Conditional Statements in WHERE Clauses

Question

Conditional Statements in WHERE Clauses

Sotiris Filippidis

SSC Enthusiast

Points: 140
More actions
July 9, 2004 at 6:07 pm

#164221

Comments posted to this topic are about the content posted at http://www.sqlservercentral.com/col
Sotiris L. Filippidis

Viewing 15 posts - 1 through 15 (of 35 total)

You must be logged in to reply to this topic. Login to reply

laperrej SSC Journeyman Points: 94 More actions · Answer 1

Although an interesting technique, from a database performance point of view, this is not a very good idea. SARGs are turned into non-SARGs, so indexes won't be used (for an example, rewrite

SELECT * FROM authors WHERE au_lname LIKE 'm%' AND state = 'CA'

in the pubs database with conditional statements in the WHERE clause and compare the execution plans. You will see that the query above uses an index, the query with conditional statements in the WHERE clause does not.

Sotiris Filippidis SSC Enthusiast Points: 140 More actions · Answer 2

You are right. My point, though, was to show how an operator-driven query could be performed, actually neglecting any performance issues. There are some cases, for example, a custom query builder inside an application, where a technique like that could prove handy, especially when dealing with ad hoc reporting inside a database with limited data but needing to have many different views.

I will investigate further and maybe come out with an even better solution. Thank you for your comments!

Sotiris L. Filippidis

AKM SSCommitted Points: 1538 More actions · Answer 3

I would be interested to see the relative performance of the article's method versus the following approach, which is equally powerful but (I think) can be optimised much more easily by the query engine:

-- declare condition variables as in the article

if @companynameOp='' and @countryOp=''

  select customerid, companyname, country from customers

else if @companynameOp='' and @countryOp='eq'

  select customerid, companyname, country from customers where country like @country

else if @companynameOp='' and @countryOp='bg'

  select customerid, companyname, country from customers where country like @country+'%'

-- etc etc

else if @companynameOp='ct' and @countryOp='ne'

  select customerid, companyname, country from customers where companyname like '%' + @companyName +'%' and country not like @country

-- etc etc

OK it's much more typing but that shouldn't really be considered an issue. As you can see, we are handling each possible pair of comparison operators separately, and each select statement can use indexes, be optimised, etc.

Sotiris Filippidis SSC Enthusiast Points: 140 More actions · Answer 4

When I was assigned this task, I first thought of a similar approach. But when you have over 10 fields to filter by this can be rather big. You also have to type a lot if / when something like a field name changes.

I created this example having in mind not only the use of operators but also the extension of this query to accept different logical operators between criteria too. Think of it as a dynamic environment where you can add filters to fields and also add operators like AND, OR between criteria. This would be too much typing if you wanted to cover all possible combinations.

Sotiris L. Filippidis

Henrik Staun Poulsen-105872 Mr or Mrs. 500 Points: 541 More actions · Answer 5

Performance might be an issue, but this solution is great when security is an issue. It is very easy to grant execution rights to a sproc, and then block the Select command.

See http://www.novicksoftware.com/Articles/crud-operations-using-sql-server-stored-procedures-part-1.htm about application roles.

Adam Machanic SSCoach Points: 15370 More actions · Answer 6

Please check out MVP Erland Sommarskog's article on this subject:

http://www.sommarskog.se/dyn-search.html

--
Adam Machanic
whoisactive

RonKyle SSC-Dedicated Points: 31554 More actions · Answer 7

We create a temp table which is then filled in with the resulting values which are then joined through a stored proc to a fixed statement.

In this case, we would populate a temp table with the company codes that meet the desired criteria. We then join that table to the query. The temp table acts as a filter.

It's much simpler than the way described in the article.

thormj Ten Centuries Points: 1217 More actions · Answer 8

I like RonKyle's solution -- its one I haven't seen before.

But why not a query like this:

select * from customers where

((@countryOP='bg' AND country like @country+'%') or
(@countryOP='eq' AND country = @country) or
(@countryOP='ne' AND country <> @country) or
(@countryOP='')) AND
((@companyOP='') or
(@companyOP='bg' AND comany like @company+'%') or
...

9/10ths of the statements would be false, but how much of a problem is that?

Adam Machanic SSCoach Points: 15370 More actions · Answer 9

thormj,

The problem is that unfortunately that query will be unable to seek indexes as the optimizer will evaluate ALL of the possible conditions (therefore, a scan will be necessary).

--
Adam Machanic
whoisactive

thormj Ten Centuries Points: 1217 More actions · Answer 10

Thank you very much; every day I find I know very little about sql (Mainly a C/C++ dude).

Hey RonKyle, could you post an example of the join that you're talking about so we can talk about how it would use the indices? It looks like that might be the cleanest solution yet.

Thanks!

Beedles About Valued Member Points: 52 More actions · Answer 11

Hi all,

Did RonKyle post a sample of his solution as I have a little task which could use his technique.

If not perhaps he could post a sample so we may see how it works and give some feedback.

Regards Steve

Gary Xu Grasshopper Points: 15 More actions · Answer 12

My first ever post. Here are my 2 cents:

Sotiris' article has some good acdemic interest. But just as laperrej pointed out, we can't use anything like that in production because of performance issues.
AKM's answer is not a bad one, for a large, busy system. I would actually push it futher: use a master stored procedure to call sub stored procedures. Yes, it would be lots of typing. But performance is always the key.
Dynamic construct a query is not a bad idea. When you use EXEC sp_executesql @CMD, the plan is most likely cached and reused. Comparing with executing a wrong plan, the time consumed to generate a plan is neglectable
If I am tasked to write such a query, I would do this:

CREATE PROC usp_GetComstomers
@CompY VARCHAR(100)='%',
@ContY VARCHAR(100)='%',
@CompN VARCHAR(100)='NoNothingNoneWhateverString',
@ContN VARCHAR(100)='NoNothingNoneWhateverString'
AS
/*
Example:
EXEC usp_GetComstomers '%','Germany','A%'
EXEC usp_GetComstomers '%the%'
*/
SET NOCOUNT ON
SELECT *
FROM dbo.Customers
WHERE CompanyName LIKE @CompY
AND Country LIKE @ContY
AND CompanyName NOT LIKE @CompN
AND Country NOT LIKE @ContN
This would be fast and not too bad looking. 🙂

Carlo Romagnano SSC-Insane Points: 22663 More actions · Answer 13

In this case, I prefer dynamic sql and add only conditions I need. Performance are granted more than complex and useless conditions. 🙂

Ben-425648 SSC Enthusiast Points: 101 More actions · Answer 14

How about:...?

set @companyname = 'A%'

set @companynameOp = 'ne'

set @country = 'Germany'

set @countryOp = 'eq'

set@companyName =

case @companyNameOp

when '' then null

when 'eq' then -- Operator is "equals"

@companyName

when 'bg' then -- Operator is "begins with"

@companyName +'%'

when 'ed' then -- Operator is "ends with"

'%' + @companyName

when 'ct' then -- Operator is "contains"

'%' + @companyName + '%'

set@country =

case @countryOp

when '' then null

when 'eq' then -- Operator is "equals"

@country

when 'bg' then -- Operator is "begins with"

@country +'%'

when 'ed' then -- Operator is "ends with"

'%' + @country

when 'ct' then -- Operator is "contains"

'%' + @country + '%'

-- Ok, now let's form our query.

select

customerid, companyname, country

from

customers

where

(companyName like @companyName or @companyName is null)

AND

(country like @country or @country is null)