How to Convert Postgresql 9.4'S Jsonb Type to Float
I'm Trying the Following Query: Select (Json_Data->'Position'->'Lat') + 1.0 as Lat from Updates Limit 5; (The +1.0 Is Just There to Force Conversion to Float...
I'm trying the following query:
SELECT (json_data->'position'->'lat') + 1.0 AS lat FROM updates LIMIT 5;
(The +1.0 is just there to force conversion to float. My actual queries are far more complex, this query is just a test case for the problem.)
I get the error:
ERROR: operator does not exist: jsonb + numeric
If I add in explicit casting:
SELECT (json_data->'position'->'lat')::float + 1.0 AS lat FROM updates LIMIT 5;
the error becomes:
ERROR: operator does not exist: jsonb + double precesion
I understand that most jsonb values cannot be cast into floats, but in this case I know that the lats are all JSON numbers.
Is there a function which casts jsonb values to floats (or return NULLs for the uncastable)?
7 Answers
There are two operations to get value from JSON. The first one -> will return JSON. The second one ->> will return text.
Details: JSON Functions and Operators
Try
SELECT (json_data->'position'->>'lat')::float + 1.0 AS lat
FROM updates
LIMIT 5
AFAIK there's no json->float casting in Postgres, so you could try an explicit (json_data->'position'->'lat')::text::float cast
Per documentation, there are also the functions
jsonb_populate_record()
jsonb_populate_recordset()
Analog to their json twins (present since pg 9.3)
json_populate_record()
json_populate_recordset()
You need a predefined row type. Either use the row-type of an existing table or define one with CREATE TYPE. Or substitute with a temporary table ad hoc:
CREATE TEMP TABLE x(lat float);
Can be a single column or a long list of columns.
Only those columns are filled, where the name matches a key in the json object. The value is coerced to the column type and has to be compatible or an exception is raised. Other keys are ignored.
SELECT lat + 1 -- no need for 1.0, this is float already
FROM updates u
, jsonb_populate_record(NULL::x, u.json_data->'position')
LIMIT 5;
Using an implicit LATERAL JOIN here.
Similarly, use jsonb_populate_recordset() to decompose arrays into multiple rows per entry.
This works the same way in Postgres 9.3 with json. There is the added benefit that casting to / from text internally is not necessary for numeric data in jsonb.
Must Read
Now we can do it!
In nowadays we can cast directly from JSONb to SQL datatypes. I am using PostgreSQL v12.3, where it is working fine:
SELECT (j->'i')::int, (j->>'i')::int, (j->'f')::float, (j->>'f')::float
FROM (SELECT '{"i":123,"f":12.34}'::jsonb) t(j);
Sub-questions:
From which version is it possible?
It is a syntax sugar or a real conversion?
If real "binary JSONb → binary SQL" conversion, where the micro-optimizations?
For example, what wold be faster (?) tham "binary JSONb → string → binary SQL"? boolean→boolean, number→numeric, number→int, number→bigint; number→flloat, number→double.Why not optimized for NULL?
Curiosily the "NULL to SqlType" not works, "ERROR: cannot cast jsonb null to type integer".