Question

我有一个表， int 值被用作位域（每个位都是一个标志）。

现在我想用二进制操作（在我的例子中为OR）聚合它们，以便：

SELECT 1 AS bitfield
INTO #TABLE
UNION ALL SELECT 1 + 2 + 8 + 32
UNION ALL SELECT 2 + 128
UNION ALL SELECT 2 + 32

SELECT AND_AGGR(bitfield) -- Invalid because AND_AGGR doesn't exist
FROM #TABLE

DROP #TABLE

会产生值171

这样做的好方法是希望不需要很多|和MAX（但如果必须，你必须这样做）？

我自己使用MS SQL Server 2008，但其他服务器上的解决方案也很有用。

Answer 1

在MySQL和PostgreSQL上，您可以使用BIT_OR。

我认为SQL Server没有这个聚合函数。

如你所说，你可以用很多MAX和&执行此操作：

MAX(x & 1) + MAX(x & 2) + ... + MAX(x & 128)

Answer 2

如果您期望结果为171，那么您的意思是二进制OR而不是AND吗？

在任何情况下，此解决方案都会将值聚合到变量中：

SELECT 1 AS bitfield
INTO #TABLE
UNION ALL SELECT 1 + 2 + 8 + 32
UNION ALL SELECT 2 + 128
UNION ALL SELECT 2 + 32

DECLARE @i int = 0

SELECT @i = @i | bitfield
FROM #TABLE

SELECT @i

DROP TABLE  #table

如果您希望按其他字段对聚合进行分组，则可能无法满足您的要求。

在大桌子上也不太可能表现良好。

Answer 3

在 MS SQL Server 中

let i = 0, j = s.length - 1;

while (i < j) {
    [s[i], s[j]] = [s[j], s[i]];
    i++;
    j--;
}

return s;

// **Run time 120 ms and  memory 46B**

印刷品

DECLARE @agg  VARCHAR(MAX) = '0001,0011,0101,0101,0101'
SELECT CONVERT(binary(4), VALUE, 2) , VALUE  FROM STRING_SPLIT( @agg , ',')

DECLARE @sum AS BIGINT = 0
DECLARE @mul AS BIGINT = 0xffffffff
SELECT  @sum |= v
   , @mul &= v
FROM STRING_SPLIT( @agg , ',')
CROSS APPLY (VALUES (CONVERT(binary(4), VALUE, 2))) _(v)

PRINT FORMAT(@sum,'X8')
PRINT FORMAT(@mul,'X8')

你需要更复杂的词：

            VALUE
---------- ------------
0x00010000 0001
0x00110000 0011
0x01010000 0101
0x01010000 0101
0x01010000 0101

01110000
00010000

使用付款期位图时

CREATE OR ALTER FUNCTION dbo.BOR( @agg VARCHAR(MAX))
RETURNS BIGINT
AS 
BEGIN
DECLARE @sum AS BIGINT = 0
SELECT  @sum |= CONVERT(BIGINT, VALUE)
FROM STRING_SPLIT( @agg , ',')
RETURN @sum
END

GO
CREATE OR ALTER FUNCTION dbo.BAND( @agg VARCHAR(MAX))
RETURNS BIGINT
AS 
BEGIN
DECLARE @mul AS BIGINT = 0xffffffffffffffff
SELECT  @mul &= CONVERT(BIGINT, VALUE)
FROM STRING_SPLIT( @agg , ',')
RETURN @mul
END
GO

但有时只需要一个人思考，你就可以用 SUM 做到这一点

;WITH delayedPayment AS
(SELECT * FROM ( VALUES 
    ( 123, 67, '2020-2-1')
   ,( 123, 67, '2020-4-1')
   ,( 123, 67, '2020-5-1')
   ,( 123, 67, '2020-6-1')
   ,( 123, 68, '2020-6-1')  -- another agreement
   ,( 123, 67, '2020-12-1')
           
   ,( 456, 69, '2020-4-1')
   ,( 456, 69, '2020-8-1')
   ,( 456, 69, '2020-10-1')
   ,( 456, 69, '2020-11-1')) _(cuno, loan, missedDuedate)
)
, delayPattern AS
(SELECT cuno
   ,  sum_months
   ,  bor_months
   ,  IIF( FORMAT( CAST(bor_months AS BIGINT), 'X16') LIKE '%111%', 'dalyad 3+ month in row', NULL) delayState
   FROM (SELECT cuno
         , SUM(POWER( 16.0, CONVERT( BIGINT, DATEDIFF( month, missedDuedate, '2020-12-1')))) sum_months
         , dbo.BOR( STRING_AGG( CONVERT( BIGINT, POWER( 16.0, DATEDIFF( month, missedDuedate, '2020-12-1'))),',')) bor_months
      FROM delayedPayment
      GROUP BY cuno
   ) s
)
SELECT cuno
   ,  FORMAT( CAST(sum_months AS BIGINT), 'X16') sum_months
   ,  FORMAT( CAST(bor_months AS BIGINT), 'X16') bor_months
   ,  delayState
FROM delayPattern

cuno    sum_months          bor_months          delayState
123     00000*10112*000001  00000*10111*000001  dalyad 3+ month in row
456     0000000100010110    0000000100010110    NULL

将打印

, delayPattern AS -- optimal
(SELECT cuno
   ,  bor_months
   ,  IIF( FORMAT( CAST(bor_months AS BIGINT), 'X16') LIKE '%111%', 'dalyad 3+ month in row', NULL) delayState
   FROM (SELECT cuno
         , SUM(POWER( 16.0, missedmonth)) bor_months
      FROM ( SELECT DISTINCT cuno
               , missedmonth
            FROM delayedPayment
            CROSS APPLY (VALUES ( DATEDIFF( month, missedDuedate, '2020-12-1'))) _(missedmonth)
            GROUP BY cuno, missedmonth
            ) ss
      GROUP BY cuno
   ) s
)

SELECT cuno
   ,  FORMAT( CAST(bor_months AS BIGINT), 'X16') bor_months
   ,  delayState
FROM delayPattern

注意：我正在使用十六进制格式和 POWER(16.0, X) ，只是为了偷懒， POWER(2.0, X) 将是正确的，但是您需要 bin->string formatter。像这样：

cuno    bor_months  delayState
123 0000010111000001    dalyad 3+ month in row
456 0000000100010110    NULL

使用二进制OR聚合位域值

3 个答案: