Perfectly parsing SQL requires only ten simple steps

Many programmers regard SQL as a scourge. SQL is one of the few declarative languages ​​that operates completely differently from the command line languages, object-oriented programming languages, and even functional languages ​​we are familiar with (although some people believe that the SQL language is also a functional language ).

We write SQL every day and apply it in the open source software jOOQ. So I want to introduce the beauty of SQL to those friends who still have headaches about it, so this article is specially written for the following readers:

1. I will use SQL at work but don’t fully understand it. people.

2. People who are proficient in using SQL but do not understand its syntax and logic.

3. People who want to teach others SQL.

This article focuses on the SELECT sentence pattern, and other DML (Data Manipulation Language data manipulation language commands) will be introduced in other articles.

1. SQL is a declarative language

First of all, we must keep this concept in mind: "declaration". The SQL language is an example of declaring for the computer what kind of results you want to get from the raw data, rather than telling the computer how to get the results. Isn't this great?

(Translator's Note: Simply put, the SQL language declares the properties of the result set. The computer will select the data that meets the declaration from the database based on what the SQL declares, rather than traditional programming thinking. to instruct the computer how to operate. )

SELECT first_name, last_name FROM employees WHERE salary > 100000

The above example is easy to understand, we don’t care about these employee records from Where does it come from? All we need is data on those with high salaries (Translator's Note: salary>100000).

Where did we learn this?

If the SQL language is so simple, then what makes people "turn pale after hearing SQL"? The main reason is that we subconsciously think about problems according to the imperative programming way of thinking. It's like this: "Computer, do this step first, then that step, but before that check whether condition A and condition B are met." For example, using variables to pass parameters, using loop statements, iteration, calling functions, etc. are all thinking habits of this imperative programming.

2. SQL syntax is not executed in grammatical order

SQL statements have a feature that confuses most people, that is: the execution order of SQL statements is related to the grammatical order of its statements. Not consistent. The syntax sequence of SQL statements is:

• SELECT[DISTINCT]

• FROM

• WHERE

• GROUP BY

• HAVING

• UNION

• ORDER BY

For the convenience of understanding, not all SQL syntax structures are listed above, but it is enough to show that the syntax order of SQL statements is completely different from its execution order. Taking the above statement as an example, its execution order is:

• FROM

• WHERE

• ##GROUP BY

• HAVING

• SELECT

• DISTINCT

• UNION

• ORDER BY

Regarding the execution order of SQL statements, there are three things worthy of our attention:

1. FROM is the first step in SQL statement execution, not SELECT. The first step for a database to execute a SQL statement is to load data from the hard disk into the data buffer so that it can be operated on. (Translator's Note: The original text is "The first thing that happens is loading data from the disk into memory, in order to operate on such data.", but this is not the case. Taking common databases such as Oracle as an example, the data is extracted from the hard disk. Go to the data buffer to perform operations.)

2. SELECT is executed after most statements are executed. Strictly speaking, it is executed after FROM and GROUP BY. It is very important to understand this, which is why you cannot use a field in WHERE that is aliased in SELECT as a condition.

SELECT A.x + A.y AS z
FROM A
WHERE z = 10 -- z 在此处不可用，因为SELECT是最后执行的语句！

If you want to reuse alias z, you have two options. Either rewrite the expression represented by z:

SELECT A.x + A.y AS z
FROM AWHERE (A.x + A.y) = 10

...or resort to derived tables, common data expressions, or views to avoid alias reuse. See the examples below.

3. Whether in syntax or execution order, UNION is always ranked before ORDER BY. Many people think that every UNION segment can be sorted using ORDER BY, but according to the SQL language standard and the implementation differences of SQL in each database, this is not true. Although some databases allow SQL statements to sort subqueries or derived tables, this does not mean that the sorting will maintain the sorted order after the UNION operation.

Note: Not all databases use the same parsing method for SQL statements. For example, MySQL, PostgreSQL and SQLite will not perform as mentioned in the second point above.

What have we learned?

既然并不是所有的数据库都按照上述方式执行 SQL 预计，那我们的收获是什么？我们的收获是永远要记得： SQL 语句的语法顺序和其执行顺序并不一致，这样我们就能避免一般性的错误。如果你能记住 SQL 语句语法顺序和执行顺序的差异，你就能很容易的理解一些很常见的 SQL 问题。

当然，如果一种语言被设计成语法顺序直接反应其语句的执行顺序，那么这种语言对程序员是十分友好的，这种编程语言层面的设计理念已经被微软应用到了 LINQ 语言中。

3、 SQL 语言的核心是对表的引用（table references）

由于 SQL 语句语法顺序和执行顺序的不同，很多同学会认为SELECT 中的字段信息是 SQL 语句的核心。其实真正的核心在于对表的引用。

根据 SQL 标准，FROM 语句被定义为：

44dc542b5d74380f2cd0250ba54efa49 ::= FROM de33f891b201ddbdecd12e0f3bc36d8b [ { ccb011809fb18504b450f66338ebdd64 de33f891b201ddbdecd12e0f3bc36d8b }... ]

FROM 语句的“输出”是一张联合表，来自于所有引用的表在某一维度上的联合。我们们慢慢来分析：

FROM a, b

上面这句 FROM 语句的输出是一张联合表，联合了表 a 和表 b 。如果 a 表有三个字段， b 表有 5 个字段，那么这个“输出表”就有 8 （ =5+3）个字段。

这个联合表里的数据是 ab，即 a 和 b 的笛卡尔积。换句话说，也就是 a 表中的每一条数据都要跟 b 表中的每一条数据配对。如果 a 表有3 条数据， b 表有 5 条数据，那么联合表就会有 15 （ =53）条数据。

FROM 输出的结果被 WHERE 语句筛选后要经过 GROUP BY 语句处理，从而形成新的输出结果。我们后面还会再讨论这方面问题。

如果我们从集合论（关系代数）的角度来看，一张数据库的表就是一组数据元的关系，而每个 SQL 语句会改变一种或数种关系，从而产生出新的数据元的关系（即产生新的表）。

我们学到了什么？

思考问题的时候从表的角度来思考问题提，这样很容易理解数据如何在 SQL 语句的“流水线”上进行了什么样的变动。

4、 灵活引用表能使 SQL 语句变得更强大

灵活引用表能使 SQL 语句变得更强大。一个简单的例子就是 JOIN 的使用。严格的说 JOIN 语句并非是 SELECT 中的一部分，而是一种特殊的表引用语句。 SQL 语言标准中表的连接定义如下：

<table reference> ::=    <table name>
  | <derived table>
  | <joined table>

就拿之前的例子来说：

FROM a, b

a 可能输如下表的连接：

a1 JOIN a2 ON a1.id = a2.id

将它放到之前的例子中就变成了：

FROM a1 JOIN a2 ON a1.id = a2.id, b

尽管将一个连接表用逗号跟另一张表联合在一起并不是常用作法，但是你的确可以这么做。结果就是，最终输出的表就有了 a1+a2+b 个字段了。

（译者注：原文这里用词为 degree ，译为维度。如果把一张表视图化，我们可以想象每一张表都是由横纵两个维度组成的，横向维度即我们所说的字段或者列，英文为columns；纵向维度即代表了每条数据，英文为 record ，根据上下文，作者这里所指的应该是字段数。）

在 SQL 语句中派生表的应用甚至比表连接更加强大，下面我们就要讲到表连接。

我们学到了什么？

思考问题时，要从表引用的角度出发，这样就很容易理解数据是怎样被 SQL 语句处理的，并且能够帮助你理解那些复杂的表引用是做什么的。

更重要的是，要理解 JOIN 是构建连接表的关键词，并不是 SELECT 语句的一部分。有一些数据库允许在 INSERT 、 UPDATE 、 DELETE 中使用 JOIN 。

5、 SQL 语句中推荐使用表连接

我们先看看刚刚这句话：

FROM a, b

高级 SQL 程序员也许学会给你忠告：尽量不要使用逗号来代替 JOIN 进行表的连接，这样会提高你的 SQL 语句的可读性，并且可以避免一些错误。

利用逗号来简化 SQL 语句有时候会造成思维上的混乱，想一下下面的语句：

FROM a, b, c, d, e, f, g, h
WHERE a.a1 = b.bxAND a.a2 = c.c1AND d.d1 = b.bc
-- etc...

我们不难看出使用 JOIN 语句的好处在于：

• 安全。 JOIN 和要连接的表离得非常近，这样就能避免错误。

• 更多连接的方式，JOIN 语句能去区分出来外连接和内连接等。

我们学到了什么？

记着要尽量使用 JOIN 进行表的连接，永远不要在 FROM 后面使用逗号连接表。

6、 SQL 语句中不同的连接操作

SQL 语句中，表连接的方式从根本上分为五种：

• EQUI JOIN

• SEMI JOIN

• ANTI JOIN

• CROSS JOIN

• DIVISION

EQUI JOIN

这是一种最普通的 JOIN 操作，它包含两种连接方式：

• INNER JOIN（或者是 JOIN ）

• OUTER JOIN（包括： LEFT 、 RIGHT、 FULL OUTER JOIN）

用例子最容易说明其中区别：

-- This table reference contains authors and their books.
-- There is one record for each book and its author.
-- authors without books are NOT included
author JOIN book ON author.id = book.author_id

-- This table reference contains authors and their books
-- There is one record for each book and its author.
-- ... OR there is an "empty" record for authors without books
-- ("empty" meaning that all book columns are NULL)
author LEFT OUTER JOIN book ON author.id = book.author_id

SEMI JOIN

这种连接关系在 SQL 中有两种表现方式：使用 IN，或者使用 EXISTS。“ SEMI ”在拉丁文中是“半”的意思。这种连接方式是只连接目标表的一部分。这是什么意思呢？再想一下上面关于作者和书名的连接。我们想象一下这样的情况：我们不需要作者 / 书名这样的组合，只是需要那些在书名表中的书的作者信息。那我们就能这么写：

-- Using IN
FROM author
WHERE author.id IN (SELECT book.author_id FROM book)

-- Using EXISTS
FROM author
WHERE EXISTS (SELECT 1 FROM book WHERE book.author_id = author.id)

尽管没有严格的规定说明你何时应该使用 IN ，何时应该使用 EXISTS ，但是这些事情你还是应该知道的：

• IN比 EXISTS 的可读性更好

• EXISTS 比IN 的表达性更好（更适合复杂的语句）

• 二者之间性能没有差异（但对于某些数据库来说性能差异会非常大）

因为使用 INNER JOIN 也能得到书名表中书所对应的作者信息，所以很多初学者机会认为可以通过 DISTINCT 进行去重，然后将 SEMI JOIN 语句写成这样：

-- Find only those authors who also have books
SELECT DISTINCT first_name, last_name
FROM author
JOIN book ON author.id = book.author_id

这是一种很糟糕的写法，原因如下：

• SQL 语句性能低下：因为去重操作（ DISTINCT ）需要数据库重复从硬盘中读取数据到内存中。（译者注： DISTINCT 的确是一种很耗费资源的操作，但是每种数据库对于 DISTINCT 的操作方式可能不同）。

• 这么写并非完全正确：尽管也许现在这么写不会出现问题，但是随着 SQL 语句变得越来越复杂，你想要去重得到正确的结果就变得十分困难。

更多的关于滥用 DISTINCT 的危害可以参考这篇博文

（http://blog.jooq.org/2013/07/30/10-common-mistakes-java-developers-make-when-writing-sql/）。

ANTI JOIN

这种连接的关系跟 SEMI JOIN 刚好相反。在 IN 或者 EXISTS 前加一个 NOT 关键字就能使用这种连接。举个例子来说，我们列出书名表里没有书的作者：

-- Using IN
FROM author
WHERE author.id NOT IN (SELECT book.author_id FROM book)

-- Using EXISTS
FROM author
WHERE NOT EXISTS (SELECT 1 FROM book WHERE book.author_id = author.id)

关于性能、可读性、表达性等特性也完全可以参考 SEMI JOIN。

这篇博文介绍了在使用 NOT IN 时遇到 NULL 应该怎么办，因为有一点背离本篇主题，就不详细介绍，有兴趣的同学可以读一下

（http://blog.jooq.org/2012/01/27/sql-incompatibilities-not-in-and-null-values/）。

CROSS JOIN

这个连接过程就是两个连接的表的乘积：即将第一张表的每一条数据分别对应第二张表的每条数据。我们之前见过，这就是逗号在 FROM 语句中的用法。在实际的应用中，很少有地方能用到 CROSS JOIN，但是一旦用上了，你就可以用这样的 SQL语句表达：

author CROSS JOIN book

DIVISION

DIVISION 的确是一个怪胎。简而言之，如果 JOIN 是一个乘法运算，那么 DIVISION 就是 JOIN 的逆过程。DIVISION 的关系很难用 SQL 表达出来，介于这是一个新手指南，解释 DIVISION 已经超出了我们的目的。但是有兴趣的同学还是可以来看看这三篇文章

（http://blog.jooq.org/2012/03/30/advanced-sql-relational-pision-in-jooq/）

（http://en.wikipedia.org/wiki/Relational_algebra#Division）

（https://www.simple-talk.com/sql/t-sql-programming/pided-we-stand-the-sql-of-relational-pision/）。

推荐阅读 →_→ 《画图解释SQL联合语句》

我们学到了什么？

学到了很多！让我们在脑海中再回想一下。 SQL 是对表的引用， JOIN 则是一种引用表的复杂方式。但是 SQL 语言的表达方式和实际我们所需要的逻辑关系之间是有区别的，并非所有的逻辑关系都能找到对应的 JOIN 操作，所以这就要我们在平时多积累和学习关系逻辑，这样你就能在以后编写 SQL 语句中选择适当的 JOIN 操作了。

7、 SQL 中如同变量的派生表

在这之前，我们学习到过 SQL 是一种声明性的语言，并且 SQL 语句中不能包含变量。但是你能写出类似于变量的语句，这些就叫做派生表：

说白了，所谓的派生表就是在括号之中的子查询：

-- A derived table
FROM (SELECT * FROM author)

需要注意的是有些时候我们可以给派生表定义一个相关名（即我们所说的别名）。

-- A derived table with an aliasFROM (SELECT * FROM author) a

派生表可以有效的避免由于 SQL 逻辑而产生的问题。举例来说：如果你想重用一个用 SELECT 和 WHERE 语句查询出的结果，这样写就可以（以 Oracle 为例）：

-- Get authors' first and last names, and their age in days
SELECT first_name, last_name, age
FROM (
  SELECT first_name, last_name, current_date - date_of_birth age
  FROM author
)
-- If the age is greater than 10000 days
WHERE age > 10000

需要我们注意的是：在有些数据库，以及 SQL ： 1990 标准中，派生表被归为下一级——通用表语句（ common table experssion）。这就允许你在一个 SELECT 语句中对派生表多次重用。上面的例子就（几乎）等价于下面的语句：

WITH a AS (
  SELECT first_name, last_name, current_date - date_of_birth age
  FROM author
)
SELECT *
FROM a
WHERE age > 10000

当然了，你也可以给“ a ”创建一个单独的视图，这样你就可以在更广泛的范围内重用这个派生表了。更多信息可以阅读下面的文章（http://en.wikipedia.org/wiki/View_%28SQL%29）。

我们学到了什么？

我们反复强调，大体上来说 SQL 语句就是对表的引用，而并非对字段的引用。要好好利用这一点，不要害怕使用派生表或者其他更复杂的语句。

8、 SQL 语句中 GROUP BY 是对表的引用进行的操作

让我们再回想一下之前的 FROM 语句：

FROM a, b

现在，我们将 GROUP BY 应用到上面的语句中：

GROUP BY A.x, A.y, B.z

上面语句的结果就是产生出了一个包含三个字段的新的表的引用。我们来仔细理解一下这句话：当你应用 GROUP BY 的时候， SELECT 后没有使用聚合函数的列，都要出现在 GROUP BY 后面。（译者注：原文大意为“当你是用 GROUP BY 的时候，你能够对其进行下一级逻辑操作的列会减少，包括在 SELECT 中的列”）。

• 需要注意的是：其他字段能够使用聚合函数：

SELECT A.x, A.y, SUM(A.z)FROM AGROUP BY A.x, A.y

• 还有一点值得留意的是： MySQL 并不坚持这个标准，这的确是令人很困惑的地方。（译者注：这并不是说 MySQL 没有 GROUP BY 的功能）但是不要被 MySQL 所迷惑。 GROUP BY 改变了对表引用的方式。你可以像这样既在 SELECT 中引用某一字段，也在 GROUP BY 中对其进行分组。

我们学到了什么？

GROUP BY，再次强调一次，是在表的引用上进行了操作，将其转换为一种新的引用方式。

9、 SQL 语句中的 SELECT 实质上是对关系的映射

我个人比较喜欢“映射”这个词，尤其是把它用在关系代数上。（译者注：原文用词为 projection ，该词有两层含义，第一种含义是预测、规划、设计，第二种意思是投射、映射，经过反复推敲，我觉得这里用映射能够更直观的表达出 SELECT 的作用）。一旦你建立起来了表的引用，经过修改、变形，你能够一步一步的将其映射到另一个模型中。 SELECT 语句就像一个“投影仪”，我们可以将其理解成一个将源表中的数据按照一定的逻辑转换成目标表数据的函数。

通过 SELECT语句，你能对每一个字段进行操作，通过复杂的表达式生成所需要的数据。

SELECT 语句有很多特殊的规则，至少你应该熟悉以下几条：

1. 你仅能够使用那些能通过表引用而得来的字段；

2. 如果你有 GROUP BY 语句，你只能够使用 GROUP BY 语句后面的字段或者聚合函数；

3. 当你的语句中没有 GROUP BY 的时候，可以使用开窗函数代替聚合函数；

4. 当你的语句中没有 GROUP BY 的时候，你不能同时使用聚合函数和其它函数；

5. 有一些方法可以将普通函数封装在聚合函数中；

6. ……

一些更复杂的规则多到足够写出另一篇文章了。比如：为何你不能在一个没有 GROUP BY 的 SELECT 语句中同时使用普通函数和聚合函数？（上面的第 4 条）

原因如下：

1. Intuitively, this approach does not make sense logically.

2. If intuition doesn’t convince you, grammar certainly will. The SQL: 1999 standard introduced GROUPING SETS, and the SQL: 2003 standard introduced group sets: GROUP BY(). Whenever an aggregate function appears in your statement and there is no explicit GROUP BY statement, an ambiguous, empty GROUPING SET will be applied to the SQL. Therefore, the original logical order rules are broken, and the mapping (ie SELECT) relationship will first affect the logical relationship, and secondly the grammatical relationship. (Translator's Note: The original text of this paragraph is rather difficult, but it can be simply understood as follows: In a SQL statement that has both aggregate functions and ordinary functions, if there is no GROUP BY for grouping, the SQL statement will treat the entire table as a group by default. When an aggregate function performs aggregate statistics on a certain field, each record in the referenced table loses its meaning. All data is aggregated into one statistical value. It is meaningless for you to use other functions for each record at this time. ).

Confused? Yes, me too. Let’s go back and look at something simpler.

What did we learn?

The SELECT statement is probably the most difficult part of the SQL statement, even though it looks simple. The function of other statements is actually different forms of references to the table. The SELECT statement integrates these references and maps the source table to the target table through logical rules. Moreover, this process is reversible, and we can clearly know how the data in the target table comes from.

If you want to learn the SQL language well, you must understand other statements before using the SELECT statement. Although SELECT is the first keyword in the grammatical structure, it should be the last one we master.

10. Several simple keywords in SQL statements: DISTINCT, UNION, ORDER BY and OFFSET

After learning the complex SELECT Yu opera, let’s look at something simple:

• Set operations (DISTINCT and UNION)

• Sort operations (ORDER BY, OFFSET…FETCH)

Set operation (set operation):

The main operation of set operation is on the set. In fact, it refers to an operation on the table. Conceptually, they are easy to understand:

• DISTINCT Deduplicate the data after mapping

• UNION Splice the two subqueries together and remove duplicates

• UNION ALL Splice the two subqueries together without removing duplicates

• EXCEPT Concatenate the results in the second word query from Remove

• INTERSECT from the first subquery, retain the results in both subqueries and remove duplicates

sorting operation ( ordering operation):

The sorting operation has nothing to do with logical relationships. This is a SQL-specific feature. The sorting operation is not only performed at the end of the SQL statement, but also is performed last during the execution of the SQL statement. Using ORDER BY and OFFSET...FETCH is the most effective way to ensure that the data can be arranged in order. All other sorting methods are somewhat random, although the sorting results they produce are reproducible.

OFFSET…SET is a statement with no unified syntax. Different databases have different expression methods, such as LIMIT…OFFSET of MySQL and PostgreSQL, TOP…START AT of SQL Server and Sybase, etc. For specific information about the different syntaxes of OFFSET..FETCH, please refer to this article

（http://www.jooq.org/doc/3.1/manual/sql-building/sql-statements/select- statement/limit-clause/).

Let us enjoy using SQL at work!

Just like other languages, if you want to learn SQL language well, you need a lot of practice. The 10 simple steps above can help you gain a better understanding of the SQL statements you write every day. On the other hand, a lot of experience can be accumulated from common mistakes. The following two articles introduce some common SQL mistakes made by JAVA and other developers:

• ##10 Common Mistakes Java Developers Make when Writing SQL

• 10 More Common Mistakes Java Developers Make when Writing SQL

##Related recommendations:

Query optimization for parsing tree hierarchical data in SQL

In data query, starting from 2008, SQL Server provides a new data type hierarchyid, which is specially used for operations. ....

The above is the detailed content of Perfectly parsing SQL requires only ten simple steps. For more information, please follow other related articles on the PHP Chinese website!