Convert two columns into key-value json object?

2019-05-02 10:46发布

问题:

Using FOR JSON AUTO or FOR JSON PATH on the following record set (which representing a product's attributes):

attribute | value
-----------------
color     | red
size      | small

will produce:

[{"attribute":"color","value":"red"},{"attribute":"size","value":"small"}]

Is there any way to produce the following Instead:

{"color":"red","size":"small"}

Note that as every product attribute is different than others; so this record set is different for every product. PIVOTing is not an option as it needs dynamic sql! Seems we need a function to be able to CROSS it with products table to produce for example a product catalog.

回答1:

Instead of JSON functions of SQL Server 2016, I used string concatenation function string_agg in SQL Server 2017 as seen in following script

/*create table ProductAttributes (
    product int,
    attribute varchar(40),
    value varchar(40)
)
insert into ProductAttributes select 1, 'color', 'red'
insert into ProductAttributes select 1, 'size', 'small'
insert into ProductAttributes select 2, 'processor', 'intel'
insert into ProductAttributes select 2, 'ram', '16'
insert into ProductAttributes select 2, 'weight', '2'*/

select 
    product, '{' + STRING_AGG( '"' + attribute + '":"' + STRING_ESCAPE(value,'json') + '"' ,',') + '}' as attributes
from ProductAttributes 
group by product

Output is as follows for the two product entries product attributes 1 {"color":"red","size":"small"} 2 {"processor":"intel","ram":"16","weight":"2"}

If you are using a previous version than SQL Server 2017, you can use string concatenation using SQL XML Path as follows

SELECT
    product,
  '{' + STUFF(
    (
    SELECT
      ',' + '"' + attribute + '":"' + STRING_ESCAPE(value,'json') + '"'
    FROM ProductAttributes a
        where a.product = p.product
    FOR XML PATH(''),TYPE
    ).value('.','VARCHAR(MAX)'
    ), 1, 1, ''
  ) + '}' As attributes
from ProductAttributes p
group by product

Developers will get the same result

I've updated above SQL queries and used String_Escape() function @Eilert's comment