JSON Model is a compact and intuitive JSON syntax to describe JSON data structures.

This reference implementation allows to generate code in Python, C, JavaScript, PL/pgSQL, Perl and Java for checking a JSON value against a JSON model, and to export models to JSON Schema or Pydantic.

It is dedicated to the Public Domain.

JSON Model optimizing compiler (jmc) can be installed as a Python package or a Docker image, see Installation HOWTO.

Command jmc options include:

• main operations (default depends on other options, final guess is preprocess):
  • -P: preprocess model.
  • -C: compile to Python, C, JS, PL/pgSQL, Perl, Java.
  • -E: export to JSON Schema version draft 2020-12 or Pydantic.
• -O: optimize model: constant propagation, partial evaluation, xor to or conversion, flattening… (this is the default, -nO to disable)
• -o output: file output instead of standard
• …

For instance, let's consider a JSON model in file person.model.json:

{
  "#": "A person with a birth date",
  "name": "/^[a-z]+$/i",
  "born": "$DATE"
}

• to check directly sample JSON values against it (with the Python backend):

  jmc -r person.model.json hobbes.json oops.json

  hobbes.json: PASS
  oops.json: FAIL (.: not an expected object [.]; .: missing mandatory prop <born> [.])
  

• to compile an executable for checking a model (with the C backend), and use it for validating values:

  jmc -o ./person.out person.model.json
  ./person.out -r hobbes.json oops.json

  hobbes.json: PASS
  oops.json: FAIL (.: not an expected object [.]; .: missing mandatory prop <born> [.])
  

  The generated executable allow to collect validation performance figures (average and standard deviation) over a loop, with or without reporting:

  ./person.out -r -T 100000 hobbes.json

  hobbes.json.[0] nop PASS 0.056 ± 0.423 µs/check (0.174)
  hobbes.json.[0] rep PASS 0.071 ± 0.443 µs/check (0.174)
  hobbes.json: PASS
  

• to export this model as a JSON schema in the YaML format:

  jmc -E -F yaml person.model.json

  description: A person with a birth date
  type: object
  properties:
    name:
      type: string
      pattern: (?i)^[a-z]+$
    born:
      type: string
      format: date
  required:
  - name
  - born
  additionalProperties: false

The package provides functions to create and check models from Python:

import json_model as jm

# direct model definition with 2 mandatory properties
person_model: jm.Jsonable = {
  "name": "/^[a-z]+$/i",
  "born": "$DATE"
}

# create a dynamically compiled checker function for the model
checker = jm.model_checker_from_json(person_model)

# check valid data
good_person = { "name": "Hobbes", "born": "2020-07-29" }
print(good_person, "->", checker(good_person))

# check invalid data
bad_person = { "name": "Oops" }
print(bad_person, "->", checker(bad_person))

# collect reasons
reasons: jm.Report = []
assert not checker(bad_person, "", reasons)
print("reasons:", reasons)

See the benchmark page for artifacts which compare various JSON Model Compiler runs (C, JS, Java, Python) with Sourcemeta Blaze CLI as a baseline using test cases from JSON Schema Benchmark. Overall, JMC-C implementation fares over 50% faster than Blaze C++. It comes ahead 80% of the time on the latest benchmarks tests. Moreover, JMC-JS and JMC-Java/GSON native implementations are only 25-30% slower than Blaze C++, which given the intrinsic language capabilities is quite honorable.

See the JSON Model website, which among many resources, includes a tutorial for a hands-on overview of JSON Model, and links to research papers for explanations about the design.

• PYPI Package
• Docker Image
• JS Runtime
• Postgres Runtime
• Perl Runtime
• Java Runtime