Converting an NVARCHAR string to a UTF-8 VARBINARY data block in pure Transact-SQL

ToUTF8.sql

create function [dbo].[ToUTF8](@s nvarchar(max))
returns varbinary(max)
as
begin
    declare @i int = 1, @n int = datalength(@s)/2, @r varbinary(max) = 0x, @c int, @c2 int, @d varbinary(4)
    while @i <= @n
    begin
        set @c = unicode(substring(@s, @i, 1))
        if (@c & 0xFC00) = 0xD800
        begin
            set @i += 1
            if @i > @n
                return cast(cast('Malformed UTF-16 - two nchar sequence cut short' as int) as varbinary)
            set @c2 = unicode(substring(@s, @i, 1))
            if (@c2 & 0xFC00) <> 0xDC00
                return cast(cast('Malformed UTF-16 - continuation missing in a two nchar sequence' as int) as varbinary)
            set @c = (((@c & 0x3FF) * 0x400) | (@c2 & 0x3FF)) + 0x10000
        end

        if @c < 0x80
            set @d = cast(@c as binary(1))
        if @c >= 0x80 and @c < 0x800 
            set @d = cast(((@c * 4) & 0xFF00) | (@c & 0x3F) | 0xC080 as binary(2))
        if @c >= 0x800 and @c < 0x10000
            set @d = cast(((@c * 0x10) & 0xFF0000) | ((@c * 4) & 0x3F00) | (@c & 0x3F) | 0xe08080 as binary(3))
        if @c >= 0x10000
            set @d = cast(((@c * 0x40) & 0xFF000000) | ((@c * 0x10) & 0x3F0000) | ((@c * 4) & 0x3F00) | (@c & 0x3F) | 0xf0808080 as binary(4))
            
        set @r += @d
        set @i += 1
    end
    return @r
end

Unlike the GitHub comment system, it works correctly with characters from Unicode's astral planes.

Hi Sevaa,
I worked in AP team needs to deal with Asia Characters.
I have tested your SQL function with few Korea and Chinese characters find out the encoding does not meet the internet URL encoding standard.

The left UTF8 is from standard URL encoding, the right one is from the function
〇 E38087 E3C087
ㄱ E384B1 E3C4B1
validated with http://memory.loc.gov/diglib/codetables/9.3.html

There seems to be a pattern of the wrong encoding on the second byte 80 -> C0 and 84 -> C4.

I am not familiar bitwise and hex calculation at all. I really hope to fix it. Will you be quickly discover why?

I found this function works correctly :)
https://stackoverflow.com/questions/35200452/compute-md5-hash-of-a-utf8-string/35289890#35289890

Sorry didn't get to this earlier, only found this comment today. Fixed now, a mask was off in the 3- and 4-byte encoding.

Truly a hidden gem. Many thanks for this!

Here's a gist for the converse operation - from a UTF-8 VARBINARY to a NVARCHAR string: https://gist.github.com/sevaa/5a50ba98762762ae960b87fb148a7df2

Works perfectly 👍

Isn't the handling of surrogate pairs incorrect here? According to https://en.wikipedia.org/wiki/UTF-16#Code_points_from_U+010000_to_U+10FFFF , 0x10000 should be added to the combination produced from two surrogate code units, rather than OR'ed with this combination.

Also, this code doesn't check that the second code unit is actually surrogate (or whether it exists at all), so it can produce unexpected/broken results if unpaired surrogate code units are encountered.

I've added some checks for malformed UTF-16 surrogate pairs. The function will now throw a reasonably descriptive error instead of producing garbage outputs.

Since MSSQL 2019, there is native support for UTF-8. The following will do the same job, most likely much faster:

declare @s nvarchar(100) = N'Москва 東京 🙃', @t varchar(200) 
set @t = CAST(@s AS varchar) collate Latin1_General_100_CI_AI_SC_UTF8
print @t

The most performant way I could find for MSSQL 2012+, MSSQL 2019-
Works much much faster for long strings

CREATE FUNCTION fn_to_utf8_bytes_4000(@s NVARCHAR(4000))
RETURNS VARBINARY(8000)
AS
BEGIN
	DECLARE @multiByteChar NVARCHAR(150) = N'%[^' + NCHAR(1) + '-' + NCHAR(127) + N']%'
	DECLARE @index INT = PATINDEX(@multibytechar COLLATE Latin1_General_BIN, @s);
	IF @index = 0
		RETURN CAST(CAST(@s AS VARCHAR(8000)) AS VARBINARY(8000))

	DECLARE @result VARBINARY(8000) = CAST(CAST(SUBSTRING(@s, 1, @index - 1) AS VARCHAR(8000)) AS VARBINARY(8000))

	SET @s = SUBSTRING(@s, @index, 4000)
	DECLARE @c INT = UNICODE(@s COLLATE Latin1_General_100_CS_AS_KS_WS_SC)

	WHILE @c > 0x80
	BEGIN
multibyte:
		IF @c >= 0x10000
		BEGIN
			SET @result += CAST(((@c * 0x40) & 0xFF000000) | ((@c * 0x10) & 0x3F0000) | ((@c * 4) & 0x3F00) | (@c & 0x3F) | 0xf0808080 AS VARBINARY(4))
			SET @s = SUBSTRING(@s, 3, 4000)
		END
		ELSE
		BEGIN
			SET @result += CASE
				WHEN @c < 0x800 THEN CAST(((@c * 4) & 0xFF00) | (@c & 0x3F) | 0xC080 AS VARBINARY(2))
				ELSE CAST(((@c * 0x10) & 0xFF0000) | ((@c * 4) & 0x3F00) | (@c & 0x3F) | 0xe08080 AS VARBINARY(3))
			END
			SET @s = SUBSTRING(@s, 2, 4000)
		END
		SET @c = UNICODE(@s COLLATE Latin1_General_100_CS_AS_KS_WS_SC)
	END

	SET @index = PATINDEX(@multibytechar COLLATE Latin1_General_BIN, @s);
	IF @index = 0
		RETURN @result + CAST(CAST(@s AS VARCHAR(8000)) AS VARBINARY(8000))

	SET @result += CAST(CAST(SUBSTRING(@s, 1, @index - 1) AS VARCHAR(8000)) AS VARBINARY(8000))
	SET @s = SUBSTRING(@s, @index, 4000)
	SET @c = UNICODE(@s COLLATE Latin1_General_100_CS_AS_KS_WS_SC)
	GOTO multibyte

	RETURN @result;
END

CREATE FUNCTION fn_to_utf8_bytes_max(@s NVARCHAR(MAX))
RETURNS VARBINARY(MAX)
AS
BEGIN
	DECLARE @size INT = LEN(@s)
	DECLARE @halfsize INT

	IF @size <= 4000
		RETURN dbo.fn_to_utf8_bytes_4000(@s);

	SET @halfsize = @size / 2
	SET @halfsize = IIF(UNICODE(SUBSTRING(@s, @halfsize, 1)) & 0xFC00 = 0xD800, @halfsize - 1, @halfsize)

	RETURN dbo.fn_to_utf8_bytes_max(SUBSTRING(@s, 1, @halfsize)) + dbo.fn_to_utf8_bytes_max(SUBSTRING(@s, @halfsize + 1, @size - @halfsize));
END

Thanks.

Thank you, it works perfectly, also for chinese chars, and other special chars.

I have to prepare data on a SQL 2016 to send it to the Salesforce API as BASE64.

When I used CAST(MyText AS VARBINARY(MAX)) AS MyText_VARBINARY, and then encoded that to BASE64 with CAST('' AS XML).value('xs:base64Binary(sql:column("MyText_VARBINARY"))', 'VARCHAR(MAX)'), I got a BASE64 which had UTF16-LE encoding.

With your function ToUTF8, I got the corrct encoding!