DataType JSON auto convert to String

[motionrus]

To reproduce the problem, I created a dataset from the abc table

create table public.abc
(
  source_id varchar(255) not null,
  json jsonb,
  constraint pk_source_id primary key(source_id)
);

insert into abc (source_id, json) VALUES (1, '{"key": "value"}');

then I checked that my column type was json.

and when I get an object for handlebars, I get very upset that I get a string instead of an object.

and I can't do {{this.json.key}} because there is a string in the json

{"source_id":"1","json":"{'key': 'value'}"}

[gregersn]

After digging a bit in the source code, I found that this is related to the lack of support for complex data types as can be seen here:

superset/superset/utils/core.py

Line 158 in 419d122

# ARRAY = 4 # Mapping all the complex data types to STRING for now

and here:

superset/superset/result_set.py

Line 155 in 419d122

# TODO: revisit nested column serialization once nested types

And is mentioned in related issue, here: #19388

[albuscopperplate]

would be great this to work!

[hf-13]

+1

[rusackas]

@villebro left the latter of those comments and may have some idea how deep this rabbit hole goes. Obviously, we're open to contributions if you want to keep digging into the code :D

[rusackas]

We're still open to contributions on this one, otherwise it may be closed as inactive before terribly long.

[rusackas]

Still welcoming contributions on this one! For anyone interested in picking it up, here's some context on the scope:

The GenericDataType enum in superset/utils/core.py has JSON, ARRAY, MAP, and ROW types intentionally commented out, with complex types mapped to STRING as a stopgap. Additionally, result_set.py explicitly stringifies nested types via PyArrow. Uncommenting those enum values is the easy part — the real work involves updating the db_engine_specs to map JSON column types, preserving JSON structure through the serialization layer, and adding frontend support for rendering/handling JSON columns across charts, tables, and filters.

It's a meaningful effort, but would be a great improvement for anyone working with JSONB/nested data. PRs welcome!