SQL-日期时间转换不一致

Question

I am hoping someone can give me an explanation the the following issue involving converting a varchar to datetime.

The code below works:

SELECT
    mt.matter_code,
    ud.uds_type,
    ud.group_no,
    ud.ud_field##2,
    convert(datetime,ud.ud_field##2,103) hearingDate
FROM dbo.matdb mt
    INNER JOIN dbo.matdb_add_in ad
            ON mt.mt_int_code  = ad.mt_int_code
            AND ad.add_in_code = 'OUTAA'
    INNER JOIN dbo.uddetail ud
            ON convert(varchar,ad.mt_add_in_int_code) = ud.owner_code
            AND ud.parent_code = ad.add_in_code
            AND ud.po_type_char = 'A'
            AND ud.uds_type = 'LPR'
            --AND ud.uds_type IN (SELECT s FROM dbo.split(',','LPR'))
WHERE mt.mt_type   = 'MATA'
AND mt.matter_code = '118-1'
ORDER BY ud.ud_field##2 ASC;

However if we substitute the

'AND ud.uds_type = 'LPR'

FOR

'AND ud.uds_type IN (SELECT s FROM dbo.split(',','LPR'))'

I get the following conversion error message: 'Conversion failed when converting datetime from character string.'

Both lines of code return the EXACT same results when not using a conversion. However as soon as the conversion is introduced when using dbo.split() I get the error message mentioned above.

Am I missing some kind of reference or is this a bug of sorts?

EDIT: dbo.split function

USE [Vfile_Dev]
GO

/****** Object:  UserDefinedFunction [dbo].[Split]    Script Date: 02/18/2016 13:28:12 ******/
SET ANSI_NULLS ON
GO

SET QUOTED_IDENTIFIER ON
GO

CREATE FUNCTION [dbo].[Split] (@sep char(1), @s varchar(512))
RETURNS table
AS
RETURN (
    WITH Pieces(pn, start, stop) AS (
      SELECT 1, 1, CHARINDEX(@sep, @s)
      UNION ALL
      SELECT pn + 1, stop + 1, CHARINDEX(@sep, @s, stop + 1)
      FROM Pieces
      WHERE stop > 0
    )
    SELECT pn,
      SUBSTRING(@s, start, CASE WHEN stop > 0 THEN stop-start ELSE 512 END) AS s
    FROM Pieces
  )
GO

EDIT: If anyone is interested, I have provided a solution below which was made possible thanks to Ivan. Additional checks were made on ud.ud_field##2 to ensure that there were no blanks and the dbo.split results are now inserted into a table variable.

INSERT INTO @screenTable
SELECT s FROM dbo.split(',','LPR,COU');

SELECT
    mt.matter_code,
    ud.uds_type,
    ud.group_no,
    isnull(nullif(ud.ud_field##2,''),'01/01/1990') as hdNorm,
    convert(datetime, isnull(nullif(ud.ud_field##2,''),'1990-01-01 00:00:00.000')) AS hdConverted
FROM dbo.matdb mt
    INNER JOIN dbo.matdb_add_in ad
            ON mt.mt_int_code  = ad.mt_int_code
            AND ad.add_in_code = 'OUTAA'
    INNER JOIN dbo.uddetail ud
            ON convert(varchar, ad.mt_add_in_int_code) = ud.owner_code
            AND ud.parent_code = ad.add_in_code
            AND ud.uds_type IN (select * from @screenTable)
WHERE mt.mt_type   = 'MATA'
AND mt.matter_code = '118-1'
ORDER BY group_no ASC;

Answer 1

My guess is that ud.ud_field##2 contains more than just datetime data. ud.uds_type and one of the reasons you need the convert(datetime,ud.ud_field##2,103) . When you use a WHERE clause directly, the predicate is applied before the CONVERT function is called. This means SQL Server is only converting data that can be converted. When you use your split function, SQL Server is no longer able to do that.

If ud.ud_field##2 is being used to store multiple datatypes, then you have a design problem. There are a few different solutions (eg create multiple uddetail tables - one for each datatype, add columns to uddetail - one for each datatype, etc.).

Here's my example:

CREATE TABLE #DifferentData1
  (
     [DataType]           VARCHAR(20),-- This is our code for the type of data we have in Sometimes_datetime
     [Sometimes_datetime] VARCHAR(20)-- This is datetime or something else
  )

CREATE TABLE #DifferentData2
  (
     [DataType]        VARCHAR(20),-- This is our code for the type of data we have
     [Always_datetime] DATETIME,-- This is ALWAYS datetime
     [Other_data]      VARCHAR(50)
  )

INSERT INTO #DifferentData1
            ([DataType],
             [Sometimes_datetime])
VALUES      ( 'LPR',CONVERT(VARCHAR, GETDATE(), 103)),
            ( 'Almost LPR',CONVERT(VARCHAR, GETDATE(), 120)),
            ('Not LPR','Some other data type')

INSERT INTO #DifferentData2
            ([DataType],
             [Always_datetime],
             [Other_data])
VALUES      ( 'LPR',CONVERT(DATETIME, GETDATE(), 103),NULL),
            ( 'Almost LPR',NULL,CONVERT(VARCHAR, GETDATE(), 120)),
            ('Not LPR',NULL,'Some other data type')

-- Let's see what we have
SELECT [DataType],
       [Sometimes_datetime]
FROM   #DifferentData1

SELECT [DataType],
       [Always_datetime],
       [Other_data]
FROM   #DifferentData2

Now, we can filter with a simple WHERE clause and everything is fine for either table:

SELECT CONVERT(DATETIME, [Sometimes_datetime], 103)
FROM   #DifferentData1
WHERE  [DataType] = 'LPR'

SELECT CONVERT(DATETIME, [Always_datetime], 103)
FROM   #DifferentData2
WHERE  [DataType] = 'LPR'

This works for #DifferentData1 because the predicate (WHERE clause) is applied before the CONVERT . However, when using the split function, SQL Sever is attempting to CONVERT after the predicate is applied. This causes it to fail. #DifferentData2 will still work as intended:

-- This fails
SELECT CONVERT(DATETIME, [Sometimes_datetime], 103)
FROM   #DifferentData1
WHERE  [DataType] IN
       (SELECT [SplitData]
        FROM   [dbo].[SplitString]('LPR', ','))
OPTION (MERGE JOIN) 

-- This works!
SELECT CONVERT(DATETIME, [Always_datetime], 103)
FROM   #DifferentData2
WHERE  [DataType] IN
       (SELECT [SplitData]
        FROM   [dbo].[SplitString]('LPR', ','))
OPTION (MERGE JOIN) 

Note: I had to force a MERGE JOIN, because the predicate might be applied before the CONVERT if a LOOP JOIN is used. If your data set is small, it may work to simply force a LOOP JOIN, but this is NOT recommended as a general solution:

SELECT CONVERT(DATETIME, [Sometimes_datetime], 103)
FROM   #DifferentData1
WHERE  [DataType] IN
       (SELECT [SplitData]
        FROM   [dbo].[SplitString]('LPR', ','))
OPTION (LOOP JOIN)