有没有办法使用窗口函数来简化这个查询？

Question

我有一些数据，真实世界的例子。为了以特定方式显示数据，数据的存储方式使得我必须对其进行操作才能获得我想要的结果。从本质上讲，有两种不同类型的发票：预发票 (P)，其中在处理订单之前开具账单；和标准发票 (I)，其中剩余部分在发货后开具账单。在一般情况下，您会期望标准发票会减去发票前金额，例如，如果发票前金额为 2 美元而订单金额为 10 美元，则标准发票将为 8 美元。不过，标准发票存储为 10 美元。

下面是一个完全可操作的查询（该查询是可操作的！这是一个陷阱！），它既填充数据又 returns 我想要的结果。目标是获取发票金额，如果是预发票，则 return 该金额；但如果是标准发票，则 "use up" 预发票的价值和 return 新金额，最小为零。我已经包括了六种情况，因为预发票可以是任何金额，并且在技术上可能随时需要。

如有任何帮助，我们将不胜感激。我尝试了一些窗口函数，包括 UNBOUNDED PRECEDING 类型的东西，但似乎我总是需要递归并进入无限循环。因此，下面是我的蛮力方法。

DECLARE @PData TABLE (
     CustNum        INT
    ,TransNum       INT
    ,InvType        NVARCHAR(1)
    ,InvAmt         DECIMAL(5,2)
    ,CRank          INT
    ,TRank          INT
    ,ModInvAmt      DECIMAL(5,2)
    )

INSERT INTO @PData (
     CustNum
    ,TransNum
    ,InvType
    ,InvAmt
    )

    VALUES
     (124, 1,'P',2)
    ,(124, 2,'I',10)
    ,(124, 3,'I',10)
    ,(153, 4,'I',10)
    ,(153, 5,'P',2)
    ,(153, 6,'I',10)
    ,(324, 7,'I',10)
    ,(324, 8,'I',10)
    ,(324, 9,'P',2)
    ,(441,10,'P',12)
    ,(441,11,'I',10)
    ,(441,12,'I',10)
    ,(455,13,'I',10)
    ,(455,14,'P',12)
    ,(455,15,'I',10)
    ,(667,16,'I',10)
    ,(667,17,'I',10)
    ,(667,18,'P',12)

UPDATE pd1
    SET CRank = pd2.CDR
        FROM @PData pd1
            JOIN (SELECT CustNum, TransNum, DENSE_RANK() OVER (ORDER BY CustNum) AS CDR
                    FROM @PData) pd2
                ON pd1.TransNum = pd2.TransNum

UPDATE pd1
    SET TRank = pd2.TDR
        FROM @PData pd1
            JOIN (SELECT CustNum, TransNum, DENSE_RANK() OVER (PARTITION BY CustNum ORDER BY TransNum) AS TDR
                    FROM @PData) pd2
                ON pd1.TransNum = pd2.TransNum

DECLARE  @Counter1      INT
        ,@Counter2      INT
        ,@CBal          DECIMAL(5,2)
        ,@TNum          INT
        ,@IAmt          DECIMAL(5,2)

SET @Counter1 = 0

WHILE @Counter1 < (SELECT MAX(CRank) FROM @PData)
    BEGIN
        SET @Counter1 += 1
        SET @CBal = 0
        SET @Counter2 = 0
            WHILE @Counter2 < (SELECT MAX(TRank) FROM @PData WHERE CRank = @Counter1)
                BEGIN
                    SET @Counter2 += 1
                    SET @TNum = (SELECT TransNum FROM @PData WHERE CRank = @Counter1 AND TRank = @Counter2)
                    SET @IAmt = (SELECT InvAmt FROM @PData WHERE TransNum = @TNum)
                    IF (SELECT InvType FROM @PData WHERE TransNum = @TNum) = 'P'
                        BEGIN
                            UPDATE @PData SET ModInvAmt = @IAmt WHERE TransNum = @TNum
                            SET @CBal += -@IAmt 
                        END
                    ELSE
                        BEGIN
                            UPDATE @PData SET ModInvAmt = (ABS(@IAmt+@CBal)+(@IAmt+@CBal))/2 -- MINIMUM = 0
                                 WHERE TransNum = @TNum
                            SET @CBal += (@IAmt - (ABS(@IAmt+@CBal)+(@IAmt+@CBal))/2)
                        END                 
                END
    END

SELECT   CustNum
        ,TransNum
        ,InvType
        ,InvAmt
        ,ModInvAmt
 FROM @PData

这是我得到的结果：

我通常不会报告原始发票金额——只报告新发票金额——但我将其包含在此处，以便更清楚地了解它是如何变化的。

 CustNum TransNum InvType InvAmt  ModInvAmt
 124      1         P     2.00      2.00
 124      2         I    10.00      8.00
 124      3         I    10.00     10.00
 153      4         I    10.00     10.00
 153      5         P     2.00      2.00
 153      6         I    10.00      8.00
 324      7         I    10.00     10.00
 324      8         I    10.00     10.00
 324      9         P     2.00      2.00
 441     10         P    12.00     12.00
 441     11         I    10.00      0.00
 441     12         I    10.00      8.00
 455     13         I    10.00     10.00
 455     14         P    12.00     12.00
 455     15         I    10.00      0.00
 667     16         I    10.00     10.00
 667     17         I    10.00     10.00
 667     18         P    12.00     12.00

Answer 1

我会用常见的 table 表达式递归地做。

此致彼得

DECLARE @PData TABLE (
     CustNum        INT
    ,TransNum       INT
    ,InvType        NVARCHAR(1)
    ,InvAmt         DECIMAL(5,2)
    )

INSERT INTO @PData (
     CustNum
    ,TransNum
    ,InvType
    ,InvAmt
    )

    VALUES
 (124, 1,'P',2)
,(124, 2,'I',10)
,(124, 3,'I',10)
,(153, 4,'I',10)
,(153, 5,'P',2)
,(153, 6,'I',10)
,(324, 7,'I',10)
,(324, 8,'I',10)
,(324, 9,'P',2)
,(441,10,'P',12)
,(441,11,'I',10)
,(441,12,'I',10)
,(455,13,'I',10)
,(455,14,'P',12)
,(455,15,'I',10)
,(455,19,'I',10)
,(667,16,'I',10)
,(667,17,'I',10)
,(667,18,'P',12)

;WITH Data as (
    SELECT
        CustNum,
        TransNum,
        InvType,
        InvAmt, 
        ROW_NUMBER() OVER (PARTITION BY custNum ORDER BY Transnum ASC) row,
        CASE WHEN InvType = 'P' THEN cast(-1*InvAmt AS DECIMAL(5,2)) ELSE 0 END prepaidAmt
    FROM 
        @PData
), modified as(
    SELECT 
        CustNum,
        TransNum,
        InvType,
        InvAmt,
        prepaidAmt,
        row, 
        InvAmt  total
    FROM Data d1 
        WHERE row = 1
    UNION ALL
    SELECT      
        d2.CustNum,
        d2.TransNum,
        d2.InvType,
        d2.InvAmt,
        CASE 
        WHEN 
            d2.InvAmt+m.prepaidAmt  < 0 
        THEN  
            CAST (d2.InvAmt+m.prepaidAmt AS DECIMAL(5,2))  
        ELSE 
            CASE 
            WHEN 
                d2.invtype = 'P' 
            THEN 
                CAST(-1*d2.invamt AS DECIMAL(5,2)) 
            ELSE 
                0 
            END   
        END ,
        d2.row, 
        CASE 
        WHEN 
            d2.InvAmt+m.prepaidAmt <0 
        THEN 
            0 
        ELSE  
            CAST( d2.InvAmt+m.prepaidAmt AS DECIMAL(5,2))  
        END
    FROM Data d2
    JOIN modified m 
    ON 
        m.CustNum = d2.CustNum and 
        m.row = d2.row-1
)
SELECT 
    m.CustNum,
    m.TransNum,
    m.InvType,
    m.InvAmt,
    m.total
FROM modified m
ORDER BY 
    custnum, 
    transnum

有没有办法使用窗口函数来简化这个查询？

Is there a way to use windowing functions to simplify this query?

tsql

ssms-2014