更新Postgres裏一列只返回錯誤一個值（Postgres的9.5.1）

Question

假設我有創建的表如下：

create table test.sales 
(
customer text, 
purchased text, 
date_purchased date, 
rownumber integer, 
primary key (customer, date_purchased) 
) 
; 

insert into test.sales 
values 
('kevin', 'books', '2017-01-01'::date, null), 
('kevin', 'movies', '2017-01-02'::date, null), 
('paul', 'books', '2017-01-05'::date, null), 
('paul', 'movies', '2017-01-07'::date, null) 

在這一點上，rownumber總是NULL，我想設置值rownumber與row_number() over (partition by customer order by date_purchased) as rownumber。我要對這個方法如下：

update test.sales as a 
set (rownumber) = 
(
select 
    row_number() over (partition by customer order by date_purchased) as rownumber 
from test.sales as b 
-- These fields correspond to the primary keys of the table. 
where a.customer = b.customer 
    and a.date_purchased = b.date_purchased 
) 

但出於某種原因，這個返回：

customer purchased date_purchased rownumber 
kevin  books   1/1/17   1 
kevin  movies  1/2/17   1 
paul   books   1/5/17   1 
paul   movies  1/7/17   1 

我期待這樣的：

customer purchased date_purchased rownumber 
kevin  books   1/1/17   1 
kevin  movies  1/2/17   2 
paul   books   1/5/17   1 
paul   movies  1/7/17   2 

請注意，在實際結果，rownumber始終是1.爲什麼這是？

Answer 1

您正在使用標量子查詢。使用相關子查詢來代替：

---

UPDATE test.sales dst 
SET rownumber = src.rownumber 
FROM (SELECT customer,date_purchased 
     , row_number() over (partition by customer order by date_purchased) as rownumber 
     from test.sales 
     ) src 
WHERE dst.customer = src.customer 
    AND dst.date_purchased = src.date_purchased 
     ;