Sql server 在基于集合的SQL中从VARCHAR字段提取段/值的最佳方法

以以下数据为例：

SELECT 'HelpDesk Call Reference F0012345, Call Update, 40111' AS [Subject]
UNION ALL
SELECT 'HelpDesk Call Reference F0012346, Call Resolved, 40112' AS [Subject]
UNION ALL
SELECT 'HelpDesk Call Reference F0012347, New call logged, 40113' AS [Subject]

我想做的是提取以下数据：

WITH    ExampleData
          AS ( SELECT   'HelpDesk Call Reference F0012345, Call Update, 40111' AS [Subject]
               UNION ALL
               SELECT   'HelpDesk Call Reference F0012346, Call Resolved, 40112'
               UNION ALL
               SELECT   'HelpDesk Call Reference F0012347, New call logged, 40113'
             )
    SELECT  dbo.fnParseString(2, ',', REPLACE([Subject], 'HelpDesk Call Reference ', 'HelpDesk Call Reference, ')) AS [Ref] ,
            dbo.fnParseString(3, ',', REPLACE([Subject], 'HelpDesk Call Reference ', 'HelpDesk Call Reference, ')) AS [Type] ,
            dbo.fnParseString(4, ',', REPLACE([Subject], 'HelpDesk Call Reference ', 'HelpDesk Call Reference, ')) AS [OurRef]
    FROM    ExampleData

如您所见，我需要将Ref、Type和OurRef提取为单独的列，以确保在处理生成的电子邮件时高效地使用基于集合的SQL

通常在这种情况下，我会使用如下函数：

CREATE FUNCTION dbo.fnParseString (
    @Section SMALLINT ,
    @Delimiter CHAR ,
    @Text VARCHAR(MAX)
)
RETURNS VARCHAR(8000)
AS 
    BEGIN
        DECLARE @NextPos SMALLINT;
        DECLARE @LastPos SMALLINT;
        DECLARE @Found SMALLINT;

        SELECT  @NextPos = CHARINDEX(@Delimiter, @Text, 1) ,
                @LastPos = 0 ,
                @Found = 1

        WHILE @NextPos > 0
            AND ABS(@Section) <> @Found 
            SELECT  @LastPos = @NextPos ,
                    @NextPos = CHARINDEX(@Delimiter, @Text, @NextPos + 1) ,
                    @Found = @Found + 1

        RETURN LTRIM(RTRIM(CASE
            WHEN @Found <> ABS(@Section) OR @Section = 0 THEN NULL
            WHEN @Section > 0 THEN SUBSTRING(@Text, @LastPos + 1, CASE WHEN @NextPos = 0 THEN DATALENGTH(@Text) - @LastPos ELSE @NextPos - @LastPos - 1 END)
            ELSE SUBSTRING(@Text, @LastPos + 1, CASE WHEN @NextPos = 0 THEN DATALENGTH(@Text) - @LastPos ELSE @NextPos - @LastPos - 1 END)
        END))
    END

正如您所看到的，我有一个解决方案，可以得到我想要的最终结果，但是使用凌乱的udf并不理想&我想知道是否有更好的方法来处理这样的事情-也许是内联正则表达式？也就是说，我认为

PATINDEX（）

接受正则表达式作为搜索字符串-这与

SUBSTRING（）

结合可以满足我的需要，但我真的不知道从哪里开始

编辑：请注意，这是一个简化的示例，主题是可变的，我还将采用相同的技术来解析主体，主体将有8项数据，我需要使用各种分隔符来解析，因此这排除了使用

ParseName（）

，因为它只允许4个部分，我不能使用固定长度（即

substring（）

），因为长度会非常不同（特别是如果涉及到不同的帮助台（它们是），这就是为什么我会按照

PATINDEX（）

&

substring（）

的思路思考的原因。我建议使用这个：

;WITH CTE
AS
(
SELECT 'HelpDesk Call Reference F0012345, Call Update, 40111' AS [Subject]
UNION ALL
SELECT 'HelpDesk Call Reference F0012346, Call Resolved, 40112' AS [Subject]
UNION ALL
SELECT 'HelpDesk Call Reference F0012347, New call logged, 40113' AS [Subject]
)
, CTEPart
as
(
SELECT [Subject], REPLACE(SUBSTRING([Subject], 25, 1000), ', ', '.') Part
FROM CTE
)
SELECT
    [Subject],
    PARSENAME(Part, 1) AS [Ref],
    PARSENAME(Part, 2) AS [Type],
    PARSENAME(Part, 3) AS [OurRef]
FROM CTEPart

我建议使用以下方法：

;WITH CTE
AS
(
SELECT 'HelpDesk Call Reference F0012345, Call Update, 40111' AS [Subject]
UNION ALL
SELECT 'HelpDesk Call Reference F0012346, Call Resolved, 40112' AS [Subject]
UNION ALL
SELECT 'HelpDesk Call Reference F0012347, New call logged, 40113' AS [Subject]
)
, CTEPart
as
(
SELECT [Subject], REPLACE(SUBSTRING([Subject], 25, 1000), ', ', '.') Part
FROM CTE
)
SELECT
    [Subject],
    PARSENAME(Part, 1) AS [Ref],
    PARSENAME(Part, 2) AS [Type],
    PARSENAME(Part, 3) AS [OurRef]
FROM CTEPart

---

此示例为Oracle查询。使用的所有函数均为ANSI SQL标准，可在任何SQL中使用。此示例仅剪切字符串的REF部分。您只需对Type、OutRef等重复所有步骤即可。此示例假设您的REF始终包含0-0，并且REF之后始终有“，”可以用WhitePac替换可以使用NVL（）cna:INSTR（str，NVL（'，'，''）。我认为这种方法比将值硬编码到SUBSTR中更通用…：

SELECT str, SUBSTR(str, ref_start_pos, ref_end_pos) final_ref
 FROM
 (
  SELECT str, ref_start_pos, INSTR(str, ',', ref_start_pos)-ref_start_pos AS ref_end_pos
    FROM
    (
     SELECT str, INSTR(str, '0')-1 AS ref_start_pos
       FROM
       (
        SELECT 'HelpDesk Call Reference F0012345, Call Update, 40111' AS str
          FROM dual
        UNION ALL
        SELECT 'HelpDesk Call Reference F0012346, Call Resolved, 40112' 
          FROM dual
       )
     )
   )
  /

  SQL>

  STR                                                    |  FINAL_REF
  ------------------------------------------------------------------------
  HelpDesk Call Reference F0012345, Call Update, 40111   |  F0012345
  HelpDesk Call Reference F0012346, Call Resolved, 40112 |  F0012346

SQL Server版本（由OP添加）：

---

此示例为Oracle查询。使用的所有函数均为ANSI SQL标准，可在任何SQL中使用。此示例仅剪切字符串的REF部分。您只需对Type、OutRef等重复所有步骤即可。此示例假设您的REF始终包含0-0，并且REF之后始终有“，”可以用WhitePac替换可以使用NVL（）cna:INSTR（str，NVL（'，'，''）。我认为这种方法比将值硬编码到SUBSTR中更通用…：

SELECT str, SUBSTR(str, ref_start_pos, ref_end_pos) final_ref
 FROM
 (
  SELECT str, ref_start_pos, INSTR(str, ',', ref_start_pos)-ref_start_pos AS ref_end_pos
    FROM
    (
     SELECT str, INSTR(str, '0')-1 AS ref_start_pos
       FROM
       (
        SELECT 'HelpDesk Call Reference F0012345, Call Update, 40111' AS str
          FROM dual
        UNION ALL
        SELECT 'HelpDesk Call Reference F0012346, Call Resolved, 40112' 
          FROM dual
       )
     )
   )
  /

  SQL>

  STR                                                    |  FINAL_REF
  ------------------------------------------------------------------------
  HelpDesk Call Reference F0012345, Call Update, 40111   |  F0012345
  HelpDesk Call Reference F0012346, Call Resolved, 40112 |  F0012346

SQL Server版本（由OP添加）：

---

在做了额外的工作之后，我们决定不使用Art的答案中的方法（即使它有效）

我们需要一种更强大的方法来验证和提取子字符串，因此我通过CLR路由使用正则表达式（感谢您为我指明了正确的方向）

我采取的方法如下：

WITH    ExampleData
          AS ( SELECT   'HelpDesk Call Reference F0012345, Call Update, 40111' AS [Subject]
               UNION ALL
               SELECT   'HelpDesk Call Reference F0012346, Call Resolved, 40112'
               UNION ALL
               SELECT   'HelpDesk Call Reference F0012347, New call logged, 40113'
             )
    SELECT  dbo.fnParseString(2, ',', REPLACE([Subject], 'HelpDesk Call Reference ', 'HelpDesk Call Reference, ')) AS [Ref] ,
            dbo.fnParseString(3, ',', REPLACE([Subject], 'HelpDesk Call Reference ', 'HelpDesk Call Reference, ')) AS [Type] ,
            dbo.fnParseString(4, ',', REPLACE([Subject], 'HelpDesk Call Reference ', 'HelpDesk Call Reference, ')) AS [OurRef]
    FROM    ExampleData

首先，我编译了以下CLR：（从C#示例转换为VB）

然后我编写了以下包装函数以简化使用：

EXEC sp_configure 
    'clr enabled' ,
    '1'

GO

RECONFIGURE
USE [db_Utility]

GO
CREATE ASSEMBLY SQL_CLR_RegExp FROM 'D:\Program Files\Microsoft SQL Server\MSSQL10_50.MSSQLSERVER\MSSQL\Binn\SQL_CLR_RegExp.dll' WITH
PERMISSION_SET = SAFE

GO
-- =============================================
-- Returns 1 or 0 if input matches pattern
-- VB function: RegexMatch(ByVal input As SqlChars, ByVal pattern As SqlString) As SqlBoolean
-- =============================================
CREATE FUNCTION [dbo].[RegexMatch]
    (
      @input [nvarchar](MAX) ,
      @pattern [nvarchar](MAX)
    )
RETURNS [bit]
    WITH EXECUTE AS CALLER
AS EXTERNAL NAME 
    [SQL_CLR_RegExp].[SQL_CLR_RegExp.UserDefinedFunctions].[RegexMatch]
GO

-- =============================================
-- Returns a comma separated string of found objects
-- VB function: RegexReplace(ByVal expression As SqlString, ByVal pattern As SqlString, ByVal replace As SqlString) As SqlString
-- =============================================
CREATE FUNCTION [dbo].[RegexReplace]
    (
      @expression [nvarchar](MAX) ,
      @pattern [nvarchar](MAX) ,
      @replace [nvarchar](MAX)
    )
RETURNS [nvarchar](MAX)
    WITH EXECUTE AS CALLER
AS EXTERNAL NAME 
    [SQL_CLR_RegExp].[SQL_CLR_RegExp.UserDefinedFunctions].[RegexReplace]
GO
-- =============================================
-- Returns a comma separated string of found objects
-- VB function: RegexSelectAll(ByVal input As SqlChars, ByVal pattern As SqlString, ByVal matchDelimiter As SqlString) As SqlString
-- =============================================
CREATE FUNCTION [dbo].[RegexSelectAll]
    (
      @input [nvarchar](MAX) ,
      @pattern [nvarchar](MAX) ,
      @matchDelimiter [nvarchar](MAX)
    )
RETURNS [nvarchar](MAX)
    WITH EXECUTE AS CALLER
AS EXTERNAL NAME 
    [SQL_CLR_RegExp].[SQL_CLR_RegExp.UserDefinedFunctions].[RegexSelectAll]
GO
-- =============================================
-- Returns finding matchIndex of a zero based index
-- RegexSelectOne(ByVal input As SqlChars, ByVal pattern As SqlString, ByVal matchIndex As SqlInt32) As SqlString
-- =============================================
CREATE FUNCTION [dbo].[RegexSelectOne]
    (
      @input [nvarchar](MAX) ,
      @pattern [nvarchar](MAX) ,
      @matchIndex [int]
    )
RETURNS [nvarchar](MAX)
    WITH EXECUTE AS CALLER
AS EXTERNAL NAME 
    [SQL_CLR_RegExp].[SQL_CLR_RegExp.UserDefinedFunctions].[RegexSelectOne]
GO 

SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
-- =============================================
-- Author:      <Jordon Pilling>
-- Create date: <30/01/2013>
-- Description: <Calls RegexSelectOne with start and end text and cleans the result>
-- =============================================
CREATE FUNCTION [dbo].[RegexSelectOneWithScrub]
(
    @Haystack VARCHAR(MAX),
    @StartNeedle VARCHAR(MAX),
    @EndNeedle VARCHAR(MAX)
)
RETURNS VARCHAR(MAX)
AS
BEGIN
    DECLARE @ReturnStr VARCHAR(MAX)

    --#### Extract text from HayStack using Start and End Needles
    SET @ReturnStr = dbo.RegexSelectOne(@Haystack, REPLACE(@StartNeedle, ' ','\s') + '((.|\n)+?)' + REPLACE(@EndNeedle, ' ','\s'), 0)

    --#### Remove the Needles
    SET @ReturnStr = REPLACE(@ReturnStr, @StartNeedle, '')
    SET @ReturnStr = REPLACE(@ReturnStr, @EndNeedle, '')

    --#### Trim White Space
    SET @ReturnStr = LTRIM(RTRIM(@ReturnStr))

    --#### Trim Line Breaks and Carriage Returns
    SET @ReturnStr = dbo.SuperTrim(@ReturnStr)

    RETURN @ReturnStr

END
GO

DECLARE @Subject VARCHAR(250) = 'HelpDesk Call Reference F0012345, Call Update, 40111' 
DECLARE @Ref VARCHAR(250) = NULL

IF dbo.RegexMatch(@Subject, '^HelpDesk\sCall\sReference\sF[0-9]{7},\s(Call\sResolved|Call\sUpdate|New\scall\slogged),(|\s+)([0-9]+|unknown)$') = 1
    SET @Ref = ISNULL(dbo.RegexSelectOneWithScrub(@Subject, 'HelpDesk Call Reference', ','), 'Invalid (#1)')
ELSE
    SET @Ref = 'Invalid (#2)'

SELECT @Ref

---

这在多个搜索中使用要快得多，在处理大量具有不同起始和结束短语的文本时功能更强大。

经过额外的工作，我们决定在Art的答案中不使用这种方法（即使它有效）

我们需要一种更强大的方法来验证和提取子字符串，因此我通过CLR路由使用正则表达式（感谢您为我指明了正确的方向）

我采取的方法如下：

WITH    ExampleData
          AS ( SELECT   'HelpDesk Call Reference F0012345, Call Update, 40111' AS [Subject]
               UNION ALL
               SELECT   'HelpDesk Call Reference F0012346, Call Resolved, 40112'
               UNION ALL
               SELECT   'HelpDesk Call Reference F0012347, New call logged, 40113'
             )
    SELECT  dbo.fnParseString(2, ',', REPLACE([Subject], 'HelpDesk Call Reference ', 'HelpDesk Call Reference, ')) AS [Ref] ,
            dbo.fnParseString(3, ',', REPLACE([Subject], 'HelpDesk Call Reference ', 'HelpDesk Call Reference, ')) AS [Type] ,
            dbo.fnParseString(4, ',', REPLACE([Subject], 'HelpDesk Call Reference ', 'HelpDesk Call Reference, ')) AS [OurRef]
    FROM    ExampleData

首先，我编译了以下CLR：（从C#示例转换为VB）

然后我编写了以下包装函数以简化使用：

EXEC sp_configure 
    'clr enabled' ,
    '1'

GO

RECONFIGURE
USE [db_Utility]

GO
CREATE ASSEMBLY SQL_CLR_RegExp FROM 'D:\Program Files\Microsoft SQL Server\MSSQL10_50.MSSQLSERVER\MSSQL\Binn\SQL_CLR_RegExp.dll' WITH
PERMISSION_SET = SAFE

GO
-- =============================================
-- Returns 1 or 0 if input matches pattern
-- VB function: RegexMatch(ByVal input As SqlChars, ByVal pattern As SqlString) As SqlBoolean
-- =============================================
CREATE FUNCTION [dbo].[RegexMatch]
    (
      @input [nvarchar](MAX) ,
      @pattern [nvarchar](MAX)
    )
RETURNS [bit]
    WITH EXECUTE AS CALLER
AS EXTERNAL NAME 
    [SQL_CLR_RegExp].[SQL_CLR_RegExp.UserDefinedFunctions].[RegexMatch]
GO

-- =============================================
-- Returns a comma separated string of found objects
-- VB function: RegexReplace(ByVal expression As SqlString, ByVal pattern As SqlString, ByVal replace As SqlString) As SqlString
-- =============================================
CREATE FUNCTION [dbo].[RegexReplace]
    (
      @expression [nvarchar](MAX) ,
      @pattern [nvarchar](MAX) ,
      @replace [nvarchar](MAX)
    )
RETURNS [nvarchar](MAX)
    WITH EXECUTE AS CALLER
AS EXTERNAL NAME 
    [SQL_CLR_RegExp].[SQL_CLR_RegExp.UserDefinedFunctions].[RegexReplace]
GO
-- =============================================
-- Returns a comma separated string of found objects
-- VB function: RegexSelectAll(ByVal input As SqlChars, ByVal pattern As SqlString, ByVal matchDelimiter As SqlString) As SqlString
-- =============================================
CREATE FUNCTION [dbo].[RegexSelectAll]
    (
      @input [nvarchar](MAX) ,
      @pattern [nvarchar](MAX) ,
      @matchDelimiter [nvarchar](MAX)
    )
RETURNS [nvarchar](MAX)
    WITH EXECUTE AS CALLER
AS EXTERNAL NAME 
    [SQL_CLR_RegExp].[SQL_CLR_RegExp.UserDefinedFunctions].[RegexSelectAll]
GO
-- =============================================
-- Returns finding matchIndex of a zero based index
-- RegexSelectOne(ByVal input As SqlChars, ByVal pattern As SqlString, ByVal matchIndex As SqlInt32) As SqlString
-- =============================================
CREATE FUNCTION [dbo].[RegexSelectOne]
    (
      @input [nvarchar](MAX) ,
      @pattern [nvarchar](MAX) ,
      @matchIndex [int]
    )
RETURNS [nvarchar](MAX)
    WITH EXECUTE AS CALLER
AS EXTERNAL NAME 
    [SQL_CLR_RegExp].[SQL_CLR_RegExp.UserDefinedFunctions].[RegexSelectOne]
GO 

SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
-- =============================================
-- Author:      <Jordon Pilling>
-- Create date: <30/01/2013>
-- Description: <Calls RegexSelectOne with start and end text and cleans the result>
-- =============================================
CREATE FUNCTION [dbo].[RegexSelectOneWithScrub]
(
    @Haystack VARCHAR(MAX),
    @StartNeedle VARCHAR(MAX),
    @EndNeedle VARCHAR(MAX)
)
RETURNS VARCHAR(MAX)
AS
BEGIN
    DECLARE @ReturnStr VARCHAR(MAX)

    --#### Extract text from HayStack using Start and End Needles
    SET @ReturnStr = dbo.RegexSelectOne(@Haystack, REPLACE(@StartNeedle, ' ','\s') + '((.|\n)+?)' + REPLACE(@EndNeedle, ' ','\s'), 0)

    --#### Remove the Needles
    SET @ReturnStr = REPLACE(@ReturnStr, @StartNeedle, '')
    SET @ReturnStr = REPLACE(@ReturnStr, @EndNeedle, '')

    --#### Trim White Space
    SET @ReturnStr = LTRIM(RTRIM(@ReturnStr))

    --#### Trim Line Breaks and Carriage Returns
    SET @ReturnStr = dbo.SuperTrim(@ReturnStr)

    RETURN @ReturnStr

END
GO

DECLARE @Subject VARCHAR(250) = 'HelpDesk Call Reference F0012345, Call Update, 40111' 
DECLARE @Ref VARCHAR(250) = NULL

IF dbo.RegexMatch(@Subject, '^HelpDesk\sCall\sReference\sF[0-9]{7},\s(Call\sResolved|Call\sUpdate|New\scall\slogged),(|\s+)([0-9]+|unknown)$') = 1
    SET @Ref = ISNULL(dbo.RegexSelectOneWithScrub(@Subject, 'HelpDesk Call Reference', ','), 'Invalid (#1)')
ELSE
    SET @Ref = 'Invalid (#2)'

SELECT @Ref

---

这在多个搜索中使用起来要快得多，在处理大量具有不同屏幕开始和结束短语的文本时功能更强大。

@Ben Thank:）您的-没有黑客攻击。+1很好，但只要数据一致且不超过4部分，此解决方案就可以工作。非常抱歉，您在我编辑问题时回答了我的问题–请参阅问题的编辑部分–基本上，我没有使用PARSENAME或FixedLength子字符串的奢侈：'（@Ben Thank:）您的-没有黑客攻击。+1很好，但只要数据一致且不超过4部分，此解决方案就可以工作。非常抱歉，您在我编辑问题时回答了我的问题–请参阅问题的编辑部分–基本上，我没有使用PARSENAME或FixedLength子字符串的奢侈：'（如果你有复杂的解析逻辑，我建议使用

CLR

function.CLR和Regex。如果你的格式没有相同的结构，你应该发布它们。你为什么要在TSQL中这样做？正如其他人所建议的那样，使用.NET Regex代码可能比使用TSQL容易得多，因为TSQL是一种不适合处理文本的语言。我会D考虑CLR程序或外部脚本/程序来解析数据。既然提到了发送电子邮件，在这种情况下，外部程序可能更容易。@ PANDLIFE我们最初希望在SQL中这样做，因为这是一个现有的系统，已经有很多SQL程序来处理不同类型的电子邮件。（而且效果非常好），但是，是的，我同意这是一个更复杂的问题，其他的和CLR可能是解决的方法-只需要有人把它作为答案，这样我就可以相应地标记它。@HeavenCore好的，让我们来具体化问题。你有由任何分隔符分隔的字符串。每个字符串可以有不同的部分计数，还是相同的部分计数ch一开始就知道？如果你有复杂的解析逻辑，我建议使用

CLR

function.CLR和Regex。如果你的格式没有相同的结构，你应该发布它们。为什么要在TSQL中这样做？正如其他人所建议的，使用.NET Regex代码可能比使用TSQL容易得多，因为TSQL是一种糟糕的语言