using regular expression in sql - sql

Question

I have the following rows in my table
COL1 EXTRA DOUBLE TEST
12 TEST
123 EXTRA
125 EXTRA 95 DOUBLE
EXTRA 45 99 DOUBLE
I am using regular expressions to filter out the rows and move them appropriately to different columns. So:
For the first row, I want 12 to be extracted and put in column TEST.
For the second row, I want 123 to be extracted and put in column EXTRA.
For 3rd row, I want 125 to be extracted and put in column EXTRA.
I want to ignore 95.
For the last row, I want 45 to be extracted and put in column EXTRA.
I can extract the values and put them in appropriate columns through my query, I am using this regular expression for extracting the values:
'%[0-9]%[^A-Z]%[0-9]%'
the problem with this regular expression is that it extracts 12, but it does not extract 123 from the second row, if I change the regular expression to:
'%[0-9]*%[^A-Z]%[0-9]%'
then it extracts 123, but for the third row, it concatenates 125 with 95 so I get 12595. Is there any way I can avoid 95 and just get the value 125? If I remove the star then it does not do any concatenation.
Any help will be appreciated. I posted this question before, but some of you were asking for more explanation so I posted a new question for that.

Answer 1

I believe that the regex that you are looking for is below. This will match digits followed by numbers, followed by ignoring any future number patterns. However, I believe that when you use the %regex%regex%..., that it runs each regex separately, so I am not sure about the nuances of regex in SQL. However, if you run this against rubular.com it seems to solve the problem you are asking. Hopefully it can be of some use in your regex search :)
([0-9]*)([^A-Z])(?>[0-9]*)
However, I did just look at your other examples of the letters coming first, and that would not work here. But, maybe this can still be of use to you

Answer 2

SQL Server does not natively support Regex. It does support some limited pattern matching through Like and Patindex.
If you really want to use Regex inside of SQL Server you can use a .NET language like C# to create a special CLR and import that into SQL Server, but that comes with a number of drawbacks. If you want to use Regex the better way is to have an application that runs on top of SQL Server. That application could be written in any language that can interface ODBC like C# or Python, and in fact in an intro article I talk about interfacing Python with SQL Server to use regex on Simple-Talk.
But, the patterns you provide are using SQL Servers more limited pattern matching capabilities rather than Regex, so that seems to be what you want. There is a full description at Pattern Matching in Search Conditions
As for solving your particular problem, you don't seem to have one particular pattern but several possible patterns anway. That type of situation is almost impossible to handle with a single SQL Server pattern and the regex logic gets unnecessarily complicated too. So, if I were in your position I would not try to create a single pattern but a series of cases and then extract the number you need based on that.

Answer 3

Assuming this is SQL 2005 (or later I guess... I can only speak for 2005), and all different permutations of COL1 data are in your question:
UPDATE NameOfYourTable
SET TEST = SUBSTRING(Col1, 0, LEN(Col1) - (LEN(Col1) - PATINDEX('%[0-9] TEST%', Col1) - 1))
WHERE COL1 LIKE '%[0-9] TEST%'
UPDATE NameOfYourTable
SET EXTRA = SUBSTRING(Col1, 0, LEN(Col1) - (LEN(Col1) - PATINDEX('%[0-9] EXTRA%', Col1) - 1))
WHERE COL1 LIKE '%[0-9] EXTRA%'
UPDATE NameOfYourTable
SET EXTRA = SUBSTRING(Col1, PATINDEX('%[0-9]%', Col1), LEN(Col1) - (LEN(Col1) - PATINDEX('%[0-9] [0-9]%', Col1) + LEN('EXTRA ')))'
WHERE COL1 LIKE 'EXTRA [0-9]%'
Somehow though, I really don't think this is going to resolve your problem. I would strongly advise you to make sure this will catch all the cases you need to handle by running this on some test data.
If you have a lot of different cases to handle, then the better alternative I think would be to make a small console program in something like C# (that has much better RegExp support) to sift through your data and apply updates that way. Trying to handle numerous permutations of COL1 data is going to be a nightmare in SQL.
Also read these on LIKE, PATINDEX and their (limited) pattern-matching abilities:
LIKE: http://msdn.microsoft.com/en-us/library/ms179859(v=sql.90).aspx
PATINDEX: http://msdn.microsoft.com/en-us/library/ms188395(v=sql.90).aspx