jtd: JSON Validation for Python

JSON Type Definition, aka RFC 8927, is an easy-to-learn, standardized way to define a schema for JSON data. You can use JSON Typedef to portably validate data across programming languages, create dummy data, generate code, and more.

This jtd package is a Python implementation of JSON Type Definition, a schema language for JSON. It lets you validate input data against JSON Type Definition schemas.

If you're looking to generate code from schemas, check out "Generating Python from JSON Typedef schemas" in the JSON Typedef docs.

Installation

You can install this package with pip:

pip install jtd

Documentation

Detailed API documentation is available online at:

https://jtd.readthedocs.io

For more high-level documentation about JSON Typedef in general, or JSON Typedef in combination with Python in particular, see:

• The JSON Typedef Website
• "Generating Python from JSON Typedef schemas"

Basic Usage

Here's an example of how you can use this package to validate JSON data against a JSON Typedef schema:

import jtd

schema = jtd.Schema.from_dict({
    'properties': {
        'name': { 'type': 'string' },
        'age': { 'type': 'uint32' },
        'phones': {
            'elements': {
                'type': 'string'
            }
        }
    }
})

# jtd.validate returns an array of validation errors. If there were no problems
# with the input, it returns an empty array.

# Outputs: []
print(jtd.validate(schema=schema, instance={
  'name': 'John Doe',
  'age': 43,
  'phones': ['+44 1234567', '+44 2345678'],
}))

# This next input has three problems with it:
#
# 1. It's missing "name", which is a required property.
# 2. "age" is a string, but it should be an integer.
# 3. "phones[1]" is a number, but it should be a string.
#
# Each of those errors corresponds to one of the errors returned by validate.

# Outputs:
#
# [
#   ValidationError(
#     instance_path=[], schema_path=['properties', 'name']
#   ),
#   ValidationError(
#     instance_path=['age'], schema_path=['properties', 'age', 'type']
#   ),
#   ValidationError(
#     instance_path=['phones', '1'], schema_path=['properties', 'phones', 'elements', 'type']
#   ),
# ]
print(jtd.validate(schema=schema, instance={
  'age': "43",
  'phones': ["+44 1234567", 442345678],
}))

Advanced Usage: Limiting Errors Returned

By default, jtd.validate returns every error it finds. If you just care about whether there are any errors at all, or if you can't show more than some number of errors, then you can get better performance out of jtd.validate using the max_errors option.

For example, taking the same example from before, but limiting it to 1 error, we get:

# Outputs:
#
# [ValidationError(instance_path=[], schema_path=['properties', 'name'])]
options = jtd.ValidationOptions(max_errors=1)
print(jtd.validate(schema=schema, options=options, instance={
  'age': '43',
  'phones': ['+44 1234567', 442345678],
}))

Advanced Usage: Handling Untrusted Schemas

If you want to run jtd against a schema that you don't trust, then you should:

1. Ensure the schema is well-formed, using the validate() method on jtd.Schema. That will check things like making sure all refs have corresponding definitions.

2. Call jtd.validate with the max_depth option. JSON Typedef lets you write recursive schemas -- if you're evaluating against untrusted schemas, you might go into an infinite loop when evaluating against a malicious input, such as this one:

   {
     "ref": "loop",
     "definitions": {
       "loop": {
         "ref": "loop"
       }
     }
   }

   The max_depth option tells jtd.validate how many refs to follow recursively before giving up and throwing jtd.MaxDepthExceededError.

Here's an example of how you can use jtd to evaluate data against an untrusted schema:

import jtd

# validate_untrusted returns true if `data` satisfies `schema`, and false if it
# does not. Throws an error if `schema` is invalid, or if validation goes in an
# infinite loop.
def validate_untrusted(schema, data):
    schema.validate()

    # You should tune max_depth to be high enough that most legitimate schemas
    # evaluate without errors, but low enough that an attacker cannot cause a
    # denial of service attack.
    options = jtd.ValidationOptions(max_depth=32)
    return len(jtd.validate(schema=schema, instance=data, options=options)) == 0
}

# Returns true
validate_untrusted(jtd.Schema.from_dict({ 'type': 'string' }), 'foo')

# Returns false
validate_untrusted(jtd.Schema.from_dict({ 'type': 'string' }), None)

# Throws "invalid schema"
validate_untrusted(jtd.Schema.from_dict({ 'type': 'nonsense' }), 'foo')

# Throws an instance of jtd.MaxDepthExceededError
validate_untrusted({
  "ref": "loop",
  "definitions": {
    "loop": {
      "ref": "loop"
    }
  }
}, None)