json-backed-dict

A dict subclass that automatically persists every mutation to a JSON file.

from json_backed_dict import JsonBackedDict as JBD

d = JBD('config.json')
d['host'] = 'localhost'   # written to disk immediately
d['port'] = 5432          # written to disk immediately

Features

• Automatic persistence — every __setitem__, __delitem__, update, pop, etc. atomically writes the file
• Nested mutation tracking — mutating a nested dict or list is also persisted
• Temporal type round-tripping — datetime, date, time, and timedelta values survive save/load cycles as their Python types
• Atomic writes — uses a temp file + os.replace so a crash mid-write never corrupts the file
• Fast — backed by orjson for serialization

Installation

pip install json-backed-dict

Requires Python 3.9+.

Usage

Basic operations

from json_backed_dict import JsonBackedDict as JBD

d = JBD('data.json')

d['name'] = 'Alice'
d['scores'] = [10, 20, 30]
d.update({'active': True, 'level': 5})

print(d['name'])      # 'Alice'
print(d.get('age'))   # None
del d['level']

All standard dict methods work: keys(), values(), items(), pop(), setdefault(), popitem(), clear(), update().

Loading an existing file

If the file already exists, it is loaded on construction. The initial argument is ignored when a file is present.

d = JBD('data.json')          # loads existing file
d = JBD('data.json', initial={'x': 1})  # initial ignored, file loaded

Seeding a new file

d = JBD('settings.json', initial={'debug': False, 'timeout': 30})

initial is only used when creating a new file.

Nested mutations

Nested dicts and lists returned by __getitem__ and get() are proxy objects. Mutating them persists the change to the root file automatically.

d = JBD('data.json', initial={'config': {'timeout': 10}, 'tags': ['a', 'b']})

d['config']['timeout'] = 30   # persisted
d['tags'].append('c')         # persisted
d['config'].update({'retries': 3})  # persisted

Temporal types

datetime, date, time, and timedelta values are serialized to strings and deserialized back to their Python types on load. No manual conversion needed.

from datetime import datetime, date, time, timedelta

d = JBD('data.json')
d['created_at'] = datetime(2024, 6, 15, 10, 30, 45)
d['due_date']   = date(2024, 7, 1)
d['start_time'] = time(9, 0)
d['ttl']        = timedelta(hours=24)

d2 = JBD('data.json')
isinstance(d2['created_at'], datetime)   # True
isinstance(d2['ttl'], timedelta)         # True

Note: This means strings that look like dates, times, or timedeltas cannot be stored as plain strings — they will be coerced to the corresponding Python type on the next load. For example, storing "2024-01-15" will be read back as date(2024, 1, 15).

Supported value types

All dict keys must be str. Attempting to store any other type raises TypeError before any mutation occurs.

Limitations

Thread-safe, but not process-safe. Methods implemented by JsonBackedDict acquire an instance-level lock for their full duration, making individual operations on a single instance atomic. Concurrent writes from multiple processes can still interleave: each write creates a temp file and calls os.replace, so the last writer wins and earlier changes are silently lost.

Not an IPC mechanism. If another process modifies the backing file, the current instance will never see those changes. To pick up external changes, construct a new JBD from the same path.

License

MIT