Using PATINDEX to find varying length patterns in T-SQL - sql

Question

I'm looking to pull floats out of some varchars, using PATINDEX() to spot them. I know in each varchar string, I'm only interested in the first float that exists, but they might have different lengths.
e.g.
'some text 456.09 other text'
'even more text 98273.453 la la la'
I would normally match these with a regex
"[0-9]+[.][0-9]+"
However, I can't find an equivalent for the + operator, which PATINDEX accepts. So they would need to be matched (respectively) with:
'[0-9][0-9][0-9].[0-9][0-9]' and '[0-9][0-9][0-9][0-9][0-9].[0-9][0-9][0-9]'
Is there any way to match both of these example varchars with one single valid PATINDEX pattern?

Answer 1

I blogged about this a while ago.
Extracting numbers with SQL server
Declare #Temp Table(Data VarChar(100))
Insert Into #Temp Values('some text 456.09 other text')
Insert Into #Temp Values('even more text 98273.453 la la la')
Insert Into #Temp Values('There are no numbers in this one')
Select Left(
SubString(Data, PatIndex('%[0-9.-]%', Data), 8000),
PatIndex('%[^0-9.-]%', SubString(Data, PatIndex('%[0-9.-]%', Data), 8000) + 'X')-1)
From #Temp

Answer 2

Wildcards.
SELECT PATINDEX('%[0-9]%[0-9].[0-9]%[0-9]%','some text 456.09 other text')
SELECT PATINDEX('%[0-9]%[0-9].[0-9]%[0-9]%','even more text 98273.453 la la la')

Answer 3

Yes you need to link to the clr to get regex support. But if PATINDEX does not do what you need then regex was designed exactly for that.
http://msdn.microsoft.com/en-us/magazine/cc163473.aspx

Answer 4

Should be checked for robustness (what if you only have an int, for example), but this is just to put you on a track:
if exists (select routine_name from information_schema.routines where routine_name = 'GetFirstFloat')
drop function GetFirstFloat
go
create function GetFirstFloat (#string varchar(max))
returns float
as
begin
declare #float varchar(max)
declare #pos int
select #pos = patindex('%[0-9]%', #string)
select #float = ''
while isnumeric(substring(#string, #pos, 1)) = 1
begin
select #float = #float + substring(#string, #pos, 1)
select #pos = #pos + 1
end
return cast(#float as float)
end
go
select dbo.GetFirstFloat('this is a string containing pi 3.14159216 and another non float 3 followed by a new fload 5.41 and that''s it')
select dbo.GetFirstFloat('this is a string with no float')
select dbo.GetFirstFloat('this is another string with an int 3')

Answer 5

Given that the pattern is going to be varied in length, you're not going to have a rough time getting this to work with PATINDEX. There is another post that I wrote, which I've modified to accomplish what you're trying to do here. Will this work for you?
CREATE TABLE #nums (n INT)
DECLARE #i INT
SET #i = 1
WHILE #i < 8000
BEGIN
INSERT #nums VALUES(#i)
SET #i = #i + 1
END
CREATE TABLE #tmp (
id INT IDENTITY(1,1) not null,
words VARCHAR(MAX) null
)
INSERT INTO #tmp
VALUES('I''m looking for a number, regardless of length, even 23.258 long'),('Maybe even pi which roughly 3.14159265358,'),('or possibly something else that isn''t a number')
UPDATE #tmp SET words = REPLACE(words, ',',' ')
;WITH CTE AS (SELECT ROW_NUMBER() OVER (ORDER BY ID) AS rownum, ID, NULLIF(SUBSTRING(' ' + words + ' ' , n , CHARINDEX(' ' , ' ' + words + ' ' , n) - n) , '') AS word
FROM #nums, #tmp
WHERE ID <= LEN(' ' + words + ' ') AND SUBSTRING(' ' + words + ' ' , n - 1, 1) = ' '
AND CHARINDEX(' ' , ' ' + words + ' ' , n) - n > 0),
ids AS (SELECT ID, MIN(rownum) AS rownum FROM CTE WHERE ISNUMERIC(word) = 1 GROUP BY id)
SELECT CTE.rownum, cte.id, cte.word
FROM CTE, ids WHERE cte.id = ids.id AND cte.rownum = ids.rownum
The explanation and origin of the code is covered in more detail in the origional post

Answer 6

PATINDEX is not powerful enough to do that. You should use regular expressions.
SQL Server has Regular expression support since SQL Server 2005.