Re: json/jsonb inconsistence - 2

On 05/29/2014 08:15 AM, Andrew Dunstan wrote:
>
> On 05/29/2014 08:00 AM, Teodor Sigaev wrote:
>> postgres=# select '["\u0000"]'::json->0;
>>  ?column?
>> ----------
>>  "\u0000"
>> (1 row)
>>
>> Time: 1,294 ms
>> postgres=# select  '["\u0000"]'::jsonb->0;
>>  ?column?
>> -----------
>>  "\\u0000"
>> (1 row)
>>
>> It seems to me that escape_json() is wrongly used in 
>> jsonb_put_escaped_value(), right name of escape_json() is a 
>> escape_to_json().
>
>
> That's a bug. I will look into it. I think we might need to 
> special-case \u0000 on output, just as we do on input.


Actually, this is just the tip of the iceberg.

Here's what 9.3 does:
   andrew=# select array_to_json(array['a','\u0000','b']::text[]);        array_to_json   ---------------------
["a","\\u0000","b"]


I'm now wondering if we should pass though any unicode escape 
(presumably validated to some extent). I guess we can't change this in 
9.2/9.3 because it would be a behaviour change.

These unicode escapes have given us more trouble than any other part of 
the JSON spec :-(

cheers

andrew