事实证明，即使没有真正的“脚本”扩展，如PL/SQL或PSM(它们被设计成真正的编程语言，所以这有点作弊)，SQL也可以是图灵完备的。

在这组幻灯片中，Andrew Gierth通过构造一个循环标记系统证明了CTE和Windowing SQL是图灵完备的，这个循环标记系统已被证明是图灵完备的。然而，CTE特性是重要的部分——它允许您创建可以引用自己的命名子表达式，从而递归地解决问题。

值得注意的有趣的事情是，添加CTE并不是为了将SQL转换为编程语言——只是为了将声明式查询语言转换为更强大的声明式查询语言。有点像c++，它的模板被证明是图灵完备的，尽管它们并不打算创建一个元编程语言。

哦，SQL中Mandelbrot集合的例子也非常令人印象深刻:)

严格地说，SQL现在是一种图灵完备的语言，因为最新的SQL标准包含了“持久存储模块”(psm)。简而言之，PSM是Oracle中PL/SQL语言的标准版本(以及当前DBMS的其他类似过程扩展)。

随着这些psm的加入，SQL变得图灵完备

https://web.archive.org/web/20110807062050/http://channel9.msdn.com/forums/TechOff/431432-SQL-Turing-Completeness-question

是关于这个话题的讨论。引用:

SQL as such (i.e. the SQL92 standard) is not turing complete. However, many of the languages derived from SQL, such as Oracle's PL/SQL and SQL Server's T-SQL and others are turing complete. PL/SQL and T-SQL certainly qualify as programming languages, whether SQL92 itself qualifies is open for debate. Some people claim that any piece of code that tells a computer what to do qualifies as a programming language; by that definition SQL92 is one, but so is e.g. HTML. The definition is rather vague, and it's imo a pointless thing to argue about.

如果能证明一种编程语言在计算上等价于图灵机，那么它就是图灵完备的。

TSQL是图灵完备的，因为我们可以在TSQL中创建一个BrainFuck解释器。

BrainFuck解释器在SQL - GitHub

所提供的代码在内存中工作，不修改数据库。

-- Brain Fuck interpreter in SQL

DECLARE @Code  VARCHAR(MAX) = ', [>,] < [.<]'
DECLARE @Input VARCHAR(MAX) = '!dlroW olleH';

-- Creates a "BrainFuck" DataBase.
-- CREATE DATABASE BrainFuck;

-- Creates the Source code table
DECLARE @CodeTable TABLE (
    [Id]      INT IDENTITY(1,1) PRIMARY KEY NOT NULL,
    [Command] CHAR(1) NOT NULL
);

-- Populate the source code into CodeTable
DECLARE @CodeLen INT = LEN(@Code);
DECLARE @CodePos INT = 0;
DECLARE @CodeChar CHAR(1);

WHILE @CodePos < @CodeLen
BEGIN
    SET @CodePos  = @CodePos + 1;
    SET @CodeChar = SUBSTRING(@Code, @CodePos, 1);
    IF @CodeChar IN ('+', '-', '>', '<', ',', '.', '[', ']')
        INSERT INTO @CodeTable ([Command]) VALUES (@CodeChar)
END

-- Creates the Input table
DECLARE @InputTable TABLE (
    [Id]   INT IDENTITY(1,1) PRIMARY KEY NOT NULL,
    [Char] CHAR(1) NOT NULL
);

-- Populate the input text into InputTable
DECLARE @InputLen INT = LEN(@Input);
DECLARE @InputPos INT = 0;

WHILE @InputPos < @InputLen
BEGIN
    SET @InputPos = @InputPos + 1;
    INSERT INTO @InputTable ([Char])
    VALUES (SUBSTRING(@Input, @InputPos, 1))
END

-- Creates the Output table
DECLARE @OutputTable TABLE (
    [Id]   INT IDENTITY(1,1) PRIMARY KEY NOT NULL,
    [Char] CHAR(1) NOT NULL
);

-- Creates the Buffer table
DECLARE @BufferTable TABLE (
    [Id]     INT IDENTITY(1,1) PRIMARY KEY NOT NULL,
    [Memory] INT DEFAULT 0  NOT NULL
);
INSERT INTO @BufferTable ([Memory])
VALUES (0);

-- Initialization of temporary variables 
DECLARE @CodeLength INT = (SELECT COUNT(*) FROM @CodeTable);
DECLARE @CodeIndex  INT = 0;
DECLARE @Pointer    INT = 1;
DECLARE @InputIndex INT = 0;
DECLARE @Command    CHAR(1);
DECLARE @Depth      INT;

-- Main calculation cycle
WHILE @CodeIndex < @CodeLength
BEGIN
    -- Read the next command.
    SET @CodeIndex = @CodeIndex + 1;
    SET @Command = (SELECT [Command] FROM @CodeTable WHERE [Id] = @CodeIndex);

    -- Increment the pointer.
    IF @Command = '>'
    BEGIN
        SET @Pointer = @Pointer + 1;
        IF (SELECT [Id] FROM @BufferTable WHERE [Id] = @Pointer) IS NULL
            INSERT INTO @BufferTable ([Memory]) VALUES (0);
    END

    -- Decrement the pointer.
    ELSE IF @Command = '<'
        SET @Pointer = @Pointer - 1;

    -- Increment the byte at the pointer.
    ELSE IF @Command = '+'
        UPDATE @BufferTable SET [Memory] = [Memory] + 1 WHERE [Id] = @Pointer;

    -- Decrement the byte at the pointer.
    ELSE IF @Command = '-'
        UPDATE @BufferTable SET [Memory] = [Memory] - 1 WHERE [Id] = @Pointer;

    -- Output the byte at the pointer.
    ELSE IF @Command = '.'
        INSERT INTO @OutputTable ([Char]) (SELECT CHAR([Memory]) FROM @BufferTable WHERE [Id] = @Pointer);

    -- Input a byte and store it in the byte at the pointer.
    ELSE IF @Command = ','
    BEGIN
        SET @InputIndex = @InputIndex + 1;
        UPDATE @BufferTable SET [Memory] = COALESCE((SELECT ASCII([Char]) FROM @InputTable WHERE [Id] = @InputIndex), 0) WHERE [Id] = @Pointer;
    END

    -- Jump forward past the matching ] if the byte at the pointer is zero.
    ELSE IF @Command = '[' AND COALESCE((SELECT [Memory] FROM @BufferTable WHERE [Id] = @Pointer), 0) = 0
    BEGIN
        SET @Depth = 1;
        WHILE @Depth > 0
        BEGIN
            SET @CodeIndex = @CodeIndex + 1;
            SET @Command = (SELECT [Command] FROM @CodeTable WHERE [Id] = @CodeIndex);
            IF @Command = '[' SET @Depth = @Depth + 1;
            ELSE IF @Command = ']' SET @Depth = @Depth - 1;
        END
    END

    -- Jump backwards to the matching [ unless the byte at the pointer is zero.
    ELSE IF @Command = ']' AND COALESCE((SELECT [Memory] FROM @BufferTable WHERE [Id] = @Pointer), 0) != 0
    BEGIN
        SET @Depth = 1;
        WHILE @Depth > 0
        BEGIN
            SET @CodeIndex = @CodeIndex - 1;
            SET @Command = (SELECT [Command] FROM @CodeTable WHERE [Id] = @CodeIndex);
            IF @Command = ']' SET @Depth = @Depth + 1;
            ELSE IF @Command = '[' SET @Depth = @Depth - 1;
        END
    END
END;

-- Collects and prints the output
DECLARE @Output VARCHAR(MAX);
SELECT @Output = COALESCE(@Output, '') + [Char]
FROM @OutputTable;

PRINT @Output;
Go

最初在SQL-86中定义的ANSI选择语句不是图灵完全语句，因为它总是终止(除了递归cte，并且只有在实现支持任意深度递归时)。因此，不可能模拟任何其他图灵机。存储过程是图灵完整的，但这是作弊;-)

事实证明，即使没有真正的“脚本”扩展，如PL/SQL或PSM(它们被设计成真正的编程语言，所以这有点作弊)，SQL也可以是图灵完备的。

在这组幻灯片中，Andrew Gierth通过构造一个循环标记系统证明了CTE和Windowing SQL是图灵完备的，这个循环标记系统已被证明是图灵完备的。然而，CTE特性是重要的部分——它允许您创建可以引用自己的命名子表达式，从而递归地解决问题。

值得注意的有趣的事情是，添加CTE并不是为了将SQL转换为编程语言——只是为了将声明式查询语言转换为更强大的声明式查询语言。有点像c++，它的模板被证明是图灵完备的，尽管它们并不打算创建一个元编程语言。

哦，SQL中Mandelbrot集合的例子也非常令人印象深刻:)