blitzjson

Drop-in replacement for Python's json module with native Django type support. Built with Rust via PyO3 for maximum performance.

Features

• Drop-in replacement: Same API as json.dumps(), json.loads(), etc.
• Native Django types: Handles datetime, date, time, timedelta, UUID, Decimal, QuerySet, Model, Promise without custom encoders
• Rust-powered: Direct serialization to JSON buffer via CPython FFI, no intermediate Python objects
• Zero dependencies: No runtime Python dependencies required
• Full json API: ensure_ascii, indent, sort_keys, allow_nan, default (recursive)
• Streaming: stream_dump_queryset() for memory-efficient large QuerySet serialization
• Django integration: BlitzJsonResponse, BlitzJSONEncoder, install() monkey-patch

Installation

pip install blitzjson

Quick Start

# Before
import json
from datetime import datetime

json.dumps({"created": datetime.now()})  # TypeError!

# After
import blitzjson as json

json.dumps({"created": datetime.now()})  # Works!

Monkey-patching (try before you commit)

If you want to test blitzjson in your existing project without changing any import json statements, use the install() function to monkey-patch Python's built-in json module:

# In your settings.py or at the top of manage.py
import blitzjson
blitzjson.install()

# Now ALL code that uses `import json` will use blitzjson instead
import json
json.dumps({"created": datetime.now()})  # Works! No TypeError!

Or test it interactively:

>>> import blitzjson
>>> blitzjson.install()
>>> import json
>>> json.dumps({"created": datetime.now(), "uuid": uuid4()})
'{"created": "2024-01-15T10:30:45+00:00", "uuid": "550e8400-e29b-41d4-a716-446655440000"}'

>>> # Revert when done testing
>>> blitzjson.uninstall()

Supported Types

Django-Specific

API

dumps(obj, **kwargs)

import blitzjson as json

# Standard usage
json.dumps({"key": "value"})

# With Django types
json.dumps({"created": datetime.now(), "uuid": uuid4()})

loads(s, **kwargs)

obj = json.loads('{"key": "value"}')

# With object_hook
obj = json.loads('{"key": "value"}', object_hook=lambda d: {k.upper(): v for k, v in d.items()})

dump(obj, fp, **kwargs) / load(fp, **kwargs)

File-based serialization/deserialization.

dumpb(obj, pretty=False)

Serialize to bytes (faster than dumps for network responses).

dump_queryset(queryset) / dump_queryset_bytes(queryset)

Optimized serialization for Django QuerySets.

Benchmarks

vs json + DjangoJSONEncoder (CPython 3.14, Linux x86_64, consumer hardware):

Note: Benchmarks run on consumer hardware (AMD Ryzen 9 7900X). CI runners (Intel Xeon server CPUs) may show different results due to lower single-thread performance. For stable benchmarks, run locally with uv run python benchmarks/bench_serialization.py.

dumps

Benchmark                           json+DJE    blitzjson    Speedup
------------------------------------------------------------------------
Simple dict (4 fields)                 2.5µs        0.8µs      3.0x
Nested dict (deep)                     5.4µs        1.8µs      3.0x
Large list (1000 items)              752.1µs      193.2µs      3.9x
String-heavy dict (20 keys)            4.8µs        1.5µs      3.1x
Datetime dict                         10.3µs        3.6µs      2.8x
UUID dict                              7.3µs        3.9µs      1.9x
Decimal dict                           4.4µs        1.6µs      2.7x
Mixed dict (all types)                 7.3µs        2.8µs      2.6x

loads

Benchmark                               json    blitzjson    Speedup
------------------------------------------------------------------------
Simple dict (4 fields)                 1.5µs        0.8µs      2.0x
Nested dict (deep)                     5.1µs        3.2µs      1.6x
Large list (1000 items)              716.0µs      421.2µs      1.7x
String-heavy dict (20 keys)            4.3µs        2.7µs      1.6x

Django Response

Benchmark                             JsonResponse      BlitzJson    Speedup
------------------------------------------------------------------------
API response (50 users)                 44.9µs         19.6µs      2.3x

Requirements

• Python 3.10+
• No Rust installation required (pre-built wheels)

License

MIT - Ricardo Robles Fernández