSE 0 END /* end of the year calculation*/
, #birthdate),
#current_date) AS varchar(2)) + ' m, '
+ CAST(DATEDIFF(day, DATEADD(month
, datediff(MONTH, DATEADD(year
, /* the year calculation*/ DATEDIFF(year, #birthdate, #current_date) - CASE WHEN((MONTH(#birthdate) * 100 + DAY(#birthdate)) > (MONTH(#current_date) * 100 + DAY(#current_date))) THEN 1 ELSE 0 END /* end of the year calculation*/
, #birthdate)
, #current_date)
, dateadd(year
, /* the year calculation*/ DATEDIFF(year, #birthdate, #current_date) - CASE WHEN((MONTH(#birthdate) * 100 + DAY(#birthdate)) > (MONTH(#current_date) * 100 + DAY(#current_date))) THEN 1 ELSE 0 END /* end of the year calculation*/
, #birthdate)
)
, #current_date) AS varchar(2)) + ' d' as [Age]
)
GO
This function has to be called like this:
SELECT Age = (SELECT Age FROM dbo.fn_age(birthDate, current_timestamp))
FROM Person
Comparison with other alternatives
When writing this problem as a normal scalar function I would create something like this:
CREATE FUNCTION dbo.fn_age_slow(#birthdate datetime, #current_date datetime )
RETURNS VARCHAR(10)
WITH SCHEMABINDING
AS
begin
return CAST(DATEDIFF(year, #birthdate, #current_date) - CASE WHEN((MONTH(#birthdate) * 100 + DAY(#birthdate)) > (MONTH(#current_date) * 100 + DAY(#current_date))) THEN 1 ELSE 0 END AS varchar(3)) + ' y, '
+ CAST(DATEDIFF(MONTH, DATEADD(year
, /* the year calculation*/ DATEDIFF(year, #birthdate, #current_date) - CASE WHEN((MONTH(#birthdate) * 100 + DAY(#birthdate)) > (MONTH(#current_date) * 100 + DAY(#current_date))) THEN 1 ELSE 0 END /* end of the year calculation*/
, #birthdate),
#current_date) AS varchar(2)) + ' m, '
+ CAST(DATEDIFF(day, DATEADD(month
, datediff(MONTH, DATEADD(year
, /* the year calculation*/ DATEDIFF(year, #birthdate, #current_date) - CASE WHEN((MONTH(#birthdate) * 100 + DAY(#birthdate)) > (MONTH(#current_date) * 100 + DAY(#current_date))) THEN 1 ELSE 0 END /* end of the year calculation*/
, #birthdate)
, #current_date)
, dateadd(year
, /* the year calculation*/ DATEDIFF(year, #birthdate, #current_date) - CASE WHEN((MONTH(#birthdate) * 100 + DAY(#birthdate)) > (MONTH(#current_date) * 100 + DAY(#current_date))) THEN 1 ELSE 0 END /* end of the year calculation*/
, #birthdate)
)
, #current_date) AS varchar(2)) + ' d'
end
GO
As you can see it does exactly the same as the table valued function. (it is also schema bound which makes the functions faster in some cases)
When running the following script against the first function (on my pc)
declare #a varchar(10) = ''
, #d datetime = '20120101'
, #i int = 1
, #begin datetime
select #begin = CURRENT_TIMESTAMP
while #i < 1000000
begin
select #a = Age
, #d = #begin - #i%1000
, #i += 1
from dbo.fn_age(#d, #begin)
end
select CURRENT_TIMESTAMP - #begin
GO
==> 00:00:07.920
declare #a varchar(10) = ''
, #d datetime = '19500101'
, #i int = 1
, #begin datetime
select #begin = CURRENT_TIMESTAMP
while #i < 1000000
begin
select #a = dbo.fn_age_slow(#d, #begin)
, #d = #begin - #i%1000
, #i += 1
end
select CURRENT_TIMESTAMP - #begin
==> 00:00:14.023
This is in no way a proper benchmark but it should give you an idea about the performance difference.