我正在编写一个迭代数组的多态 PL/pgSQL 函数。我对使用
FOREACH
很感兴趣,但是我不知道如何声明具有正确类型的临时变量。
我的函数如下,有关更多信息,请参阅第 4 行的注释。
CREATE OR REPLACE FUNCTION uniq(ary anyarray) RETURNS anyarray AS $$
DECLARE
ret ary%TYPE := '{}';
v ???; -- how do I get the element type of @ary@?
BEGIN
IF ary IS NULL THEN
return NULL;
END IF;
FOREACH v IN ARRAY ary LOOP
IF NOT v = any(ret) THEN
ret = array_append(ret, v);
END IF;
END LOOP;
RETURN ret;
END;
$$ LANGUAGE plpgsql;
如果没有“模板”变量或参数,则无法声明多态类型的变量。
手册中章节声明函数参数有相关示例,但是这个技巧没有涉及到:
将数据类型为IN
的另一个
INOUT
、
OUT
或
ANYELEMENT
参数添加到函数定义中。它自动解析为匹配的元素类型,并且可以直接(ab)用作函数体内的变量或用作更多变量的模板:
CREATE OR REPLACE FUNCTION uniq1(ary ANYARRAY, v ANYELEMENT = NULL)
RETURNS anyarray
LANGUAGE plpgsql AS
$func$
DECLARE
ret ary%TYPE := '{}';
some_var v%TYPE; -- we could declare more variables now
-- but we don't need to
BEGIN
IF ary IS NULL THEN
RETURN NULL;
END IF;
FOREACH v IN ARRAY ary LOOP -- instead, we can use v directly
IF NOT v = any(ret) THEN
ret := array_append(ret, v);
END IF;
END LOOP;
RETURN ret;
END
$func$;
相关:这样的复制类型仅适用于
DECLARE
部分,并且是不同的类型转换。这里的手册中有说明。 分配一个默认值,因此添加的参数不必包含在函数调用中:
ANYELEMENT
= NULL
致电(不变):
SELECT uniq1('{1,2,1}'::int[]);
SELECT uniq1('{foo,bar,bar}'::text[]);
更好的功能
CREATE OR REPLACE FUNCTION uniq2(ary ANYARRAY, elem ANYELEMENT = NULL
, OUT ret ANYARRAY)
RETURNS anyarray
LANGUAGE plpgsql AS
$func$
BEGIN
IF ary IS NULL
THEN RETURN;
ELSE ret := '{}'; -- init
END IF;
FOREACH elem IN ARRAY ary LOOP
IF elem = ANY(ret) THEN -- do nothing
ELSE
ret := array_append(ret, elem);
END IF;
END LOOP;
END
$func$;
但这仍然没有涵盖所有包含空元素的情况。功能正常
CREATE OR REPLACE FUNCTION uniq3(ary ANYARRAY, elem ANYELEMENT = NULL
, OUT ret ANYARRAY)
RETURNS anyarray
LANGUAGE plpgsql AS
$func$
BEGIN
IF ary IS NULL
THEN RETURN;
ELSE ret := '{}'; -- init
END IF;
FOREACH elem IN ARRAY ary LOOP
IF elem IS NULL THEN -- special test for NULL
IF array_length(array_remove(ret, NULL), 1) = array_length(ret, 1) THEN
ret := array_append(ret, NULL);
END IF;
ELSIF elem = ANY(ret) THEN -- do nothing
ELSE
ret := array_append(ret, elem);
END IF;
END LOOP;
END
$func$;
检查数组中的 NULL 有点痛苦:所有这些功能都
只是概念证明。我不会使用都不会。相反:
使用简单 SQL 的卓越解决方案WITH ORDINALITY
来保留元素的原始顺序。 详细解释:单值基本代码:
SELECT ARRAY (
SELECT elem
FROM (
SELECT DISTINCT ON (elem) elem, i
FROM unnest('{1,2,1,NULL,4,NULL}'::int[]) WITH ORDINALITY u(elem, i)
ORDER BY elem, i
) sub
ORDER BY i) AS uniq;
退货:
uniq
------------
{1,2,NULL,4}
关于DISTINCT ON
:内置查询:
SELECT *
FROM test t
, LATERAL (
SELECT ARRAY (
SELECT elem
FROM (
SELECT DISTINCT ON (elem) elem, i
FROM unnest(t.arr) WITH ORDINALITY u(elem, i)
ORDER BY elem, i
) sub
ORDER BY i) AS arr
) a;
这有一个很小的极端情况:它返回一个空数组 NULL 数组。覆盖所有基础:
SELECT t.*, CASE WHEN t.arr IS NULL THEN NULL ELSE a.arr END AS arr
FROM test t
, LATERAL (
SELECT ARRAY (
SELECT elem
FROM (
SELECT DISTINCT ON (elem) elem, ord
FROM unnest(t.arr) WITH ORDINALITY u(elem, ord)
ORDER BY elem, ord
) sub
ORDER BY ord) AS arr
) a;
或者:
SELECT *
FROM test t
LEFT JOIN LATERAL (
SELECT ARRAY (
SELECT elem
FROM (
SELECT DISTINCT ON (elem) elem, i
FROM unnest(t.arr) WITH ORDINALITY u(elem, i)
ORDER BY elem, i
) sub
ORDER BY i) AS arr
) a ON t.arr IS NOT NULL;
在 Postgres 9.3 或更早版本中,您可以替换为 generate_subscripts()
:
SELECT *
FROM test t
, LATERAL (
SELECT ARRAY (
SELECT elem
FROM (
SELECT DISTINCT ON (t.arr[i]) t.arr[i] AS elem, i
FROM generate_subscripts(t.arr, 1) i
ORDER BY t.arr[i], i
) sub
ORDER BY i
) AS arr
) a;
我们在sqlfiddle中需要这个,目前只支持pg 9.3,所以WITH ORDINALITY
不可用:
anyarray
参数的基类型变量(文档没有提到这种可能性)。 您可以将
FOR LOOP
与整型变量一起使用:
CREATE OR REPLACE FUNCTION uniq(ary anyarray) RETURNS anyarray AS $$
DECLARE
ret ary%TYPE := '{}';
i int;
BEGIN
IF ary IS NULL THEN
return NULL;
END IF;
FOR i IN array_lower(ary, 1) .. array_upper(ary, 1) LOOP
IF NOT ary[i] = any(ret) THEN
ret = array_append(ret, ary[i]);
END IF;
END LOOP;
RETURN ret;
END;
$$ LANGUAGE plpgsql;
但是,循环和变量可能不是必需的:
create or replace function uniq_without_loop(arr anyarray)
returns anyarray language plpgsql as $$
begin
return (
select array_agg(distinct elem)
from unnest(arr) elem);
end $$;
上述函数保持数组顺序不变的版本:
create or replace function unsorted_uniq_without_loop(arr anyarray)
returns anyarray language plpgsql as $$
begin
return (
select array_agg(elem)
from (
select elem
from (
select distinct on(elem) elem, row_number() over ()
from unnest(array[arr]) elem
) sub
order by row_number
) sub);
end $$;