Question

我有一个SQL表，我在其中存储必须唯一的大字符串值。为了确保唯一性，我在列上有一个唯一索引，我在其中存储大字符串的MD5哈希的字符串表示。

保存这些记录的C＃应用程序使用以下方法进行散列：

public static string CreateMd5HashString(byte[] input)
{
    var hashBytes = MD5.Create().ComputeHash(input);
    return string.Join("", hashBytes.Select(b => b.ToString("X")));
}

为了调用此功能，我首先使用UTF-8编码将string转换为byte[]：

// this is what I use in my app
CreateMd5HashString(Encoding.UTF8.GetBytes("abc"))
// result: 90150983CD24FB0D6963F7D28E17F72

现在我希望能够使用HASHBYTES function在SQL中实现此哈希函数，但我得到了不同的值：

print hashbytes('md5', N'abc')
-- result: 0xCE1473CF80C6B3FDA8E3DFC006ADC315

这是因为SQL计算字符串的UTF-16表示的MD5。如果我CreateMd5HashString(Encoding.Unicode.GetBytes("abc"))，我会在C＃中得到相同的结果。

我无法改变在应用程序中完成散列的方式。

有没有办法让SQL Server计算字符串的UTF-8字节的MD5哈希？

我查了类似的问题，我尝试使用整理，但到目前为止没有运气。

Answer 1

您需要创建一个UDF，以便将UTARCHAR数据转换为UTF-8表示形式的字节。假设它被称为dbo.NCharToUTF8Binary，那么你可以这样做：

hashbytes('md5', dbo.NCharToUTF8Binary(N'abc', 1))

这是一个UDF，它会这样做：

create function dbo.NCharToUTF8Binary(@txt NVARCHAR(max), @modified bit)
returns varbinary(max)
as
begin
-- Note: This is not the fastest possible routine. 
-- If you want a fast routine, use SQLCLR
    set @modified = isnull(@modified, 0)
    -- First shred into a table.
    declare @chars table (
    ix int identity primary key,
    codepoint int,
    utf8 varbinary(6)
    )
    declare @ix int
    set @ix = 0
    while @ix < datalength(@txt)/2  -- trailing spaces
    begin
        set @ix = @ix + 1
        insert @chars(codepoint)
        select unicode(substring(@txt, @ix, 1))
    end

    -- Now look for surrogate pairs.
    -- If we find a pair (lead followed by trail) we will pair them
    -- High surrogate is \uD800 to \uDBFF
    -- Low surrogate  is \uDC00 to \uDFFF
    -- Look for high surrogate followed by low surrogate and update the codepoint   
    update c1 set codepoint = ((c1.codepoint & 0x07ff) * 0x0800) + (c2.codepoint & 0x07ff) + 0x10000
    from @chars c1 inner join @chars c2 on c1.ix = c2.ix -1
    where c1.codepoint >= 0xD800 and c1.codepoint <=0xDBFF
    and c2.codepoint >= 0xDC00 and c2.codepoint <=0xDFFF
    -- Get rid of the trailing half of the pair where found
    delete c2 
    from @chars c1 inner join @chars c2 on c1.ix = c2.ix -1
    where c1.codepoint >= 0x10000

    -- Now we utf-8 encode each codepoint.
    -- Lone surrogate halves will still be here
    -- so they will be encoded as if they were not surrogate pairs.
    update c 
    set utf8 = 
    case 
    -- One-byte encodings (modified UTF8 outputs zero as a two-byte encoding)
    when codepoint <= 0x7f and (@modified = 0 OR codepoint <> 0)
    then cast(substring(cast(codepoint as binary(4)), 4, 1) as varbinary(6))
    -- Two-byte encodings
    when codepoint <= 0x07ff
    then substring(cast((0x00C0 + ((codepoint/0x40) & 0x1f)) as binary(4)),4,1)
    + substring(cast((0x0080 + (codepoint & 0x3f)) as binary(4)),4,1)
    -- Three-byte encodings
    when codepoint <= 0x0ffff
    then substring(cast((0x00E0 + ((codepoint/0x1000) & 0x0f)) as binary(4)),4,1)
    + substring(cast((0x0080 + ((codepoint/0x40) & 0x3f)) as binary(4)),4,1)
    + substring(cast((0x0080 + (codepoint & 0x3f)) as binary(4)),4,1)
    -- Four-byte encodings 
    when codepoint <= 0x1FFFFF
    then substring(cast((0x00F0 + ((codepoint/0x00040000) & 0x07)) as binary(4)),4,1)
    + substring(cast((0x0080 + ((codepoint/0x1000) & 0x3f)) as binary(4)),4,1)
    + substring(cast((0x0080 + ((codepoint/0x40) & 0x3f)) as binary(4)),4,1)
    + substring(cast((0x0080 + (codepoint & 0x3f)) as binary(4)),4,1)

    end
    from @chars c

    -- Finally concatenate them all and return.
    declare @ret varbinary(max)
    set @ret = cast('' as varbinary(max))
    select @ret = @ret + utf8 from @chars c order by ix
    return  @ret

end

Answer 2

SQL Server本身不支持使用UTF-8字符串和it hasn't for quite a while。如您所见，NCHAR and NVARCHAR use UCS-2 rather than UTF-8。

如果您坚持使用HASHBYTES功能，则必须能够从C＃代码传递UTF-8 byte[]作为VARBINARY以保留编码。 HASHBYTES accepts VARBINARY in place of NVARCHAR.这可以通过接受NVARCHAR的CLR函数完成，并将Encoding.UTF8.GetBytes的结果作为VARBINARY返回。

话虽如此，我强烈建议在应用程序而不是数据库中保持这些类型的业务规则。特别是因为应用程序已经在执行此逻辑。

Answer 3

SELECT HashBytes('MD5', CAST (N'中文' COLLATE Latin1_General_100_CI_AI_SC_UTF8 AS varchar(4000)))

仅在sql server 2019上

参考： https://www.mssqltips.com/sqlservertip/6168/impact-of-utf8-support-in-sql-server-2019/

计算UTF8字符串的MD5哈希值

3 个答案: