输出子句添加了不在原始源中的额外字符

Question

使用 OUTPUT 子句插入临时表后，我得到了一个不在源中的额外字符。

为什么？

--

例如，在从现有 PhysicalTable_1 表记录插入 LName = ' John ' 后，目标 PhysicalTable_1 表以及 #Temp 表具有 ' ?John ' 或 ' I?John ' 它有时会出现在 Lname 中，有时会出现在 FName 或 Email 以及其他领域。

PhysicalTable_1 中的数据示例 - FName = '​Raul' 插入后的目标记录看起来像 = '?Raul'

我正在使用这个：

 CREATE TABLE #Temp 
    (
    ID INT NOT NULL,
    LName VARCHAR(75) NULL,
    FName VARCHAR(75) NULL,
    Email VARCHAR(125) NULL
    )
    
    CREATE TABLE PhysicalTable_2
    (
    ID INT NOT NULL,
    LName VARCHAR(75) NULL,
    FName VARCHAR(75) NULL,
    Email VARCHAR(125) NUL
    )
    
 CREATE TABLE PhysicalTable_1
    (
    ID INT NOT NULL,
    LName NVARCHAR(500) NULL,
    FName NVARCHAR(500) NULL,
    Email NVARCHAR(500) NULL
    )

    INSERT INTO PhysicalTable_2
    (
      LName, FName, Email
    )
    OUTPUT INSERTED.LName, INSERTED.FName, INSERTED.Email
    INTO #Temp
    
    SELECT LName, FName, Email
    FROM PhysicalTable_1

我还尝试将#Temp 表的所有字符串字段数据类型更改为 NVARCHAR。 目的地中的一些记录最终还是有多余的字符

Answer 1

问题是您的 PhysicalTable_1 在 LName 中包含不可打印的 unicode 字符。 您将 Table1 的 unicode LName NVARCHAR 列插入到 Table2 的 ascii/nonunicode LName VARCHAR 列中。 Nonunicode 是 sql server 中 unicode 大小的一半，一些字节必须被“截断”，并且由于大小减小，不可打印的字符变得明显。

--characters to binary
SELECT CAST(N'P' AS VARBINARY(10)) AS UnicodeP, CAST('P' AS VARBINARY(10)) AS AsciiP --unicode is double the size of ascii


CREATE TABLE #temp(UnicodeP NVARCHAR(10), AsciiP VARCHAR(10));

INSERT INTO #temp(UnicodeP, AsciiP) VALUES (N'P', 'P'); --nothing special, normal insertion
INSERT INTO #temp(UnicodeP, AsciiP) VALUES ('P', N'P'); --omitting the N for unicode and using N for ascii, still works ok, implicit conversion

SELECT * FROM #temp;

--use binary from the very first SELECT CAST(....
INSERT INTO #temp(UnicodeP, AsciiP) VALUES (0x5000, 0x50); --still fine
SELECT * FROM #temp;

--prepend a nonprintable character (BOM) to unicode P, just insert into the UnicodeP only
INSERT INTO #temp(UnicodeP, AsciiP) VALUES (0xFEFF + 0x5000, NULL); --still fine
SELECT * FROM #temp;

--if you copy and paste the last UnicodeP, where AsciiP is NULL, you will not notice any visual difference


--update the ascii from unicode , where ascii is null
UPDATE #temp
SET AsciiP = UnicodeP --implicit conversion, ascii is half the unicode, some bytes have to go away
WHERE AsciiP IS NULL;

--since unicode was implicitly converted to ascii,  some bytes are "stripped out"  The nonprintable 0xFEFF needs to be "cut in half" and it becomes an unidentified char
SELECT UnicodeP, CAST(UnicodeP AS VARBINARY(10)) AS UnicodePbinary, AsciiP, CAST(AsciiP AS VARBINARY(10)) as AsciiPbinary
FROM #temp;


DROP TABLE #temp;

*编辑，隐式 unicode 到 nonunicode 和 asciiOrnothing

SELECT NCHAR(rownum) AS TheChar, CAST(NCHAR(rownum) AS CHAR(1)) AS ImplicitConversion, 
    CASE WHEN NCHAR(rownum) < N'Ā' collate Latin1_General_BIN2 THEN NCHAR(rownum) ELSE '' END AS AsciiOrNothing,
    UNICODE(NCHAR(rownum)) AS CharInteger,
    --or
    CASE WHEN UNICODE(/*TheChar*/ NCHAR(rownum)) <= 255 THEN NCHAR(rownum) ELSE '' END AS AsciiOrNothing2
FROM 
(
SELECT ROW_NUMBER() OVER(ORDER BY (SELECT null)) AS rownum
FROM (
    --10K
    SELECT TOP (100) name from master.dbo.spt_values) AS a
    CROSS JOIN (SELECT TOP (100) name from master.dbo.spt_values) AS b
) AS src
ORDER BY rownum