How to replace anything between 2 specific characters in SQL Server - sql

Question

I'm trying to replace anything between 2 specific characters in a string that contains multiples of those 2 caracters. Take it as a csv format.
Here an example of what i got as data in that field:
0001, ABCD1234;0002, EFGH432562;0003, IJKL1345hsth;...
What I need to retreive from it is all parts before the ',' but not what are between ',' and ';'
I tried with those formula but no success
SELECT REPLACE(fieldname, ',[A-Z];', ' ') FROM ...
or
SELECT REPLACE(fieldname, ',*;', ' ') FROM ...
I need to get
0001 0002 0003
Is there a way to achieve that?

Answer 1

You can CROSS APPLY to a STRING_SPLIT that uses STRING_AGG (since Sql Server 2017) to stick the numbers back together.
select id, codes
from your_table
cross apply (
select string_agg(left(value, patindex('%_,%', value)), ' ') as codes
from string_split(fieldname, ';') s
where value like '%_,%'
) ca;
GO
id
codes
1
0001 0002 0003
Demo on db<>fiddle here
Extra
Here is a version that also works in Sql Server 2014.
Inspired by the research from #AaronBertrand
The UDF uses a recursive CTE to split the string.
And the FOR XML trick is used to stick the numbers back together.
CREATE FUNCTION dbo.fnString_Split
(
#str nvarchar(4000),
#delim nchar(1)
)
RETURNS TABLE
WITH SCHEMABINDING
AS
RETURN
(
WITH RCTE AS (
SELECT
1 AS ordinal
, ISNULL(NULLIF(CHARINDEX(#delim, #str),0), LEN(#str)) AS pos
, LEFT(#str, ISNULL(NULLIF(CHARINDEX(#delim, #str),0)-1, LEN(#str))) AS value
UNION ALL
SELECT
ordinal+1
, ISNULL(NULLIF(CHARINDEX(#delim, #str, pos+1), 0), LEN(#str))
, SUBSTRING(#str, pos+1, ISNULL(NULLIF(CHARINDEX(#delim, #str, pos+1),0)-pos-1, LEN(#str)-pos ))
FROM RCTE
WHERE pos < LEN(#str)
)
SELECT ordinal, value
FROM RCTE
);
SELECT id, codes
FROM your_table
CROSS APPLY (
SELECT RTRIM((
SELECT LEFT(value, PATINDEX('%_,%', value))+' '
FROM dbo.fnString_Split(fieldname, ';') AS spl
WHERE value LIKE '%_,%'
ORDER BY ordinal
FOR XML PATH(''), TYPE).value(N'./text()[1]', N'nvarchar(max)')
) AS codes
) ca
OPTION (MAXRECURSION 250);
id
codes
1
0001 0002 0003
Demo on db<>fiddle here
Alternative version of the UDF (no recursion)
CREATE FUNCTION dbo.fnString_Split
(
#str NVARCHAR(4000),
#delim NCHAR(1)
)
RETURNS #tbl TABLE (ordinal INT, value NVARCHAR(4000))
WITH SCHEMABINDING
AS
BEGIN
DECLARE #value NVARCHAR(4000)
, #pos INT = 0
, #ordinal INT = 0;
WHILE (LEN(#str) > 0)
BEGIN
SET #ordinal += 1;
SET #pos = ISNULL(NULLIF(CHARINDEX(#delim, #str),0), LEN(#str)+1);
SET #value = LEFT(#str, #pos-1);
SET #str = SUBSTRING(#str, #pos+1, LEN(#str));
INSERT INTO #tbl (ordinal, value)
VALUES (#ordinal, #value);
END;
RETURN;
END;

Answer 2

If you're on SQL Server 2017 and don't need a guarantee that the order will be maintained, then LukStorms' answer is perfectly adequate.
However, if you:
care about an order guarantee; or,
are on an older version than 2017 (and can't use STRING_AGG); or,
are on an even older version than 2016 or are in an older compatibility level (and can't use STRING_SPLIT):
Here's an ordered split function that can help (it's long and ugly but you only have to create it once):
CREATE FUNCTION dbo.SplitOrdered
(
#list nvarchar(max),
#delim nvarchar(10)
)
RETURNS TABLE
WITH SCHEMABINDING
AS
RETURN
(
WITH w(n) AS (SELECT 0 FROM (VALUES (0),(0),(0),(0)) w(n)),
k(n) AS (SELECT 0 FROM w a, w b),
r(n) AS (SELECT 0 FROM k a, k b, k c, k d, k e, k f, k g, k h),
p(n) AS (SELECT TOP (COALESCE(LEN(#list), 0))
ROW_NUMBER() OVER (ORDER BY ##SPID) -1 FROM r),
spots(p) AS
(
SELECT n FROM p
WHERE (SUBSTRING(#list, n, LEN(#delim + 'x') - 1) LIKE #delim OR n = 0)
),
parts(p,val) AS
(
SELECT p, SUBSTRING(#list, p + LEN(#delim + 'x') - 1,
LEAD(p, 1, 2147483647) OVER (ORDER BY p) - p - LEN(#delim))
FROM spots AS s
)
SELECT listpos = ROW_NUMBER() OVER (ORDER BY p),
Item = LTRIM(RTRIM(val))
FROM parts
);
Then the query can become:
;WITH x AS
(
SELECT id, listpos,
codes = LEFT(Item, COALESCE(NULLIF(CHARINDEX(',', Item),0),1)-1)
FROM dbo.your_table
CROSS APPLY dbo.SplitOrdered(fieldname, ';') AS c
)
SELECT id, codes = (
(SELECT x2.codes + ' '
FROM x AS x2
WHERE x2.id = x.id
ORDER BY x2.listpos
FOR XML PATH(''), TYPE).value(N'./text()[1]', N'nvarchar(max)')
)
FROM x GROUP BY id;
Example borrowing from LukStorms' db<>fiddle
Note that, in addition to guaranteeing order and being backward compatible (well, only back so many versions), it also ignores garbage data, e.g. try:
0001, ABCD1234;0002 but no comma