SQL to split a non delimited string into delimited string

• mtassin

  SSC-Insane

  Points: 23099

  More actions

  February 8, 2011 at 9:26 am

  #1283801

  Jeff Moden (2/7/2011)

  ---

  Wolfmeister (2/7/2011)

  ---

  That's pretty slick! Thanks guys.

  Just making sure... do you understand how ColdCoffee's method works?

  Heck I didn't understand what he was doing with the CTEs till I realized he was just making a virtual tally table with them. I saw Tally in the query and figured he had a tally table he was assuming. 🙂

  ---

  
  --Mark Tassin
  MCITP - SQL Server DBA
  Proud member of the Anti-RBAR alliance.
  For help with Performance click this link[/url]
  For tips on how to post your problems[/url]

• ColdCoffee

  SSC-Dedicated

  Points: 39972

  More actions

  February 8, 2011 at 6:45 pm

  #1284145

  mtassin (2/8/2011)

  ---

  Jeff Moden (2/7/2011)

  ---

  Wolfmeister (2/7/2011)

  ---

  That's pretty slick! Thanks guys.

  Just making sure... do you understand how ColdCoffee's method works?

  Heck I didn't understand what he was doing with the CTEs till I realized he was just making a virtual tally table with them. I saw Tally in the query and figured he had a tally table he was assuming. 🙂

  I was not sure if the OP knows about Tally tables, so i thought i will just make a On-The-Fly Tally table.. The Exec Plan will become much better if i were using a traditional persistent Tally Table.

• Jeff Moden

  SSC Guru

  Points: 1004414

  More actions

  February 8, 2011 at 9:22 pm

  #1284185

  Wolfmeister (2/8/2011)

  ---

  SwePeso (2/8/2011)

  ---

  DECLARE @Sample VARCHAR(100) = '12345'

  ;WITH cte(CurrPos, LastPos, String)

  AS (

  SELECT1 AS CurrPos,

  DATALENGTH(@Sample) AS LastPos,

  CAST(@Sample AS VARCHAR(MAX)) AS String

  UNION ALL

  SELECTCurrPos + 1 AS CurrPos,

  LastPos,

  STUFF(String, 2 * CurrPos, 0, ',') AS String

  FROMcte

  WHERECurrPos < LastPos

  )

  SELECTString

  FROMcte

  WHERECurrPos = LastPos

  Thanks, SwePeso, that's pretty neat as well:-)

  Oh... be careful, now. Recursion in the form of a Recursive CTE can be as bad as a loop. Let's see what happens.

  First, I've converted ColdCoffee's method and Peso's method to an iTVF (inline Table Value Function) each.

  CREATE FUNCTION dbo.ColdCoffee

  (@string VARCHAR(100))

  RETURNS TABLE AS

  RETURN

  WITH

  TENS (N) AS (SELECT 0 UNION ALL SELECT 0 UNION ALL SELECT 0 UNION ALL

  SELECT 0 UNION ALL SELECT 0 UNION ALL SELECT 0 UNION ALL

  SELECT 0 UNION ALL SELECT 0 UNION ALL SELECT 0 UNION ALL SELECT 0),

  THOUSANDS (N) AS (SELECT 1 FROM TENS t1 CROSS JOIN TENS t2 CROSS JOIN TENS t3),

  TALLY (N) AS (SELECT ROW_NUMBER() OVER (ORDER BY (SELECT 0)) FROM THOUSANDS)

  select STUFF(

  (SELECT ','+SUBSTRING(@string,N,1)

  FROM TALLY T_Outer

  WHERE N <= LEN(@string)

  FOR XML PATH('') ),1 ,1 ,'') Concat_Values

  ;

  GO

  CREATE FUNCTION dbo.Peso

  (@Sample VARCHAR(100))

  RETURNS TABLE AS

  RETURN

  WITH cte(CurrPos, LastPos, String)

  AS (

  SELECT 1 AS CurrPos,

  DATALENGTH(@Sample) AS LastPos,

  CAST(@Sample AS VARCHAR(MAX)) AS String

  UNION ALL

  SELECT CurrPos + 1 AS CurrPos,

  LastPos,

  STUFF(String, 2 * CurrPos, 0, ',') AS String

  FROM cte

  WHERE CurrPos < LastPos

  )

  SELECT String

  FROM cte

  WHERE CurrPos = LastPos

  ;

  GO

  Of course, we'll need some test data. Since both functions can handle just about any string, I used all GUID's for the test with only 10,000 rows.

  SELECT TOP 10000

  SomeString = CAST(NEWID() AS VARCHAR(36))

  INTO dbo.JBMTest

  FROM sys.all_columns ac1

  CROSS JOIN sys.all_columns ac2

  To take the display processing time out of the picture, I simply assign the results to a "throw-away" variable in the following test code.

  PRINT '========== ColdCoffee =========='

  SET STATISTICS TIME ON;

  DECLARE @Bitbucket VARCHAR(100);

  SELECT @Bitbucket = ca.Concat_Values

  FROM dbo.JBMTest test

  CROSS APPLY dbo.ColdCoffee(test.SomeString) ca;

  SET STATISTICS TIME OFF;

  GO

  PRINT '========== Peso =========='

  SET STATISTICS TIME ON;

  DECLARE @Bitbucket VARCHAR(100)

  SELECT @Bitbucket = ca.String

  FROM dbo.JBMTest test

  CROSS APPLY dbo.Peso(test.SomeString) ca;

  SET STATISTICS TIME OFF;

  GO

  Here are the results from my 8 year old desktop box...

  ========== ColdCoffee ==========

  SQL Server Execution Times:

  CPU time = 844 ms, elapsed time = 899 ms.

  ========== Peso ==========

  SQL Server Execution Times:

  CPU time = 47469 ms, elapsed time = 57220 ms.

  Like I said, you may want to avoid the "Hidden RBAR" of Recursive CTE's.

  --Jeff Moden

  ---

  RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
  First step towards the paradigm shift of writing Set Based code:
  ________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.

  Change is inevitable... Change for the better is not.
  

  ---

  Helpful Links:
  How to post code problems
  How to Post Performance Problems
  Create a Tally Function (fnTally)
  

• ColdCoffee

  SSC-Dedicated

  Points: 39972

  More actions

  February 8, 2011 at 9:50 pm

  #1284192

  Results from my machine (i am not sure of my desktop's age :w00t: , but i can tell its faster than urs :-P)

  ========== ColdCoffee ==========

  SQL Server Execution Times:

  CPU time = 390 ms, elapsed time = 398 ms.

  ========== Peso ==========

  SQL Server Execution Times:

  CPU time = 16255 ms, elapsed time = 16889 ms.

• Lynn Pettis

  SSC Guru

  Points: 442467

  More actions

  February 8, 2011 at 10:04 pm

  #1284197

  ColdCoffee (2/8/2011)

  ---

  mtassin (2/8/2011)

  ---

  Jeff Moden (2/7/2011)

  ---

  Wolfmeister (2/7/2011)

  ---

  That's pretty slick! Thanks guys.

  Just making sure... do you understand how ColdCoffee's method works?

  Heck I didn't understand what he was doing with the CTEs till I realized he was just making a virtual tally table with them. I saw Tally in the query and figured he had a tally table he was assuming. 🙂

  I was not sure if the OP knows about Tally tables, so i thought i will just make a On-The-Fly Tally table.. The Exec Plan will become much better if i were using a traditional persistent Tally Table.

  Careful here, there is a tipping point when accessing a tally table from disk. It all depends on how big the table needs to be to support its use. If you need an extremely large tally table (> 1,000,000 for example, could be larger), it is possible that a dynamic tally table may actually be better. The tipping point is system dependent, but it does exist. At a previous employer I did observe this during some testing. I wish I could find the output from those tests, but not sure if I sent the info home.

• ColdCoffee

  SSC-Dedicated

  Points: 39972

  More actions

  February 8, 2011 at 10:34 pm

  #1284207

  Lynn Pettis (2/8/2011)

  ---

  ColdCoffee (2/8/2011)

  ---

  mtassin (2/8/2011)

  ---

  Jeff Moden (2/7/2011)

  ---

  Wolfmeister (2/7/2011)

  ---

  That's pretty slick! Thanks guys.

  Just making sure... do you understand how ColdCoffee's method works?

  Heck I didn't understand what he was doing with the CTEs till I realized he was just making a virtual tally table with them. I saw Tally in the query and figured he had a tally table he was assuming. 🙂

  I was not sure if the OP knows about Tally tables, so i thought i will just make a On-The-Fly Tally table.. The Exec Plan will become much better if i were using a traditional persistent Tally Table.

  Careful here, there is a tipping point when accessing a tally table from disk. It all depends on how big the table needs to be to support its use. If you need an extremely large tally table (> 1,000,000 for example, could be larger), it is possible that a dynamic tally table may actually be better. The tipping point is system dependent, but it does exist. At a previous employer I did observe this during some testing. I wish I could find the output from those tests, but not sure if I sent the info home.

  Absolutely agree here, Lynn.. Even i have had a query which performed better with dynamic tally than the traditional tally. So again, usage of type of Tally table "depends" on the context 🙂

• Wolfmeister

  SSCommitted

  Points: 1547

  More actions

  February 9, 2011 at 6:53 am

  #1284424

  Thanks all for your inputs. Much appreciated 🙂

• Jeff Moden

  SSC Guru

  Points: 1004414

  More actions

  February 9, 2011 at 7:00 am

  #1284427

  Lynn Pettis (2/8/2011)

  ---

  Careful here, there is a tipping point when accessing a tally table from disk. It all depends on how big the table needs to be to support its use. If you need an extremely large tally table (> 1,000,000 for example, could be larger), it is possible that a dynamic tally table may actually be better. The tipping point is system dependent, but it does exist. At a previous employer I did observe this during some testing. I wish I could find the output from those tests, but not sure if I sent the info home.

  Have you found that happens anywhere other than in a "splitter"?

  --Jeff Moden

  ---

  RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
  First step towards the paradigm shift of writing Set Based code:
  ________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.

  Change is inevitable... Change for the better is not.
  

  ---

  Helpful Links:
  How to post code problems
  How to Post Performance Problems
  Create a Tally Function (fnTally)
  

• Lynn Pettis

  SSC Guru

  Points: 442467

  More actions

  February 9, 2011 at 10:07 am

  #1284619

  @jeff - I haven't been able to find anything as of yet, so some of this has to be taken as personal experience. What this means is that people need to test in their own environments and keep this in mind should performance problems begin to occur in the future.

• Jeff Moden

  SSC Guru

  Points: 1004414

  More actions

  February 9, 2011 at 5:26 pm

  #1284909

  Lynn Pettis (2/9/2011)

  ---

  @Jeff - I haven't been able to find anything as of yet, so some of this has to be taken as personal experience. What this means is that people need to test in their own environments and keep this in mind should performance problems begin to occur in the future.

  Heh... that's how myths are sometimes started. I'd rather see code that proves it. 😉

  --Jeff Moden

  ---

  RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
  First step towards the paradigm shift of writing Set Based code:
  ________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.

  Change is inevitable... Change for the better is not.
  

  ---

  Helpful Links:
  How to post code problems
  How to Post Performance Problems
  Create a Tally Function (fnTally)
  

• Lynn Pettis

  SSC Guru

  Points: 442467

  More actions

  February 9, 2011 at 6:21 pm

  #1284916

  Well, it always better to have more than one person testing.

  The testing I did used actual data at work, so I may not be able reproduce it right away. Plus, have some other things on my plate right now that are a bit more critical.