Currently json_parse parse's json's that are verifiably bad or incomplete (See example below). In the prestissimo project we are wondering whether this behavior is something we should emulate and whether this behavior is something customers expect.
Secondly as you can see (row 2 vs row 3 in example below) the behavior with bad json's is also not consistent.
Expected Behavior
It would be easier and more consistent if Presto just threw on bad or incomplete json's rather than returning partial results.
Current Behavior
Below is a snippet of current behavior:
presto:default> SELECT try(json_parse(x)), TRY_CAST(TRY("json_parse"(x)) AS array(varchar)), ARRAY[X] from (values '[2] [3]', '[3] bad json]', '[4] [bad json') as t(x
);
_col0 | _col1 | _col2
-------+-------+------------------
[2] | [2] | [[2] [3]]
NULL | NULL | [[3] bad json]]
[4] | [4] | [[4] [bad json]
(3 rows)
Possible Solution
Throw on bad json's.
Context
We are currently trying to verify and get Prestissimo behavior similar to Presto behavior and running into verification errors due to this.
kgpai
commented
1 hour ago
Currently json_parse parse's json's that are verifiably bad or incomplete (See example below). In the prestissimo project we are wondering whether this behavior is something we should emulate and whether this behavior is something customers expect. Secondly as you can see (row 2 vs row 3 in example below) the behavior with bad json's is also not consistent.
Expected Behavior
It would be easier and more consistent if Presto just threw on bad or incomplete json's rather than returning partial results.
Current Behavior
Below is a snippet of current behavior:
Possible Solution
Throw on bad json's.
Context
We are currently trying to verify and get Prestissimo behavior similar to Presto behavior and running into verification errors due to this.