patindex parser

I am trying to parse a string column matching it against several patterns. I have code that does not produce the desired results.

This is a sample table:

-- demo data
IF OBJECT_ID('dbo.Input', 'U') IS NOT NULL DROP TABLE dbo.Input 
GO
CREATE TABLE [dbo].[Input]
(
    [PromotionCode] [char](10) NOT NULL,
    [Description] [char](30) NULL
 CONSTRAINT [PK_Input] PRIMARY KEY CLUSTERED (PromotionCode)
)
GO

INSERT INTO dbo.Input (PromotionCode, Description) Values ('1709106-01', 'Good Gums 10% /M100/'), ('1708162-01', 'PURE Santevia 6% /M33/E33/R33/'), ( '1709164-01', 'Sovereign 10% /M50/E50/'),( '6031709-04', 'VS JF Powder 240g /M7/R3/E10/')

And this is my query:

-- parse description using patindex
select PromotionCode, [Description], 
    --patindex('%/M[0-9][0-9]/%', [Description]) as 'WWW',
    --substring([Description], patindex( '%/M[0-9][0-9]/%', [Description]) + 2, 2) as 'XXX',
    --parse(substring([Description], patindex( '%/M[0-9][0-9]/%', [Description]) + 2, 2) as float using 'en-US') as 'ZZZ',
    CASE 
    WHEN patindex('%/M[0-9][0-9][0-9]/%', [Description]) = 0 then 0
    WHEN patindex('%/M[0-9][0-9][0-9]/%', [Description]) > 0 then parse(substring([Description], patindex( '%/M[0-9][0-9][0-9]/%', [Description]) + 2, 3) as float using 'en-US') 
    ELSE (
        CASE
        WHEN patindex('%/M[0-9][0-9]/%', [Description]) = 0 then 0
        WHEN patindex('%/M[0-9][0-9]/%', [Description]) > 0 then parse(substring([Description], patindex( '%/M[0-9][0-9]/%', [Description]) + 2, 2) as float using 'en-US') 
        ELSE (
            CASE
            WHEN patindex('%/M[0-9]/%', [Description]) = 0 then 0
            WHEN patindex('%/M[0-9]/%', [Description]) > 0 then parse(substring([Description], patindex( '%/M[0-9]/%', [Description]) + 2, 1) as float using 'en-US') 
            ELSE NULL
            END)
        END)
    END [MCB]
from dbo.Input

It is looking for an Mxxx token where xxx is a number, and is supposed to extract and retur

Your code does not work because of how the conditions are arranged in your nested CASEs. Take, for example, the first Description value in your output, PURE Santevia 6% /M33/E33/R33/. When passed through your set of conditions, it will match the very first one:

WHEN patindex('%/M[0-9][0-9][0-9]/%', [Description]) = 0

and, consequently produce the result of 0, even though it also matches this condition further down:

WHEN patindex('%/M[0-9][0-9]/%', [Description]) > 0

The easiest fix is to remove all the branches that check for equality to 0, because they are just superfluous:

CASE 
WHEN patindex('%/M[0-9][0-9][0-9]/%', [Description]) > 0 then parse(substring([Description], patindex( '%/M[0-9][0-9][0-9]/%', [Description]) + 2, 3) as float using 'en-US') 
ELSE (
    CASE
    WHEN patindex('%/M[0-9][0-9]/%', [Description]) > 0 then parse(substring([Description], patindex( '%/M[0-9][0-9]/%', [Description]) + 2, 2) as float using 'en-US') 
    ELSE (
        CASE
        WHEN patindex('%/M[0-9]/%', [Description]) > 0 then parse(substring([Description], patindex( '%/M[0-9]/%', [Description]) + 2, 1) as float using 'en-US') 
        ELSE NULL
        END)
    END)
END [MCB]

The nesting, however, is superfluous too. If one pattern is not matched, SQL Server will continue testing the next one, until it reaches the ELSE clause. Therefore, your CASE expression could look simply like this:

CASE 
    WHEN patindex('%/M[0-9][0-9][0-9]/%', [Description]) > 0 then parse(substring([Description], patindex( '%/M[0-9][0-9][0-9]/%', [Description]) + 2, 3) as float using 'en-US') 
    WHEN patindex('%/M[0-9][0-9]/%', [Description]) > 0 then parse(substring([Description], patindex( '%/M[0-9][0-9]/%', [Description]) + 2, 2) as float using 'en-US') 
    WHEN patindex('%/M[0-9]/%', [Description]) > 0 then parse(substring([Description], patindex( '%/M[0-9]/%', [Description]) + 2, 1) as float using 'en-US') 
    ELSE NULL  -- or ELSE 0, depending on the desired result when there is no match
END [MCB]

It is also possible to avoid having the nearly duplicate code in the CASE's multiple branches and have a single expression instead. This is one way of doing that:

SELECT
  PromotionCode,
  Description,
  MCB = CASE
          WHEN Description LIKE '%/M[0-9][0-9][0-9]/%'
            OR Description LIKE '%/M[0-9][0-9]/%'
            OR Description LIKE '%/M[0-9]/%'
          THEN SUBSTRING(Description, p1.Pos, p2.Pos - p1.Pos)
          ELSE NULL  -- or ELSE 0
        END
FROM
  dbo.Input
  CROSS APPLY (SELECT PATINDEX('%/M[0-9]%/%', Description) + 2) AS p1 (Pos)
  CROSS APPLY (SELECT CHARINDEX('/', Description, p1.Pos)) AS p2 (Pos)
;

The values p1.Pos and p2.Pos are calculated for every row but used only when the description matches any of the patterns specified. The p1.Pos is the position two characters after the position of the M in the Mxxx token, and p2.Pos is the first slash encountered starting from p1.Pos – in other words, the slash immediately after the token.

In case the description does not have an Mxxx token, the values of p1.Pos and p2.Pos will make no sense. However, as already explained, they will not be used in that case but the ELSE clause will work instead.

StackExchange Database Administrators Q#188271, answer score: 4