试试看：

CREATE OR REPLACE FUNCTION get_grade(int)
  RETURNS TABLE (i int, ii int, iii int, iiii int)
  LANGUAGE sql AS
$func$
SELECT count(t.priority =  66 OR NULL)::int  -- AS I
     , count(t.priority =  67 OR NULL)::int  -- AS II
     , count(t.priority =  68 OR NULL)::int  -- AS III
     , count(t.priority = 225 OR NULL)::int  -- AS IIII
FROM   dt_info t 
WHERE  t.registrant = $1;
$func$;

你 * 可以 * 用LANGUAGE plpgsql编写一个类似这样的简单查询。你也可以只使用**LANGUAGE sql**。两者都有优缺点。显示一个SQL函数。不要用引号引用语言名称。手册：
不建议使用单引号将名称括起来，并且要求大小写匹配。

函数体内部的列别名在外部不可见。在这种情况下，它们仅用作文档。请使用namedOUT参数，否则它们将获得默认名称。RETURNS TABLE使其成为返回集合的函数，并且在任何情况下都需要列名。

OUT参数在函数体中的每个DML SQL语句中都是可见的。表限定列名，否则会与OUT参数名冲突。不要使用会冲突的别名。为了清楚起见，我注解掉了您的别名（尽管现代版本的Postgres不会在这里冲突）。
返回列的数据类型必须与标头中的声明完全匹配。sum()或count()返回bigint。必须**显式转换为integer**以匹配返回类型。

不带引号的大写标识符在Postgres中被折叠成小写，只会造成混淆。

我使用了更短（也更快）的表达式来计算条件计数。在Postgres 9.4或更高版本中，请使用聚合FILTER。请参见：

• For absolute performance, is SUM faster or COUNT?
• 条件SQL计数

我不能肯定我是否理解你所说的话。下面只是一个测试。

--create table 
skytf=> create table grade (registrant integer, pirority integer);
CREATE TABLE
skytf=> insert into grade values (1,66);
INSERT 0 1
skytf=> insert into grade values (1,66);
INSERT 0 1
skytf=> insert into grade values (1,67);
INSERT 0 1
skytf=> insert into grade values (1,67);
INSERT 0 1
skytf=> insert into grade values (1,67);
INSERT 0 1
skytf=> insert into grade values (1,68);
INSERT 0 1
skytf=> insert into grade values (1,225);
INSERT 0 1
skytf=> insert into grade values (1,225);
INSERT 0 1
skytf=> insert into grade values (1,225);
INSERT 0 1
skytf=> insert into grade values (1,225);
INSERT 0 1
skytf=> select * from grade;
 registrant | pirority 
------------+----------
          1 |       66
          1 |       66
          1 |       67
          1 |       67
          1 |       67
          1 |       68
          1 |      225
          1 |      225
          1 |      225
          1 |      225

--create function
CREATE OR REPLACE FUNCTION get_grade( in_reg integer ) RETURNS  RECORD AS
$$
DECLARE
    g_user record;
BEGIN
    select 
    sum(case when t.pirority = 66 then 1 else 0 end) as I ,
    sum(case when t.pirority = 67 then 1 else 0 end) as II,
    sum(case when t.pirority = 68 then 1 else 0 end) as III,
    sum(case when t.pirority = 225 then 1 else 0 end) as IIII 
    into g_user
    from grade  t 
    where t.registrant = in_reg;
    return g_user;
END;
$$
LANGUAGE PLPGSQL;          

--execute function
skytf=> select get_grade(1);
 get_grade 
-----------
 (2,3,1,4)
(1 row)

我想你应该使用RETURN QUERY和一个setof record返回类型：

CREATE OR REPLACE FUNCTION get_grade(IN integer, out bigint, out bigint, out bigint, out bigint)
returns setof record AS
$BODY$
begin
    return query select 
        sum(case when t.pirority = 66 then 1 else 0 end) as I,
        sum(case when t.pirority = 67 then 1 else 0 end) as II,
        sum(case when t.pirority = 68 then 1 else 0 end) as III,
        sum(case when t.pirority = 225 then 1 else 0 end) as IIII 
        from dt_info t 
        where t.registrant = $1;
end
$BODY$
LANGUAGE plpgsql VOLATILE;

您也可以将其作为一个普通的SQL函数来执行：

CREATE OR REPLACE FUNCTION get_grade(IN integer, out bigint, out bigint, out bigint, out bigint)
returns setof record AS
$BODY$
    select 
        sum(case when t.pirority = 66 then 1 else 0 end) as I,
        sum(case when t.pirority = 67 then 1 else 0 end) as II,
        sum(case when t.pirority = 68 then 1 else 0 end) as III,
        sum(case when t.pirority = 225 then 1 else 0 end) as IIII 
    from dt_info t 
    where t.registrant = $1;
$BODY$
LANGUAGE sql VOLATILE;

顺便说一句，t.pirority应该是t.priority还是真实的的列名？