Migration from v1

datason v2 is a ground-up rewrite. The API has changed significantly.

Quick Summary

v1	v2
datason.serialize(obj)	datason.dumps(obj)
datason.deserialize(data)	datason.loads(json_str)
datason.dump_ml(obj)	datason.dumps(obj, **ml_config().__dict__)
datason.load_smart(data)	datason.loads(json_str)
100+ public functions	5 functions: dumps, loads, dump, load, config
Python 3.8+	Python 3.10+
Monolithic core	Plugin-based architecture

Key Changes

1. API reduced to 5 functions. All the dump_ml, load_smart, load_perfect, save_string variants are gone. Use datason.dumps() with config overrides or presets instead.

2. Configuration via kwargs or context manager, not separate functions:

   # v1
   datason.dump_ml(model_output)
   
   # v2
   from datason import ml_config
   with datason.config(**ml_config().__dict__):
       datason.dumps(model_output)
   # or inline:
   datason.dumps(model_output, date_format=DateFormat.UNIX_MS, fallback_to_string=True)
   

3. Type metadata key changed from various formats to a consistent {"__datason_type__": "...", "__datason_value__": ...} envelope.

4. Python 3.10 minimum. Drops 3.8/3.9 support. Uses match/case and X | Y union syntax.

5. Thread-safe by default. Config uses contextvars.ContextVar, registry uses threading.Lock.