1. 关联子查询

1.1. 关联子查询和自连接在很多时候都是等价的

1.2. 使用SQL进行行间比较时，发挥主要作用的技术是关联子查询，特别是与自连接相结合的“自关联子查询”

1.3. 缺点

• 1.3.1. 代码的可读性不好

  • 1.3.1.1. 特别是在计算累计值和移动平均值的例题里，与聚合一起使用后，其内部处理过程非常难理解

• 1.3.2. 性能不好

  • 1.3.2.1. 特别是在SELECT子句里使用标量子查询时，性能可能会变差

2. 增长、减少、维持现状

2.1. 使用基于时间序列的表进行时间序列分析

2.2. 示例

• 2.2.1. --求与上一年营业额一样的年份(1)：使用关联子查询

   SELECT year, sale
     FROM Sales S1
    WHERE sale = (SELECT sale
                   FROM Sales S2
                   WHERE S2.year = S1.year -1)
    ORDER BY year;

• 2.2.2. S2.year = S1.year -1这个条件起到了将要比较的数据偏移一行的作用

• 2.2.3. --求与上一年营业额一样的年份(2)：使用自连接

   SELECT S1.year, S1.sale
     FROM Sales S1,
         Sales S2
    WHERE S2.sale = S1.sale
     AND S2.year = S1.year -1
    ORDER BY year;

3. 用列表展示与上一年的比较结果

3.1. 示例

• 3.1.1. --求出是增长了还是减少了，抑或是维持现状(1)：使用关联子查询

   SELECT S1.year, S1.sale,
         CASE WHEN sale =
               (SELECT sale
                   FROM Sales S2
                 WHERE S2.year = S1.year -1) THEN'→'--持平
               WHEN sale ＞
               (SELECT sale
                   FROM Sales S2
                 WHERE S2.year = S1.year -1) THEN'↑'--增长
               WHEN sale ＜
               (SELECT sale
                   FROM Sales S2
                 WHERE S2.year = S1.year -1) THEN'↓'--减少
         ELSE'—'END AS var
     FROM Sales S1
    ORDER BY year;

• 3.1.2. --求出是增长了还是减少了，抑或是维持现状(2)：使用自连接查询（最早的年份不会出现在结果里）

   SELECT S1.year, S1.sale,
         CASE WHEN S1.sale = S2.sale THEN'→'
               WHEN S1.sale ＞ S2.sale THEN'↑'
               WHEN S1.sale ＜ S2.sale THEN'↓'
         ELSE'—'END AS var
     FROM Sales S1, Sales S2
    WHERE S2.year = S1.year -1
    ORDER BY year;

4. 时间轴有间断时

4.1. 和过去最临近的时间进行比较

4.2. 示例

• 4.2.1. --查询与过去最临近的年份营业额相同的年份

   SELECT year, sale
     FROM Sales2 S1
    WHERE sale =
     (SELECT sale
         FROM Sales2 S2
       WHERE S2.year =
         (SELECT MAX(year)  --条件2：在满足条件1的年份中，年份最早的一个
             FROM Sales2 S3
           WHERE S1.year ＞ S3.year))  --条件1：与该年份相比是过去的年份
    ORDER BY year;

• 4.2.2.  自连接版本

SELECT S1.year AS year,

         S1.year AS year
     FROM Sales2 S1, Sales2 S2
    WHERE S1.sale = S2.sale
     AND S2.year = (SELECT MAX(year)
                       FROM Sales2 S3
                     WHERE S1.year ＞ S3.year)
    ORDER BY year;

• 4.2.3. --求每一年与过去最临近的年份之间的营业额之差(1)：结果里不包含最早的年份

   SELECT S2.year AS pre_year,
         S1.year AS now_year,
         S2.sale AS pre_sale,
         S1.sale AS now_sale,
         S1.sale - S2.sale  AS diff
     FROM Sales2 S1, Sales2 S2
    WHERE S2.year = (SELECT MAX(year)
                       FROM Sales2 S3
                     WHERE S1.year ＞ S3.year)
    ORDER BY now_year;

• 4.2.4. --求每一年与过去最临近的年份之间的营业额之差(1)：结果里不包含最早的年份

   SELECT S2.year AS pre_year,
         S1.year AS now_year,
         S2.sale AS pre_sale,
         S1.sale AS now_sale,
         S1.sale - S2.sale  AS diff
     FROM Sales2 S1, Sales2 S2
    WHERE S2.year = (SELECT MAX(year)
                       FROM Sales2 S3
                     WHERE S1.year ＞ S3.year)
    ORDER BY now_year;

• 4.2.5. 使用极值函数时会发生排序

5. 移动累计值和移动平均值

5.1. 示例

• 5.1.1. --求累计值：使用窗口函数

   SELECT prc_date, prc_amt,
         SUM(prc_amt) OVER (ORDER BY prc_date) AS onhand_amt
     FROM Accounts;

• 5.1.2. 引入窗口函数的目的原本就是解决这类问题，因此这里的代码非常简洁

  • 5.1.2.1. 如果选用的数据库支持窗口函数，也可以考虑使用窗口函数

• 5.1.3. 从性能方面来看，表的扫描和数据排序也都只进行了一次

  • 5.1.3.1. 依赖于具体的数据库的

• 5.1.4. --求累计值：使用冯·诺依曼型递归集合

   SELECT prc_date, A1.prc_amt,
         (SELECT SUM(prc_amt)
           FROM Accounts A2
           WHERE A1.prc_date ＞= A2.prc_date ) AS onhand_amt
     FROM Accounts A1
    ORDER BY prc_date;

• 5.1.5. --求移动累计值(1)：使用窗口函数

   SELECT prc_date, prc_amt,
         SUM(prc_amt) OVER (ORDER BY prc_date
                           ROWS 2 PRECEDING) AS onhand_amt
     FROM Accounts;

• 5.1.6. --求移动累计值(2)：不满3行的时间区间也输出

   SELECT prc_date, A1.prc_amt,
         (SELECT SUM(prc_amt)
           FROM Accounts A2
           WHERE A1.prc_date ＞= A2.prc_date
             AND (SELECT COUNT(＊)
                   FROM Accounts A3
                   WHERE A3.prc_date
                     BETWEEN A2.prc_date AND A1.prc_date  ) ＜= 3 )
                 AS mvg_sum
     FROM Accounts A1
    ORDER BY prc_date;

• 5.1.7. A3.prc_date在以A2.prc_date为起点，以A1.prc_date为终点的区间内移动

• 5.1.8. --移动累计值(3)：不满3行的区间按无效处理

   SELECT prc_date, A1.prc_amt,
    (SELECT SUM(prc_amt)
       FROM Accounts A2
     WHERE A1.prc_date ＞= A2.prc_date
       AND (SELECT COUNT(＊)
               FROM Accounts A3
             WHERE A3.prc_date
               BETWEEN A2.prc_date AND A1.prc_date  ) ＜= 3
     HAVING  COUNT(＊) =3) AS mvg_sum  --不满3行数据的不显示
     FROM Accounts A1
    ORDER BY prc_date;

5.2. 基本思路是使用冯·诺依曼型递归集合

6. 查询重叠的时间区间

6.1. 示例

• 6.1.1. --求重叠的住宿期间

   SELECT reserver, start_date, end_date
     FROM Reservations R1
    WHERE EXISTS
         (SELECT ＊
               FROM Reservations R2
              WHERE R1.reserver ＜＞ R2.reserver  --与自己以外的客人进行比较
                AND ( R1.start_date BETWEEN R2.start_date AND R2.end_date
                                   --条件(1)：自己的入住日期在他人的住宿期间内
                   OR R1.end_date  BETWEEN R2.start_date AND R2.end_date));
                                   --条件(2)：自己的离店日期在他人的住宿期间内

• 6.1.2. --升级版：把完全包含别人的住宿期间的情况也输出

   SELECT reserver, start_date, end_date
    FROM Reservations R1
   WHERE EXISTS
         (SELECT ＊
             FROM Reservations R2
           WHERE R1.reserver ＜＞ R2.reserver
             AND (  (     R1.start_date BETWEEN R2.start_date
                                           AND R2.end_date
                       OR R1.end_date   BETWEEN R2.start_date
                                           AND R2.end_date)
                   OR (    R2.start_date BETWEEN R1.start_date
                                           AND R1.end_date
                       AND R2.end_date   BETWEEN R1.start_date
                                           AND R1.end_date)));