根据分隔符拆分两列

Question

我有一个 TSQL (MSSQL) 表，其中包含以下格式的记录

ID	第 1 列	第 2 列
1	a/b/c	苹果/香蕉/黄瓜

我想以以下格式拆分记录

ID	第 1 列	第 2 列
1	一个	苹果
1	b	香蕉
1	c	cucumber

Column1 和 Column2 使用“/”分隔符保持关系，并以相同的顺序相互关联。

我试图在 CHARINDEX 和 SUBSTRING 的帮助下拆分列，但我无法维持两列之间的关系。

Answer 1

您可以添加 function 来拆分字符串。

然后交叉应用于Column1和Column2的分割部分。

 create table test ( Id int identity primary key, Column1 varchar(30), Column2 varchar(30) ); insert into test (Column1, Column2) values ('a/b/c', 'apple/banana/cucumber'), ('d/e/f', 'orange/prune/onion');

（从这里复制的UDF）

 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 t.Id, ca.Column1, ca.Column2 from test t cross apply ( select s1.ordinal, s1.value as Column1, s2.value as Column2 from dbo.fnString_Split(t.Column1,'/') as s1 join dbo.fnString_Split(t.Column2,'/') as s2 on s1.ordinal = s2.ordinal ) ca;

ID	第 1 列	第 2 列
1	一个	苹果
1	b	香蕉
1	c	cucumber
2	d	橙
2	e	修剪
2	F	洋葱

关于db<>fiddle 的演示在这里

Answer 2

请尝试以下解决方案。

它基于 JSON，将从 SQL Server 2016 开始工作。

SQL

-- DDL and sample data population, start
DECLARE @tbl TABLE (ID INT, ColB varchar(8000), ColC varchar(8000));
INSERT INTO @tbl VALUES
(1,'a/b/c','apple/banana/cucumber');
-- DDL and sample data population, end

DECLARE @separator CHAR(1) = '/';

WITH rs AS 
(
     SELECT *
         , ar1 = '["' + REPLACE(ColB, @separator, '","') + '"]'
         , ar2 = '["' + REPLACE(ColC, @separator, '","') + '"]'
     FROM @tbl
 )
 SELECT ID, ColB.[value] AS [ColB], ColC.[value] AS ColC
 FROM rs
    CROSS APPLY OPENJSON (ar1, N'$') AS ColB
    CROSS APPLY OPENJSON (ar2, N'$') AS ColC
 WHERE ColB.[key] = ColC.[key];

Output

+----+------+----------+
| ID | ColB |   ColC   |
+----+------+----------+
|  1 | a    | apple    |
|  1 | b    | banana   |
|  1 | c    | cucumber |
+----+------+----------+

Answer 3

我知道已经有一些答案，其中一个已被接受，但有一些重要的性能因素需要考虑。 如果值的数量总是三个然后（或者总是很低，比如小于 5 或 6），那么 Cascading APPLY 技术将是迄今为止最快的。 此解决方案假定始终有 3 个项目。 可以轻松修改它以处理可变但数量很少的项目。

级联应用解决方案：

DECLARE @table TABLE
(
  SomeId  INT IDENTITY,
  S1 VARCHAR(1000),
  s2 VARCHAR(1000)
);
INSERT @table VALUES ('a/b/c','apple/banana/cucumber'),
                     ('d/d2/f','dog/donkey/fish'),('x/y/z','x-ray/yo-yo/zeta');
SELECT
  SomeId = f.SomeId,
  Col1   = f2.C1,
  Col2   = f2.C2
FROM
(
  SELECT
    t.SomeId, 
    SUBSTRING(t.S1, 1,       c1.P-1),
    SUBSTRING(t.S1, c1.P+1,  c2.P - c1.P-1),
    SUBSTRING(t.S1, c2.P+1,  8000),
    SUBSTRING(t.S2, 1,       c1.P2 - 1),
    SUBSTRING(t.S2, c1.P2+1, c2.P2 - c1.P2-1),
    SUBSTRING(t.S2, c2.P2+1, 8000)
  FROM        @table AS t
  CROSS APPLY (VALUES(CHARINDEX('/',t.S1),CHARINDEX('/',t.S2)))                AS c1(P,P2)
  CROSS APPLY (VALUES(CHARINDEX('/',t.S1,c1.P+1),CHARINDEX('/',t.S2,c1.P2+1))) AS c2(P,P2)
  CROSS APPLY (VALUES(CHARINDEX('/',t.S1,c2.P+1),CHARINDEX('/',t.S2,c2.P2+1))) AS c3(P,P2)
) AS f(SomeId,c1_1,c1_2,c1_3,c2_1,c2_2,c2_3)
CROSS APPLY (VALUES (c1_1, c2_1), (c1_2, c2_2),(c1_3,c2_3)) AS f2(c1,c2);

如果你使用 SQL Azure 你有 STRING_SPLIT和ordinal 选项。

DECLARE @table TABLE
(
  SomeId  INT IDENTITY,
  S1 VARCHAR(1000),
  s2 VARCHAR(1000)
);
INSERT @table VALUES ('a/b/c','apple/banana/cucumber'),
                     ('d/d2/f','dog/donkey/fish'),('x/y/z','x-ray/yo-yo/zeta');

SELECT TOP(1) WITH TIES
  t.SomeId, t.S1, t.S2, Col1 = split1.[value], Col2 = split2.[value]
FROM        @table                  AS t
CROSS APPLY STRING_SPLIT(t.S1 ,'/') AS split1
CROSS APPLY STRING_SPLIT(t.S2 ,'/') AS split2
ORDER BY ABS(
  ROW_NUMBER() OVER (PARTITION BY t.SomeId, split1.value ORDER BY split1.value)-
  ROW_NUMBER() OVER (PARTITION BY t.SomeId, split2.value ORDER BY split1.value));

^^^ 这仅适用于按字母顺序排列的项目（不实用）但是，在 Azure 上，您可以将 ORDER BY 更改为：

ORDER BY ABS(
  ROW_NUMBER() OVER (PARTITION BY t.SomeId, split1.value ORDER BY split1.ordianal) -
  ROW_NUMBER() OVER (PARTITION BY t.SomeId, split2.value ORDER BY split1.ordianal));

比较发布的所有技术

现在让我们比较迄今为止发布的所有解决方案，以了解显着的性能差异。 我构建了一个基本的测试工具并运行它，首先是 100K 行，然后是一百万行。 解决方案包括：

• Andy3B 的 dbo.udf_SplitList
• LukStorms 的 dbo.fnString_Split
• 递归 CTE - DhruvJoshi 的数字/计数表解决方案
• DhruvJoshi 解决方案的改进版本，具有更快的计数表
• Yitzhak Khabinsky 的 JSON 解决方案
• 我的级联应用解决方案

对于一百万行，dbo.udf_SplitList 解决方案最慢，为 246 秒。 标量函数非常慢，但递归使事情变得更糟。

dbo.fnString_Split 解决方案在 85 秒时表现更好。 为了做得更好，我们需要丢失标量 udf。

DhruvJoshi 的递归 CTE 解决方案将我们缩短到 65 秒，提高了 50%。 我通过重写数字表以不使用递归来改进他的解决方案； 这使我们的速度再提高 50%，降至 45 秒。

Yitzhak Khabinsky 是第一个基于集合的解决方案； 他正在利用 JSON。 在这里，我们有 200% 以上的性能提升，降至 16 秒。 Cascading APPLY 解决方案比 Yitzhak 的解决方案提高了 400% 以上； 降到三秒以下。

注意标量 UDF、计数和循环的递归。 基于集合的始终规则。

IF OBJECT_ID('tempdb..#table') IS NOT NULL DROP TABLE #table;
GO
SELECT TOP(100000)
  SomeId = ROW_NUMBER() OVER (ORDER BY (SELECT NULL)),
  S1     = 'A/B/C',
  S2     = REPLACE(LEFT(NEWID(),18),'-','/')
INTO #table
FROM sys.all_columns, sys.all_columns a
GO

PRINT CHAR(10)+'dbo.udf_SplitList -Andy3B'+CHAR(10)+REPLICATE('-',90);
GO
 DECLARE @st DATETIME = GETDATE(), @ID INT, @C1 VARCHAR(1000), @C2 VARCHAR(1000);
 
 SELECT @ID = t.SomeId, @C1 = x.Code, @C2 = x.Value
 FROM    #table AS t
 CROSS APPLY
 (
   SELECT Code  = C.Value,
          Value = V.Value
   FROM    dbo.udf_SplitList(t.S1, '/') AS C
   JOIN    dbo.udf_SplitList(t.S2, '/') AS V ON V.ID = C.ID
 ) AS X;

 PRINT DATEDIFF(MS,@st,GETDATE())
GO 3

PRINT CHAR(10)+'dbo.fnString_Split - LukStorms'+CHAR(10)+REPLICATE('-',90);
GO
 DECLARE @st DATETIME = GETDATE(), @ID INT, @C1 VARCHAR(1000), @C2 VARCHAR(1000);

  select @ID = t.SomeId, @C1 = ca.Column1, @C2 = ca.Column2
  from   #table AS t
  cross apply (
    select 
      s1.ordinal
    , s1.value as Column1
    , s2.value as Column2
    from dbo.fnString_Split(t.S1,'/') as s1
    join dbo.fnString_Split(t.S2,'/') as s2
      on s1.ordinal = s2.ordinal) AS ca;

 PRINT DATEDIFF(MS,@st,GETDATE())
GO 3

PRINT CHAR(10)+'Recursive CTE - TALLY TABLE DhruvJoshi'+CHAR(10)+REPLICATE('-',90);
GO
DECLARE @st DATETIME = GETDATE(), @ID INT, @C1 VARCHAR(1000), @C2 VARCHAR(1000);

;with nums as
(
  select 1 as num union all 
  select num +1 as num
  from   nums 
  where  num <80
)
select 
   @ID = X.SomeId,
   @C1 = substring(X.S1,X.b,X.e-X.b),
   @C2 = substring(Y.S2,Y.b,Y.e-Y.b)
from 
(
  select 
    t.*,
    e=N.Num,
    r=row_number()      over (partition by t.SomeId order by N.num),
    b=isnull(lag(N.num) over (partition by t.SomeId order by N.num),0)+1
  from      #table AS t
  left join nums   AS N 
    on      charindex('/',t.S1+'/',N.num)=N.num 
) AS X
left join
(
  select 
    t.*,
    e= N.Num,
    r=row_number()      over (partition by t.SomeId order by N.num),
    b=isnull(lag(N.num) over (partition by t.SomeId order by N.num),0)+1
  from      #table AS t
  left join nums   AS N 
     on charindex('/', t.S2+'/',N.num)=N.num 
) AS Y
  ON X.SomeId = Y.SomeId and X.r=Y.r;
 
 PRINT DATEDIFF(MS,@st,GETDATE())
GO 3

PRINT CHAR(10)+'TALLY TABLE without Recursion DhruvJoshi'+CHAR(10)+REPLICATE('-',90);
GO
DECLARE @st DATETIME = GETDATE(), @ID INT, @C1 VARCHAR(1000), @C2 VARCHAR(1000);

;with nums as
(
  SELECT num = ROW_NUMBER() OVER (ORDER BY (SELECT NULL))
  FROM (VALUES(1),(1),(1),(1),(1),(1),(1),(1),(1),(1)) AS e1(x),
       (VALUES(1),(1),(1),(1),(1),(1),(1),(1),(1),(1)) AS e2(x),
       (VALUES(1),(1),(1),(1),(1),(1),(1),(1),(1),(1)) AS e3(x),
       (VALUES(1),(1),(1),(1),(1),(1),(1),(1),(1),(1)) AS e4(x)
)
select 
   @ID = X.SomeId,
   @C1 = substring(X.S1,X.b,X.e-X.b),
   @C2 = substring(Y.S2,Y.b,Y.e-Y.b)
from 
(
  select 
    t.*,
    e=N.Num,
    r=row_number()      over (partition by t.SomeId order by N.num),
    b=isnull(lag(N.num) over (partition by t.SomeId order by N.num),0)+1
  from      #table AS t
  left join nums   AS N 
    on      charindex('/',t.S1+'/',N.num)=N.num 
  WHERE n.num <= LEN(t.S1)
) AS X
left join
(
  select 
    t.*,
    e= N.Num,
    r=row_number()      over (partition by t.SomeId order by N.num),
    b=isnull(lag(N.num) over (partition by t.SomeId order by N.num),0)+1
  from      #table AS t
  left join nums   AS N 
     on charindex('/', t.S2+'/',N.num)=N.num
  WHERE n.num <= LEN(t.S2)
) AS Y
  ON X.SomeId = Y.SomeId and X.r=Y.r; 
 PRINT DATEDIFF(MS,@st,GETDATE())
GO 3

PRINT CHAR(10)+'JSON - Yitzhak Khabinsky'+CHAR(10)+REPLICATE('-',90);
GO
 DECLARE @st DATETIME = GETDATE(), @ID INT, @C1 VARCHAR(1000), @C2 VARCHAR(1000);
 DECLARE @separator CHAR(1) = '/';
 
 WITH rs AS 
 (
      SELECT *
          , ar1 = '["' + REPLACE(t.S1, @separator, '","') + '"]'
          , ar2 = '["' + REPLACE(t.S2, @separator, '","') + '"]'
      FROM #table AS t
 )
 SELECT @id = SomeID, @C1 = ColB.[value], @C2 = ColC.[value] 
 FROM        rs                   AS rs
 CROSS APPLY OPENJSON (ar1, N'$') AS ColB
 CROSS APPLY OPENJSON (ar2, N'$') AS ColC
 WHERE       ColB.[key] = ColC.[key];
 
 PRINT DATEDIFF(MS,@st,GETDATE())
GO 3

PRINT CHAR(10)+'CROSS APPLY TECHNIQUE'+CHAR(10)+REPLICATE('-',90);
GO
 DECLARE @st DATETIME = GETDATE(), @ID INT, @C1 VARCHAR(1000), @C2 VARCHAR(1000);
 
 SELECT
    @ID = f.SomeId,
    @C1   = f2.C1,
    @C2   = f2.C2
  FROM
  (
    SELECT
      t.SomeId, 
      SUBSTRING(t.S1, 1,       c1.P-1),
      SUBSTRING(t.S1, c1.P+1,  c2.P - c1.P-1),
      SUBSTRING(t.S1, c2.P+1,  8000),
      SUBSTRING(t.S2, 1,       c1.P2 - 1),
      SUBSTRING(t.S2, c1.P2+1, c2.P2 - c1.P2-1),
      SUBSTRING(t.S2, c2.P2+1,  8000)
    FROM        #table AS t
    CROSS APPLY (VALUES(CHARINDEX('/',t.S1),CHARINDEX('/',t.S2)))                AS c1(P,P2)
    CROSS APPLY (VALUES(CHARINDEX('/',t.S1,c1.P+1),CHARINDEX('/',t.S2,c1.P2+1))) AS c2(P,P2)
    CROSS APPLY (VALUES(CHARINDEX('/',t.S1,c2.P+1),CHARINDEX('/',t.S2,c2.P2+1))) AS c3(P,P2)
  ) AS f(SomeId,c1_1,c1_2,c1_3,c2_1,c2_2,c2_3)
  CROSS APPLY (VALUES (c1_1, c2_1), (c1_2, c2_2),(c1_3,c2_3)) AS f2(c1,c2);
 
 PRINT DATEDIFF(MS,@st,GETDATE())
GO 3

100K 行测试结果：

dbo.udf_SplitList -Andy3B
------------------------------------------------------------------------------------------
Beginning execution loop
24593
24566
24530
Batch execution completed 3 times.

dbo.fnString_Split - LukStorms
------------------------------------------------------------------------------------------
Beginning execution loop
8147
8260
8257
Batch execution completed 3 times.

Recursive CTE - TALLY TABLE DhruvJoshi
------------------------------------------------------------------------------------------
Beginning execution loop
6867
6733
6850
Batch execution completed 3 times.

TALLY TABLE without Recursion DhruvJoshi
------------------------------------------------------------------------------------------
Beginning execution loop
4677
4630
4620
Batch execution completed 3 times.

JSON - Yitzhak Khabinsky
------------------------------------------------------------------------------------------
Beginning execution loop
1667
1653
1670
Batch execution completed 3 times.

CROSS APPLY TECHNIQUE
------------------------------------------------------------------------------------------
Beginning execution loop
283
280
284
Batch execution completed 3 times.

100万行测试结果：

dbo.udf_SplitList -Andy3B
------------------------------------------------------------------------------------------
Beginning execution loop
246057
245296
247017
Batch execution completed 3 times.

dbo.fnString_Split - LukStorms
------------------------------------------------------------------------------------------
Beginning execution loop
85340
83010
83674
Batch execution completed 3 times.

Recursive CTE - TALLY TABLE -DhruvJoshi
------------------------------------------------------------------------------------------
Beginning execution loop
67226
64910
64740
Batch execution completed 3 times.

TALLY TABLE without Recursion DhruvJoshi
------------------------------------------------------------------------------------------
Beginning execution loop
46777
44630
44623
Batch execution completed 3 times.

JSON - Yitzhak Khabinsky
------------------------------------------------------------------------------------------
Beginning execution loop
16710
16830
16520
Batch execution completed 3 times.

CROSS APPLY TECHNIQUE
------------------------------------------------------------------------------------------
Beginning execution loop
2846
2793
2850
Batch execution completed 3 times.

Completion time: 2022-01-18T22:08:50.3264912-06:00

Answer 4

1. 首先创建下面的function来拆分字符串。
2. 然后，执行 function 代码后面的代码。

    -- Function Code
    CREATE FUNCTION [dbo].[udf_SplitList]
        (   
                @InputString    varchar(MAX)    
            ,   @Separator      varchar(1)  
        )
        RETURNS @ValuesList TABLE ( ID int IDENTITY(1,1), Value varchar(MAX))
        AS
            BEGIN
                DECLARE @ListValue NVARCHAR(max)
                SET @InputString = @InputString + @Separator
                WHILE (LEN(@InputString) > 0)
                    BEGIN
                        SELECT @ListValue = SUBSTRING(@InputString , 1, CHARINDEX(@Separator, @InputString) - 1)
                        INSERT INTO @ValuesList
                        SELECT LTRIM(@ListValue)
                        SELECT @InputString = SUBSTRING(@InputString, CHARINDEX(@Separator, @InputString) + 1 , LEN(@InputString) - CHARINDEX(@Separator, @InputString)) 
                END 
                RETURN 
            END

    -- Execution Code
    DECLARE     @YourTable TABLE (ID int, CodeList varchar(MAX), ValueList varchar(MAX));
    INSERT INTO @YourTable  VALUES ( 1, 'a/b/c', 'apple/banana/cucumber');
    SELECT  X.*
    FROM    @YourTable  Y
    CROSS APPLY
        (
            SELECT 
                    Code  = C.Value
                ,   Value = V.Value
            FROM    dbo.udf_SplitList(Y.CodeList , '/') C
            JOIN    dbo.udf_SplitList(Y.ValueList, '/') V ON V.ID = C.ID
        ) X
    ;

Answer 5

在不使用 function 的情况下，您可以使用计数/数字表来执行以下相同的操作

在此处查看工作演示

; with nums as
(
select 1 as num
union all 
select num +1 as num
from 
nums 
where num <80
)
select 
   X.id,
   substring(X.column1,X.b,X.e-X.b),
   substring(Y.column2,Y.b,Y.e-Y.b)
from 
(
    select 
        t.*,
        e=N.Num,
        r=row_number() over(partition by Id order by N.num),
        b=isnull(lag(N.num) over (partition by Id order by N.num),0)+1
     from t
        left join nums N 
            on charindex('/',column1+'/',N.num)=N.num 
)X
left join
(
    select 
        t.*,
        e= N.Num,
        r=row_number() over(partition by Id order by N.num),
        b=isnull(lag(N.num) over (partition by Id order by N.num),0)+1
    from t
        left join nums N 
             on charindex('/', column2+'/',N.num)=N.num 
)Y
       on X.id=Y.id and X.r=Y.r