使用大量 if-else 语句和 JSON 化简化 PL/pgSQL 函数

Question

我有 PL/PgSQL 函数，它执行非常相似的查询（唯一不同的是列名）并将输出转换为 JSON 对象。

CREATE OR REPLACE FUNCTION get_observations(kind varchar, site_id integer, var varchar) RETURNS TABLE (fc json) AS
$func$
BEGIN
    IF kind = 'raw' THEN
        IF var = 'o2_abs' THEN
            RETURN QUERY SELECT array_to_json(array_agg(row_to_json(obs))) FROM (
                SELECT observation_date AS date, o2_abs AS value FROM oxygen WHERE new_id = site_id) AS obs;
        ELSIF var = 'o2_rel' THEN
            RETURN QUERY SELECT array_to_json(array_agg(row_to_json(obs))) FROM (
                SELECT observation_date AS date, o2_rel AS value FROM oxygen WHERE new_id = site_id) AS obs;
        ELSIF var = 'temp' THEN
            RETURN QUERY SELECT array_to_json(array_agg(row_to_json(obs))) FROM (
                SELECT observation_date AS date, t AS value FROM oxygen WHERE new_id = site_id) AS obs;
        END IF;
    ELSIF kind = 'averaged' THEN
        IF var = 'o2_abs' THEN
            RETURN QUERY SELECT array_to_json(array_agg(row_to_json(obs))) FROM (
                SELECT month AS month, o2_abs AS value FROM oxygen_month_average WHERE new_id = site_id) AS obs;
        ELSIF var = 'o2_rel' THEN
            RETURN QUERY SELECT array_to_json(array_agg(row_to_json(obs))) FROM (
                SELECT month AS month, o2_rel AS value FROM oxygen_month_average WHERE new_id = site_id) AS obs;
        ELSIF var = 'temp' THEN
            RETURN QUERY SELECT array_to_json(array_agg(row_to_json(obs))) FROM (
                SELECT month AS month, t AS value FROM oxygen_month_average WHERE new_id = site_id) AS obs;
        END IF;
    END IF;
END;
$func$ LANGUAGE plpgsql;

函数体包含大量重复代码。我看到了几种改进它的方法，但不知道PL/PgSQL 是否允许这些技巧：

1. 将obs查询结果存储在中间变量中，并在函数末尾使用array_to_json(array_agg(row_to_json(obs)))将其转换为JSON。
2. 在查询中将var varchar 转换为列名定义，以避免大多数IF/ELSE 语句；

PostgreSQL 服务器版本为 9.3.6。

表架构：

oxygen=# \d+ oxygen
Table "public.oxygen"
         Column          |  Type   |                      Modifiers       (...)
-------------------------+---------+----------------------------------------------------
 old_id                  | text    | not null
 observation_date_string | text    | not null
 t                       | real    | not null
 o2_abs                  | real    | not null
 o2_sat                  | real    |
 o2_rel                  | real    |
 observation_date        | date    |
 new_id                  | integer |
 id                      | bigint  | not null default nextval('oxygen_id_seq'::regclass)
Indexes:
    "oxygen_pkey" PRIMARY KEY, btree (id)
Foreign-key constraints:
    "oxygen_new_id_fkey" FOREIGN KEY (new_id) REFERENCES unique_geoms(new_id)
    "oxygen_old_id_fkey" FOREIGN KEY (old_id) REFERENCES location(old_id)


oxygen=# \d+ oxygen_month_average
Table "public.oxygen_month_average"
 Column |       Type       |                             Modifiers        (...)
--------+------------------+-------------------------------------------------------------------
 new_id | integer          |
 month  | integer          |
 t      | double precision |
 o2_abs | double precision |
 o2_rel | double precision |
 id     | bigint           | not null default nextval('oxygen_month_average_id_seq'::regclass)
Indexes:
    "oxygen_month_average_pkey" PRIMARY KEY, btree (id)

Answer 1

还有另一种非常有效的方法来简化大部分内容：CASE 语句：

CREATE OR REPLACE FUNCTION get_observations(
     _kind    text
   , _site_id int
   , _var     text)
  RETURNS TABLE (fc json) AS
$func$
BEGIN
   CASE _kind
    WHEN 'raw' THEN
      RETURN QUERY
      SELECT json_agg(obs) FROM (
         SELECT observation_date AS date
              , CASE _var
                 WHEN 'o2_abs' THEN o2_abs
                 WHEN 'o2_rel' THEN o2_rel
                 WHEN 'temp'   THEN t
                END AS value
         FROM   oxygen
         WHERE  new_id = _site_id
         ) AS obs;
    WHEN 'averaged' THEN
      RETURN QUERY
      SELECT json_agg(obs) FROM (
         SELECT month AS month
              , CASE _var
                 WHEN 'o2_abs' THEN o2_abs
                 WHEN 'o2_rel' THEN o2_rel
                 WHEN 'temp'   THEN t
                END AS value
         FROM   oxygen_month_average
         WHERE  new_id = _site_id
         ) AS obs;
   END CASE;
END
$func$  LANGUAGE plpgsql;

SQL Fiddle.

• 仍然是两个不同的查询，因为它们基于不同的表。要折叠这些，您也需要使用 EXECUTE 的动态 SQL，但这不会变得更短，而且通常效率较低......

• 使用“简单”（或“切换”）CASE。注意CASE 关键字在两个不同的上下文中：外部CASE 是一个plpgsql 控制结构，内部CASE is an SQL command。相关：

  • Simplify nested case when statement

• 您可以使用json_agg() 比@Igor already suggested 进一步简化： array_to_json(array_agg(row_to_json(obs))) -> json_agg(obs)

Answer 2

您可以在 psql 中使用EXECUTE statement 来生成动态查询。

但它不会缓存查询计划，因此它可以使每次调用稍微慢一些。因此，无论代码重复如何，您的代码都还不错。

但添加对无效参数的错误检查。否则，如果您在某处打错了 avergaed 之类的错字，您将很难找到错误。