'Merge Fields' - 相似 SQL 服务器功能

Question

我试图找到一种方法让 SGBD 在长文本中执行大量合并字段。

创建结构：

CREATE TABLE [dbo].[store]
(
    [id] [int] NOT NULL,
    [text] [nvarchar](MAX) NOT NULL
)

CREATE TABLE [dbo].[statement]
(
    [id] [int] NOT NULL,
    [store_id] [int] NOT NULL
)

CREATE TABLE [dbo].[statement_merges]
(
    [statement_id] [int] NOT NULL,
    [merge_field] [nvarchar](30) NOT NULL,
    [user_data] [nvarchar](MAX) NOT NULL
)

现在，创建测试值

INSERT INTO [store] (id, text) 
VALUES (1, 'Waw, Whosebug is an amazing library of lost people in the IT hell, and i have the feeling that $$PERC_SAT$$ of the users found a solution, personally I asked $$ASKED$$ questions.')

INSERT INTO [statement] (id, store_id) 
VALUES (1, 1)

INSERT INTO [statement_merges] (statement_id, merge_field, user_data) 
VALUES (1, '$$PERC_SAT$$', '85%')

INSERT INTO [statement_merges] (statement_id, merge_field, user_data) 
VALUES (1, '$$ASKED$$', '12')

目前我的应用程序正在传递最终语句，循环合并，替换存储的文本和输出

Waw, Whosebug is an amazing library of lost people in the IT hell, and i have the feeling that 85% of the users found a solution, personally I asked 12 questions.

我试图找到一种独立于代码并在单个查询中提供输出的方法，正如您所理解的那样，select 一条语句，其中存储的文本已填充了用户数据。我希望我已经清楚了。

我查看了 TRANSLATE 函数，但它看起来像是一个字符替换，所以我有两个选择：

我尝试一个递归函数，一个一个替换，直到计算出的文本中没有merge_fields；但我对这种方法的性能有疑问；
有魔法可以做到，但我需要你的知识...

考虑到我想要这个是因为真实的文本很长，我不想在我的数据库中多次存储它。您可以想象一份只有 12 个参数的 3 页合同，例如开始日期、开票金额等...为了合规性，其他一切都无法更改。

感谢您的宝贵时间！

编辑：

感谢 Randy 的帮助，这看起来可以解决问题：

WITH cte_replace_tokens AS (

    SELECT replace(r.text, m.merge_field, m.user_data) as [final], m.merge_field, s.id, 1 AS i
    FROM store r
    INNER JOIN statement s ON s.store_id = r.id
    INNER JOIN statement_merges m ON m.statement_id = s.id
    WHERE m.statement_id = 1

    UNION ALL

    SELECT replace(r.final, m.merge_field, m.user_data) as [final], m.merge_field, r.id, r.i + 1 AS i
    FROM cte_replace_tokens r
    INNER JOIN statement_merges m ON m.statement_id = r.id
    WHERE m.merge_field > r.merge_field

) 

select TOP 1 final from cte_replace_tokens ORDER BY i DESC

如果性能好，我会检查更大的数据库...

至少，我可以“填充”一个语句，我还需要弄清楚才能提取一个列表。

再次感谢！

Answer 1

不建议在 sql 引擎中执行此类任务，但如果你想这样做，你需要在函数或存储过程中使用游标循环执行，如下所示：

DECLARE @merge_field nvarchar(30)
    , @user_data nvarchar(MAX)
    , @statementid INT = 1 
    , @text varchar(MAX) = 'Waw, Whosebug is an amazing library of lost people in the IT hell, and i have the feeling that $$PERC_SAT$$ of the users found a solution, personally I asked $$ASKED$$ questions.'

DECLARE  merge_statements CURSOR FAST_FORWARD
 FOR SELECT 
    sm.merge_field
    , sm.user_data 
    FROM dbo.statement_merges AS sm
    WHERE sm.statement_id = @statementid

 OPEN merge_statements
 FETCH NEXT FROM merge_statements
 INTO @merge_field , @user_data
 WHILE @@FETCH_STATUS = 0  
  BEGIN
    set @text = REPLACE(@text , @merge_field, @user_data )
    FETCH NEXT FROM merge_statements
    INTO @merge_field , @user_data
END 
CLOSE   merge_statements
DEALLOCATE merge_statements

SELECT @text

Answer 2

如果一条记录被同一次更新多次更新，最后一次获胜。 None 的更新受其他更新影响 - 无累积效应。在某些情况下，可以使用局部变量来欺骗 SQL 以获得累积效果，但这很棘手且不推荐。（顺序变得重要并且在更新中不可靠。）

一个替代方案是 CTE 中的递归。随着每个令牌被替换，直到没有令牌为止，从先前生成一个新记录。这是一个工作示例，将 1 替换为 A，将 2 替换为 B，等等。（我想知道是否有一些棘手的 xml 也可以做到这一点。）

if not object_id('tempdb..#Raw') is null drop table #Raw
CREATE TABLE #Raw(
    [test] [varchar](100) NOT NULL PRIMARY KEY CLUSTERED,
)

if not object_id('tempdb..#Token') is null drop table #Token
CREATE TABLE #Token(
    [id] [int] NOT NULL PRIMARY KEY CLUSTERED,
    [token] [char](1) NOT NULL,
    [value] [char](1) NOT NULL,
)

insert into #Raw values('123456'), ('1122334456')
insert into #Token values(1, '1', 'A'), (2, '2', 'B'), (3, '3', 'C'), (4, '4', 'D'), (5, '5', 'E'), (6, '6', 'F');

WITH cte_replace_tokens AS (

    SELECT r.test, replace(r.test, l.token, l.value) as [final], l.id
    FROM [Raw] r
    CROSS JOIN #Token l
    WHERE l.id = 1

    UNION ALL

    SELECT r.test, replace(r.final, l.token, l.value) as [final], l.id
    FROM cte_replace_tokens r
    CROSS JOIN #Token l
    WHERE l.id = r.id + 1

) 
select * from cte_replace_tokens where id = 6

Answer 3

这是一个递归的解决方案。 SQL Fiddle

MS SQL Server 2017 架构设置:

CREATE TABLE [dbo].[store]
(
    [id] [int] NOT NULL,
    [text] [nvarchar](MAX) NOT NULL
)

CREATE TABLE [dbo].[statement]
(
    [id] [int] NOT NULL,
    [store_id] [int] NOT NULL
)

CREATE TABLE [dbo].[statement_merges]
(
    [statement_id] [int] NOT NULL,
    [merge_field] [nvarchar](30) NOT NULL,
    [user_data] [nvarchar](MAX) NOT NULL
)
INSERT INTO store (id, text) 
VALUES (1, '$$(*)$$, Whosebug...$$PERC_SAT$$...$$ASKED$$ questions.')

INSERT INTO store (id, text) 
VALUES (2, 'Use The @_@')

INSERT INTO statement (id, store_id) VALUES (1, 1)
INSERT INTO statement (id, store_id) VALUES (2, 2)
INSERT INTO statement_merges (statement_id, merge_field, user_data) VALUES (1, '$$PERC_SAT$$', '85%')
INSERT INTO statement_merges (statement_id, merge_field, user_data) VALUES (1, '$$ASKED$$', '12')
INSERT INTO statement_merges (statement_id, merge_field, user_data) VALUES (1, '$$(*)$$', 'Wow')
INSERT INTO statement_merges (statement_id, merge_field, user_data) VALUES (2, ' @_@', 'Flux!')

查询 1:

;WITH Normalized AS
( 
    SELECT
        store_id=store.id,
        store.text,
        sm.merge_field,
        sm.user_data,
        RowNumber = ROW_NUMBER() OVER(PARTITION BY store.id,sm.statement_id ORDER BY merge_field),
        statement_id = st.id
    FROM
        store store
        INNER JOIN statement st ON st.store_id = store.id
        INNER JOIN statement_merges sm ON sm.statement_id = st.id
)
, Recurse AS
(
    SELECT 
        store_id, statement_id, old_text = text, merge_field,user_data, RowNumber,
        Iteration=1,
        new_text = REPLACE(text, merge_field, user_data)
    FROM 
        Normalized 
    WHERE 
        RowNumber=1

    UNION ALL

    SELECT 
        n.store_id, n.statement_id, r.old_text, n.merge_field, n.user_data,
        RowNumber=r.RowNumber+1,
        Iteration=Iteration+1,
        new_text = REPLACE(r.new_text, n.merge_field, n.user_data)
    FROM 
        Normalized n
        INNER JOIN Recurse r ON r.RowNumber = n.RowNumber AND r.statement_id = n.statement_id
)
,ReverseOnIteration AS
(
    SELECT *,
        ReverseIteration = ROW_NUMBER() OVER(PARTITION BY statement_id ORDER BY Iteration DESC)
    FROM
        Recurse
)
SELECT 
    store_id, statement_id, new_text, old_text
FROM 
    ReverseOnIteration
WHERE
    ReverseIteration=1

Results:

| store_id | statement_id |                                 new_text |                                                     old_text |
|----------|--------------|------------------------------------------|--------------------------------------------------------------|
|        1 |            1 | Wow, Whosebug...85%...12 questions. | $$(*)$$, Whosebug...$$PERC_SAT$$...$$ASKED$$ questions. |
|        2 |            2 |                             Use TheFlux! |                                                  Use The @_@ |

Answer 4

在Randy的帮助下，我想我已经实现了我想做的事情！

据了解，我的真实案例是一份合同，其中有几条语句可能是：

自由文本
没有任何合并的存储文本
存储的文本带有一个或几次合并

这个 CTE 可以胜任！

WITH cte_replace_tokens AS (

    -- The initial query dont join on merges neither on store because can be a free text 
    SELECT COALESCE(r.text, s.part_text) AS [final], CAST('' AS NVARCHAR) AS merge_field, s.id, 1 AS i, s.contract_id
    FROM statement s
    LEFT JOIN store r ON s.store_id = r.id
    
    UNION ALL
    
    -- We loop till the last merge field, output contains iteration to be able to keep the last record ( all fields updated )
    SELECT replace(r.final, m.merge_field, m.user_data) as [final], m.merge_field, r.id, r.i + 1 AS i, r.contract_id
    FROM cte_replace_tokens r
    INNER JOIN statement_merges m ON m.statement_id = r.id
    WHERE m.merge_field > r.merge_field AND r.final LIKE '%' + m.merge_field + '%'
    -- spare lost replacements by forcing only one merge_field per loop
    AND NOT EXISTS( SELECT mm.statement_id FROM statement_merges mm WHERE mm.statement_id = m.statement_id AND mm.merge_field > r.merge_field AND mm.merge_field < m.merge_field)
) 

select s.id, 
(select top 1 final from cte_replace_tokens t WHERE t.contract_id = s.contract_id AND t.id = s.id ORDER BY i DESC) as res
FROM statement s
where contract_id = 1

Answer 5

如果带有交叉连接的 CTE 解决方案太慢，另一种解决方案是动态构建一个标量 fn，它具有令牌 table 所需的每个 REPLACE。每个记录的一个标量 fn 调用是 order(N)。我得到了和以前一样的结果。

函数很简单，可能不会太长，具体取决于令牌的大小 table 变成...256 MB 的批量限制。我已经看到尝试动态创建查询以提高性能适得其反 - 将问题转移到编译时。这里应该不是问题。

if not object_id('tempdb..#Raw') is null drop table #Raw
CREATE TABLE #Raw(
    [test] [varchar](100) NOT NULL PRIMARY KEY CLUSTERED,
)

if not object_id('tempdb..#Token') is null drop table #Token
CREATE TABLE #Token(
    [id] [int] NOT NULL PRIMARY KEY CLUSTERED,
    [token] [char](1) NOT NULL,
    [value] [char](1) NOT NULL,
)

insert into #Raw values('123456'), ('1122334456')
insert into #Token values(1, '1', 'A'), (2, '2', 'B'), (3, '3', 'C'), (4, '4', 'D'), (5, '5', 'E'), (6, '6', 'F');

DECLARE @sql varchar(max) = 'CREATE FUNCTION dbo.fn_ReplaceTokens(@raw varchar(8000)) RETURNS varchar(8000) AS BEGIN RETURN ';

WITH cte_replace_statement AS (

    SELECT a.id, CAST('replace(@raw,''' + a.token + ''',''' + a.value + ''')' as varchar(max)) as [statement]
    FROM #Token a
    WHERE a.id = 1

    UNION ALL

    SELECT n.id, CAST(replace(l.[statement], '@raw', 'replace(@raw,''' + n.token + ''',''' + n.value + ''')') as varchar(max)) as [statement]
    FROM #Token n
    INNER JOIN cte_replace_statement l
    ON n.id = l.id + 1

) 
select @sql += [statement] + ' END' from cte_replace_statement where id = 6

print @sql

if not object_id('dbo.fn_ReplaceTokens') is null drop function dbo.fn_ReplaceTokens
execute (@sql)

SELECT r.test, dbo.fn_ReplaceTokens(r.test) as [final] FROM [Raw] r

'Merge Fields' - 相似 SQL 服务器功能

'Merge Fields' - alike SQL Server function

sql

sql-server

replace

recursive-query