The function of SQL GROUP BY dependence

SQLThe standard knows an interesting feature,You can cast inGROUP BY The primary key listed in the clause（或唯一键）any functional dependencies,without the need to explicitly add that functional dependency to GROUP BY 子句中.

这意味着什么呢？Consider this simple pattern：

CREATE TABLE author (
  id INT NOT NULL PRIMARY KEY,
  name TEXT NOT NULL
);

CREATE TABLE book (
  id INT NOT NULL PRIMARY KEY,
  author_id INT NOT NULL REFERENCES author,
  title TEXT NOT NULL
);

复制代码

To count the number of books by author,We tend to write：

SELECT a.name, count(b.id)
FROM author a
LEFT JOIN book b ON a.id = b.author_id
GROUP BY 
  a.id,  -- Required, because names aren't unique
  a.name -- Required in some dialects, but not in others

复制代码

在这种情况下,We have to group by something unique,Because if both authors call John Doe,We still want them to produce different groups.因此,GROUP BY a.id 是一个必然的结果.

We are used to using both at the same timeGROUP BY a.name ,Especially in these dialects that require such,因为我们在SELECT clause listeda.name ：

• Db2
• Derby
• Exasol
• Firebird
• HANA
• Informix
• Oracle
• SQL Server

But is it really necessary？根据SQL标准,它不是,因为在author.id 和author.name 之间存在着功能依赖.换句话说,对于author.id 的每一个值,恰好有一个author.name 的可能值,或者说author.name 是一个函数.author.id

这意味着,如果我们GROUP BY 两列,Or just the primary key,这并不重要.The result in both cases must be the same,So it is possible：

SELECT a.name, count(b.id)
FROM author a
LEFT JOIN book b ON a.id = b.author_id
GROUP BY a.id
复制代码

哪些SQLThe dialect supports this？

至少有以下SQLThe dialect supports this language feature：

• CockroachDB
• H2
• HSQLDB
• MariaDB
• MySQL
• PostgreSQL
• SQLite
• Yugabyte

值得注意的是,在有GROUP BY 的情况下,MySQLIt was once simply ignored whether a column could be unambiguously predictable.Although the query below is rejected in most dialects,但在MySQL中,在引入ONLY_FULL_GROUP_BY模式之前,It was not rejected：

SELECT author_id, title, count(*)
FROM author
GROUP BY author_id
复制代码

If an author has written more than one book,我们应该为author.title ,显示什么？这没有意义,但MySQLStill used to allow it,and will cast any arbitrary value from the group.

今天,MySQLOnly casting AND is allowedGROUP BY The clause has a functional relationship to the columns,这是SQL标准所允许的.

优点和缺点

Although a shorter syntax that avoids extra columns may be easier to maintain（如果需要的话,It's easy to cast extra columns）,But in production there is a risk of some query outages,i.e. when the underlying constraint is disabled,For example for migration.Although it is unlikely to disable primary keys in a live system,But it's still possible that this could happen,如果没有主键,Queries that were previously valid will no longer be valid,原因与MySQLThe old interpretation is invalid for the same.There are no longer guaranteed functional dependencies.

其他语法

从jOOQ 3.16和#11834开始,将有可能在GROUP BY The table is referenced directly in the clause,rather than a single column.比如说：

SELECT a.name, count(b.id)
FROM author a
LEFT JOIN book b ON a.id = b.author_id
GROUP BY a
复制代码

The semantics will be：

• If the table has a primary key（无论是否复合）,在GROUP BY The primary key is used in the clause.
• 如果表没有主键,All columns of the table are listed.

由于jOOQ支持的RDBMSNone of this syntax is currently supported,So it's a pure one 合成jOOQ功能.