I'm looking to add some code to my TSQL arsenal to defend against performing aggregations in SQL when the data in a column is null. Ideally there would be a SUM_NN (for sum no null), in sql server which would raiserror if any of the values were null.
Since you can't raiserror from a UDF, the only way I could think of doing it looked like this, though I don't like this solution:
CREATE FUNCTION dbo.NULL_TEST_F(#arg FLOAT)
RETURNS FLOAT
AS
BEGIN
IF(#arg IS NULL)
SELECT 1/0
RETURN #arg
END
Note: I think this is stupid solution, but I've gotten burned way too many times when I'm missing data. Also, we're using SQL Server 2005, but I'm open to 2008 and 2012 solutions. Additionally, I'd like to know how other databases deal with this type of issue.
This was my final solution:
CREATE FUNCTION [dbo].[NullTest_F]
(
#input FLOAT,
#message VARCHAR(100)
)
RETURNS FLOAT
AS
BEGIN
DECLARE #test INT
IF(#input IS NULL)
SELECT #test = CAST(#message AS INT)
RETURN #input
END
I can then embed this with a useful error message when running aggregate functions. Example:
CREATE TABLE Data (
Date DATETIME,
DataPoint FLOAT
)
INSERT INTO Data (Date, DataPoint) VALUES ('2012-03-01', 4)
INSERT INTO Data (Date, DataPoint) VALUES ('2012-03-02', 6)
SELECT SUM(NullTest_F(DataPoint, 'Missing data at' + CONVERT(VARCHAR(10), Data))
FROM Data
Maybe this one will help:
https://stackoverflow.com/a/4681815/1371070
You could create a function like suggested in the answer linked above and call it from your aggregate in case #arg is null.
It's still the same strategy overall but It's a better error to throw than divide-by-zero, I guess.
Related
There is a nvarchar(100) column named value, when users insert into this column I need to check this code below in a trigger:
if exists
(
select *
from inserted i
where isnumeric(value)=0
)
begin
rollback transaction
raiserror('when productType is numeric, You have to insert numeric character',18,1)
return
end
but in application interface numbers inserted in persian, so always isnumeric(value)=0.
For example I need to if user insert ۴۵ in interface in my trigger value shown as 45.
So far I use CAST,CONVERT and collate Persian_100_CI_AI but I couldn't get any result.
Thanks.
Which version of SQL Server? v2017+ offers a new function TRANSLATE.
Might be, there is a more elegant way, but a pragmatic one is this:
DECLARE #PersianNumber NVARCHAR(100)=N'۴۵';
SELECT CAST(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE
(#PersianNumber,N'۰',N'0'),N'۱',N'1'),N'۲',N'2'),N'۳',N'3'),N'۴',N'4')
,N'۵',N'5'),N'۶',N'6'),N'۷',N'7'),N'۸',N'8'),N'۹',N'9') AS INT);
Take a look at this topic, it's the opposite of what you asked but it might help you if you could reverse it :
https://social.msdn.microsoft.com/Forums/sqlserver/en-US/a44ce5c1-d487-4043-be73-b64fa98ed7a5/converting-english-numbers-to-arabic-numbers-and-vice-versa
If you are using the latest version of sql server, try this link :
https://learn.microsoft.com/en-us/sql/t-sql/functions/translate-transact-sql
the obvious thing is that SQL does not have a solution out-of-the-box and you have to implement some kind of function yourself and use the returned value in the WHERE statement.
I have used Shungo's answer to implement the function you need (also works for English numbers or a mix of both):
CREATE FUNCTION IS_NORMALIZED_NUMBER (#PersianNumber NVARCHAR(MAX))
RETURNS NVARCHAR(MAX)
BEGIN
SET #PersianNumber = CAST(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE
(#PersianNumber,N'۰',N'0'),N'۱',N'1'),N'۲',N'2'),N'۳',N'3'),N'۴',N'4')
,N'۵',N'5'),N'۶',N'6'),N'۷',N'7'),N'۸',N'8'),N'۹',N'9') AS NVARCHAR(MAX));
RETURN ISNUMERIC(#PersianNumber)
END
Here is a more optimized version (which will only work for Persian numbers) :
CREATE FUNCTION IS_NUMBER (#PersianNumber NVARCHAR(MAX))
RETURNS NVARCHAR(MAX)
BEGIN
RETURN IIF(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE
(#PersianNumber,N'۰',N''),N'۱',N''),N'۲',N''),N'۳',N''),N'۴',N'')
,N'۵',N''),N'۶',N''),N'۷',N''),N'۸',N''),N'۹',N'') = N'',1 ,0 );
END
You can use TRANSLATE (Transact-SQL) function
SELECT TRANSLATE('1234', '0123456789', N'٠١٢٣٤٥٦٧٨٩') AS KurdishNumber
What I'm trying to do is search through a specific column in my database and pull out any rows that match a pattern (i.e. any rows where the nvarchar contains a guid).
I've got this function.
Create FUNCTION [dbo].[HasGuid] (
#Data varchar(50)
) RETURNS bit WITH SCHEMABINDING AS
BEGIN
return case when #Data like REPLACE('%00000000-0000-0000-0000-000000000000%', '0', '[0-9a-fA-F]') then
1
else
0
end
END
It works on most of my data
values it works on:
12341234-1234-1234-1234-123412341234.ext
sometext:12341234-1234-1234-1234-123412341234:c:\path\12341234-1234-1234-1234-123412341234\12341234-1234-1234-1234-123412341234.ext
sometext:12341234-1234-1234-1234-123412341234.ext
values it doesn't work on:
c:\path\12341234-1234-1234-1234-123412341234\12341234-1234-1234-1234-123412341234.ext
Where .ext is one of any number of different extensions and the 1234 guid is just representative and not actually the same guid multiple times in any one string.
And I'm using the function like so (should be obvious, but just to be complete)
SELECT * from SomeTable s
WHERE dbo.HasGuid(s.SomeColumn) = 0
At this point I would expect to see only the rows where that column is not a guid but I'm getting the discrepancy above. I'm not seeing what's wrong in my function, but if someone could point out what I'm missing it'd be super helpful.
It looks like it's just a shortcoming of azure sql. It works as expected outside of azure.
I figured out the actual issue. The #Data declaration was too limiting. The following is the updated function.
Create FUNCTION [dbo].[HasGuid] (#Data nvarchar(400)) RETURNS bit WITH SCHEMABINDING AS BEGIN return case when #Data like REPLACE('%00000000-0000-0000-0000-000000000000%', '0', '[0-9a-fA-F]') then 1 else 0 end END
It's still a difference between azure sql and mssql, but at least it's figured out now.
Good afternoon folks.
I'll preface this with "this was a hard question to ask". I'm running into conversion errors when I think I've got that part covered. Obviously, I do not.
The situation: VARCHAR field with INT data in it, plus some random garbage strings that are causing conversion issues. Here's an example of what I'm trying to do.
DECLARE #MyTABLE TABLE (
Value VARCHAR(50) NOT NULL
);
-- insert some strings
INSERT INTO #MyTABLE (Value) VALUES('400'), ('H-100'), ('H-200'), ('500'),
('600'), ('H-300');
-- conversion fails for the actual strings
SELECT *
FROM #MyTABLE m
WHERE CAST(m.Value AS INT) BETWEEN 1 AND 1000;
-- what I THOUGHT would fix it, but doesn't...
SELECT *
FROM (SELECT * FROM #MyTABLE WHERE Value NOT LIKE 'H%') X
WHERE CAST(X.Value AS INT) BETWEEN 1 AND 1000;
I realize there are other ways that I can do this, such as inserting all BUT the bad data into a temp table and querying that, but I'd like to know why my query doesn't work and what I could do to fix it.
EDIT - This is for SQL 2008 R2
Thanks in advance!
It is an odd bit, if you notice it's the BETWEEN clause that kills it, because the WHERE isn't being evaluated when you think it should, you can use:
SELECT *
FROM #MyTABLE
WHERE CASE WHEN ISNUMERIC(Value)=1 THEN value ELSE 0 END
BETWEEN 1 AND 1000
Demo: SQL Fiddle
Background: I have an SQL database that contain a column (foo) of a text type and not integer. In the column I store integer in a text form.
Question: Is it possible to SELECT the row that contains (in foo column) number greater/lesser than n?
PS: I have a very good reason to store them as text form. Please refrain from commenting on that.
Update: (Forgot to mention) I am storing it in SQLite3.
SELECT foo
FROM Table
WHERE CAST(foo as int)>#n
select *
from tableName
where cast(textColumn as int) > 5
A simple CAST in the WHERE clause will work as long as you are sure that the data in the foo column is going to properly convert to an integer. If not, your SELECT statement will throw an error. I would suggest you add an extra step here and take out the non-numeric characters before casting the field to an int. Here is a link on how to do something similar:
http://blog.sqlauthority.com/2007/05/13/sql-server-udf-function-to-parse-alphanumeric-characters-from-string/
The only real modification you would need to do on this function would be to change the following lines:
PATINDEX('%[^0-9A-Za-z]%', #string)
to
PATINDEX('%[^0-9]%', #string)
The results from that UDF should then be castable to an int without it throwing an error. It will further slow down your query, but it will be safer. You could even put your CAST inside the UDF and make it one call. The final UDF would look like this:
CREATE FUNCTION dbo.UDF_ParseAlphaChars
(
#string VARCHAR(8000)
)
RETURNS int
AS
BEGIN
DECLARE #IncorrectCharLoc SMALLINT
SET #IncorrectCharLoc = PATINDEX('%[^0-9]%', #string)
WHILE #IncorrectCharLoc > 0
BEGIN
SET #string = STUFF(#string, #IncorrectCharLoc, 1, '')
SET #IncorrectCharLoc = PATINDEX('%[^0-9]%', #string)
END
SET #string = #string
RETURN CAST(#string as int)
END
GO
Your final SELECT statement would look something like this:
SELECT *
FROM Table
WHERE UDF_ParseAlphaChars(Foo) > 5
EDIT
Based upon the new information that the database is SQLite, the above probably won't work directly. I don't believe SQLite has native support for UDFs. You might be able to create a type of UDF using your programming language of choice (like this: http://www.christian-etter.de/?p=439)
The other option I see to safely get all of your data (an IsNumeric would exclude certain rows from your results, which might not be what you want) would probably be to create an extra column that has the int representation of the string. It is a little more dangerous in that you need to keep two fields in sync, but it will allow you to quickly sort and filter the table data.
SELECT *
FROM Table
WHERE CAST(foo as int) > 2000
I understand that T-SQL is not object oriented. I need to write a set of functions that mimics method overloading in C#.
Is function overloading supported in T-SQL in any way? If there is a hack to do this, is it recommended?
No, there is no way to do this.
I recommend you revisit the requirement, as "make apples look like oranges" is often difficult to do, and of questionable value.
One thing I have done successfully is to write the function in such a way as to allow it to handle null values, and then call it with nulls in place of the parameters you would like to omit.
Example:
create function ActiveUsers
(
#departmentId int,
#programId int
)
returns int
as
begin
declare #count int
select #count = count(*)
from users
where
departmentId = isnull(#departmentId, departmentId)
and programId = isnull(#programId, programId)
return #count
end
go
Uses:
select ActiveUsers(1,3) -- users in department 1 and program 3
select ActiveUsers(null,3) -- all users in program 3, regardless of department
select ActiveUsers(null,null) -- all users
You could pass in a sql_variant, but it comes with all sorts of hazards around it; you can't really use strong typing like you can with OO languages and overloading.
If you need to find the base type within your function, you can use the SQL_VARIANT_PROPERTY function.
You can pass in an array of values within a single string and parse them out using this techique by Erland Sommarskog.
Create a function with a varchar(max) parameter or several if necessary, then have your parameter values in that string like:
param1;param2;parma3;param4
or
param1:type;param2:type;param3:type
or
calltype|param1;param2;param3
etc, you are only limited by your imagination...
Use the technique from the link to split apart this array and use program logic to use those values as you wish.
One solution would be to utilize the sql_variant data type. This example works as long as you use the same datatype for both values. Returns whatever datatype you send it.
create function dbo.Greater(
#val1 sql_variant
,#val2 sql_variant
) returns sql_variant
as
begin
declare #rV sql_variant
set #rV = case when #val1 >= #val2 then #val1
else #val2 end
return #rV
end
go
A solution I've had some luck with is either creating a number of functions that each takes a different data type - or casting all input to NVARCHAR(MAX).
1. creating a number of functions that each takes a different data type
CREATE FUNCTION [dbo].[FunctionNameDatetime2]
CREATE FUNCTION [dbo].[FunctionNameInt]
CREATE FUNCTION [dbo].[FunctionNameString] --(this is not a typo)
CREATE FUNCTION [dbo].[FunctionNameUniqueidentifier]
...
Problem: duplication of code, and a lot functions
2. Cast all input to NVARCHAR(MAX)
CREATE FUNCTION [dbo].[IntToNvarchar]
(
#Key INT
)
RETURNS NVARCHAR(MAX)
AS
BEGIN
RETURN ISNULL(CAST(#Key AS NVARCHAR), '');
END
CREATE FUNCTION [dbo].[FunctionName]
(
#Key NVARCHAR(MAX)
)
RETURNS CHAR(32)
AS
BEGIN
DECLARE #something CHAR(32)
do stuff ...
RETURN #something;
END
SELECT [dbo].[FunctionName]([dbo].[IntToNvarchar](25))
Problem: less elegant code than overloading
I overload Functions all the time, but I happen to know that these kind of issues are often highly dependent on platform.
On our DB2 system, I routinely overload like the following:
CREATE Function Schema1.F1 (parm date)
returns date
return date + 1month;
CREATE Function Schema1.F1 (parm timestamp)
returns date
return date(timestamp) + 1month;
This is actually quite useful when you have multiple queries which have similar formating requirements.
The only problem I have found about this so far, is you better be sure that you want the function because the standard drop function "schema"."name" fails because it cannot determine which function to drop. If anyone knows how to drop overloaded sql functions, let me know!