GenSON

GenSON is a powerful, user-friendly JSON Schema generator built in Python.

Note

This is not the Python equivalent of the Java Genson library. If you are coming from Java and need to create JSON objects in Python, you want Python's builtin json library.)

GenSON's core function is to take JSON objects and generate schemas that describe them, but it is unique in its ability to merge schemas. It was originally built to describe the common structure of a large number of JSON objects, and it uses its merging ability to generate a single schema from any number of JSON objects and/or schemas.

GenSON's schema builder follows these three rules:

1. Every object it is given must validate under the generated schema.
2. Any object that is valid under any schema it is given must also validate under the generated schema. (there is one glaring exception to this, detailed below)
3. The generated schema should be as strict as possible given the first 2 rules.

JSON Schema Implementation

GenSON is compatible with JSON Schema Draft 6 and above.

It is important to note that GenSON uses only a subset of JSON Schema's capabilities. This is mainly because it doesn't know the specifics of your data model, and it tries to avoid guessing them. Its purpose is to generate the basic structure so that you can skip the boilerplate and focus on the details of the schema.

Currently, GenSON only deals with these keywords:

• "$schema"
• "type"
• "items"
• "properties"
• "patternProperties"
• "required"
• "anyOf"

You should be aware that this limited vocabulary could cause GenSON to violate rules 1 and 2. If you feed it schemas with advanced keywords, it will just blindly pass them on to the final schema. Note that "$ref" and id are also not supported, so GenSON will not dereference linked nodes when building a schema.

Installation

$ pip install genson

CLI Tool

The package includes a genson executable that allows you to access this functionality from the command line. For usage info, run with --help:

$ genson --help

usage: genson [-h] [--version] [-d DELIM] [-e ENCODING] [-i SPACES]
              [-s SCHEMA] [-$ SCHEMA_URI]
              ...

Generate one, unified JSON Schema from one or more JSON objects and/or JSON
Schemas. Compatible with JSON-Schema Draft 4 and above.

positional arguments:
  object                Files containing JSON objects (defaults to stdin if no
                        arguments are passed).

optional arguments:
  -h, --help            Show this help message and exit.
  --version             Show version number and exit.
  -d DELIM, --delimiter DELIM
                        Set a delimiter. Use this option if the input files
                        contain multiple JSON objects/schemas. You can pass
                        any string. A few cases ('newline', 'tab', 'space')
                        will get converted to a whitespace character. If this
                        option is omitted, the parser will try to auto-detect
                        boundaries.
  -e ENCODING, --encoding ENCODING
                        Use ENCODING instead of the default system encoding
                        when reading files. ENCODING must be a valid codec
                        name or alias.
  -i SPACES, --indent SPACES
                        Pretty-print the output, indenting SPACES spaces.
  -s SCHEMA, --schema SCHEMA
                        File containing a JSON Schema (can be specified
                        multiple times to merge schemas).
  -$ SCHEMA_URI, --schema-uri SCHEMA_URI
                        The value of the '$schema' keyword (defaults to
                        'http://json-schema.org/schema#' or can be specified
                        in a schema with the -s option). If 'NULL' is passed,
                        the "$schema" keyword will not be included in the
                        result.

Note

The --encoding option is only available in Python 3.

GenSON Python API

SchemaBuilder is the basic schema generator class. SchemaBuilder instances can be loaded up with existing schemas and objects before being serialized.

>>> from genson import SchemaBuilder

>>> builder = SchemaBuilder()
>>> builder.add_schema({"type": "object", "properties": {}})
>>> builder.add_object({"hi": "there"})
>>> builder.add_object({"hi": 5})

>>> builder.to_schema()
{'$schema': 'http://json-schema.org/schema#',
 'type': 'object',
 'properties': {
    'hi': {'type': ['integer', 'string']}},
    'required': ['hi']}

>>> print(builder.to_json(indent=2))
{
  "$schema": "http://json-schema.org/schema#",
  "type": "object",
  "properties": {
    "hi": {
      "type": [
        "integer",
        "string"
      ]
    }
  },
  "required": [
    "hi"
  ]
}

SchemaBuilder API

__init__(schema_uri=None)

param schema_uri:	value of the $schema keyword. If not given, it will use the value of the first available $schema keyword on an added schema or else the default: 'http://json-schema.org/schema#'. A value of False or None will direct GenSON to leave out the "$schema" keyword.

add_schema(schema)

Merge in a JSON schema. This can be a dict or another SchemaBuilder object.

param schema:	a JSON Schema

Note

There is no schema validation. If you pass in a bad schema, you might get back a bad schema.

add_object(obj)

Modify the schema to accommodate an object.

param obj:	any object or scalar that can be serialized in JSON

to_schema()

Generate a schema based on previous inputs.

rtype:	dict

to_json()

Generate a schema and convert it directly to serialized JSON.

rtype:	str

__eq__(other)

Check for equality with another SchemaBuilder object.

param other:	another SchemaBuilder object. Other types are accepted, but will always return False

SchemaBuilder object interaction

SchemaBuilder objects can also interact with each other:

• You can pass one schema directly to another to merge them.
• You can compare schema equality directly.

>>> from genson import SchemaBuilder

>>> b1 = SchemaBuilder()
>>> b1.add_schema({"type": "object", "properties": {
...   "hi": {"type": "string"}}})
>>> b2 = SchemaBuilder()
>>> b2.add_schema({"type": "object", "properties": {
...   "hi": {"type": "integer"}}})
>>> b1 == b2
False

>>> b1.add_schema(b2)
>>> b2.add_schema(b1)
>>> b1 == b2
True
>>> b1.to_schema()
{'$schema': 'http://json-schema.org/schema#',
 'type': 'object',
 'properties': {'hi': {'type': ['integer', 'string']}}}

Seed Schemas

There are several cases where multiple valid schemas could be generated from the same object. GenSON makes a default choice in all these ambiguous cases, but if you want it to choose differently, you can tell it what to do using a seed schema.

Seeding Arrays

For example, suppose you have a simple array with two items:

['one', 1]

There are always two ways for GenSON to interpret any array: List and Tuple. Lists have one schema for every item, whereas Tuples have a different schema for every array position. This is analogous to the (now deprecated) merge_arrays option from version 0. You can read more about JSON Schema array validation here.

List Validation

{
  "type": "array",
  "items": {"type": ["integer", "string"]}
}

Tuple Validation

{
  "type": "array",
  "items": [{"type": "integer"}, {"type": "string"}]
}

By default, GenSON always interprets arrays using list validation, but you can tell it to use tuple validation by seeding it with a schema.

>>> from genson import SchemaBuilder

>>> builder = SchemaBuilder()
>>> builder.add_object(['one', 1])
>>> builder.to_schema()
{'$schema': 'http://json-schema.org/schema#',
 'type': 'array',
 'items': {'type': ['integer', 'string']}}

>>> builder = SchemaBuilder()
>>> seed_schema = {'type': 'array', 'items': []}
>>> builder.add_schema(seed_schema)
>>> builder.add_object(['one', 1])
>>> builder.to_schema()
{'$schema': 'http://json-schema.org/schema#',
 'type': 'array',
 'items': [{'type': 'string'}, {'type': 'integer'}]}

Note that in this case, the seed schema is actually invalid. You can't have an empty array as the value for an items keyword. But GenSON is a generator, not a validator, so you can fudge a little. GenSON will modify the generated schema so that it is valid, provided that there aren't invalid keywords beyond the ones it knows about.

Seeding patternProperties

Support for patternProperties is new in version 1; however, since GenSON's default behavior is to only use properties, this powerful keyword can only be utilized with seed schemas. You will need to supply an object schema with a patternProperties object whose keys are RegEx strings. Again, you can fudge here and set the values to null instead of creating valid subschemas.

>>> from genson import SchemaBuilder

>>> builder = SchemaBuilder()
>>> builder.add_schema({'type': 'object', 'patternProperties': {r'^\d+$': None}})
>>> builder.add_object({'1': 1, '2': 2, '3': 3})
>>> builder.to_schema()
{'$schema': 'http://json-schema.org/schema#', 'type': 'object', 'patternProperties':  {'^\\d+$': {'type': 'integer'}}}

There are a few gotchas you should be aware of here:

• GenSON is written in Python, so it uses the Python flavor of RegEx.
• GenSON still prefers properties to patternProperties if a property already exists that matches one of your patterns, the normal property will be updated, not the pattern property.
• If a key matches multiple patterns, there is no guarantee of which one will be updated.
• The patternProperties docs themselves have some more useful pointers that can save you time.

Typeless Schemas

In version 0, GenSON did not accept a schema without a type, but in order to be flexible in the support of seed schemas, support was added for version 1. However, GenSON violates rule #2 in its handling of typeless schemas. Any object will validate under an empty schema, but GenSON incorporates typeless schemas into the first-available typed schema, and since typed schemas are stricter than typless ones, objects that would validate under an added schema will not validate under the result.

Customizing SchemaBuilder

You can extend the SchemaBuilder class to add in your own logic (e.g. recording minimum and maximum for a number). In order to do this, you need to:

1. Create a custom SchemaStrategy class.
2. Create a SchemaBuilder subclass that includes your custom SchemaStrategy class(es).
3. Use your custom SchemaBuilder just like you would the stock SchemaBuilder.

SchemaStrategy Classes

GenSON uses the Strategy Pattern to parse, update, and serialize different kinds of schemas that behave in different ways. There are several SchemaStrategy classes that roughly correspond to different schema types. GenSON maps each node in an object or schema to an instance of one of these classes. Each instance stores the current schema state and updates or returns it when required.

You can modify the specific ways these classes work by extending them. You can inherit from any existing SchemaStrategy class, though SchemaStrategy and TypedSchemaStrategy are the most useful base classes. You should call super and pass along all arguments when overriding any instance methods.

The documentation below explains the public API and what you need to extend and override at a high level. Feel free to explore the code to see more, but know that the public API is documented here, and anything else you depend on could be subject to change. All SchemaStrategy subclasses maintain the public API though, so you can extend any of them in this way.

SchemaStrategy API

[class constant] KEYWORDS

This should be a tuple listing all of the JSON-schema keywords that this strategy knows how to handle. Any keywords encountered in added schemas will be be naively passed on to the generated schema unless they are in this list (or you override that behavior in to_schema).

When adding keywords to a new SchemaStrategy, it's best to splat the parent class's KEYWORDS into the new tuple.

[class method] match_schema(cls, schema)

Return true if this strategy should be used to handle the passed-in schema.

param schema:	a JSON Schema in dict form
rtype:	bool

[class method] match_object(cls, obj)

Return true if this strategy should be used to handle the passed-in object.

param obj:	any object or scalar that can be serialized in JSON
rtype:	bool

__init__(self, node_class)

Override this method if you need to initialize an instance variable.

param node_class:	This param is not part of the public API. Pass it along to super.

add_schema(self, schema)

Override this to modify how a schema is parsed and stored.

param schema:	a JSON Schema in dict form

add_object(self, obj)

Override this to change the way a schemas are inferred from objects.

param obj:	any object or scalar that can be serialized in JSON

to_schema(self)

Override this method to customize how a schema object is constructed from the inputs. It is suggested that you invoke super as the basis for the return value, but it is not required.

rtype:	dict

Note

There is no schema validation. If you return a bad schema from this method, SchemaBuilder will output a bad schema.

__eq__(self, other)

When checking for SchemaBuilder equality, strategies are matched using __eq__. The default implementation uses a simple __dict__ equality check.

Override this method if you need to override that behavior. This may be useful if you add instance variables that aren't relevant to whether two SchemaStrategies are considered equal.

rtype:	bool

TypedSchemaStrategy API

This is an abstract schema strategy for making simple schemas that only deal with the type keyword, but you can extend it to add more functionality. Subclasses must define the following two class constants, but you get the entire SchemaStrategy interface for free.

[class constant] JS_TYPE

This will be the value of the type keyword in the generated schema. It is also used to match any added schemas.

[class constant] PYTHON_TYPE

This is a Python type or tuple of types that will be matched against an added object using isinstance.

Extending SchemaBuilder

Once you have extended SchemaStrategy types, you'll need to create a SchemaBuilder class that uses them, since the default SchemaBuilder only incorporates the default strategies. To do this, extend the SchemaBuilder class and define one of these two constants inside it:

[class constant] EXTRA_STRATEGIES

This is the standard (and suggested) way to add strategies. Set it to a tuple of all your new strategies, and they will be added to the existing list of strategies to check. This preserves all the existing functionality.

Note that order matters. GenSON checks the list in order, so the first strategy has priority over the second and so on. All EXTRA_STRATEGIES have priority over the default strategies.

[class constant] STRATEGIES

This clobbers the existing list of strategies and completely replaces it. Set it to a tuple just like for EXTRA_STRATEGIES, but note that if any object or schema gets added that your exhaustive list of strategies doesn't know how to handle, you'll get an error. You should avoid doing this unless you're extending most or all existing strategies in some way.

Example: MinNumber

Here's some example code creating a number strategy that tracks the minimum number seen and includes it in the output schema.

Note

This example is written in Python 3.3+. Custom strategies also work in Python 2.7, but you need different syntax (super arguments & no splatting KEYWORDS).

from genson import SchemaBuilder
from genson.schema.strategies import Number

class MinNumber(Number):
    # add 'minimum' to list of keywords
    KEYWORDS = (*Number.KEYWORDS, 'minimum')

    # create a new instance variable
    def __init__(self, node_class):
        super().__init__(node_class)
        self.min = None

    # capture 'minimum's from schemas
    def add_schema(self, schema):
        super().add_schema(schema)
        if self.min is None:
            self.min = schema.get('minimum')
        elif 'minimum' in schema:
            self.min = min(self.min, schema['minimum'])

    # adjust minimum based on the data
    def add_object(self, obj):
        super().add_object(obj)
        self.min = obj if self.min is None else min(self.min, obj)

    # include 'minimum' in the output
    def to_schema(self):
        schema = super().to_schema()
        schema['minimum'] = self.min
        return schema

# new SchemaBuilder class that uses the MinNumber strategy in addition
# to the existing strategies. Both MinNumber and Number are active, but
# MinNumber has priority, so it effectively replaces Number.
class MinNumberSchemaBuilder(SchemaBuilder):
    """ all number nodes include minimum """
    EXTRA_STRATEGIES = (MinNumber,)

# this class *ONLY* has the MinNumber strategy. Any object that is not
# a number will cause an error.
class ExclusiveMinNumberSchemaBuilder(SchemaBuilder):
    """ all number nodes include minimum, and only handles number """
    STRATEGIES = (MinNumber,)

Now that we have the MinNumberSchemaBuilder class, let's see how it works.

>>> builder = MinNumberSchemaBuilder()
>>> builder.add_object(5)
>>> builder.add_object(7)
>>> builder.to_schema()
{'$schema': 'http://json-schema.org/schema#', 'type': 'integer', 'minimum': 5}
>>> builder.add_object(-2)
>>> builder.to_schema()
{'$schema': 'http://json-schema.org/schema#', 'type': 'integer', 'minimum': -2}
>>> builder.add_schema({'$schema': 'http://json-schema.org/schema#', 'type': 'integer', 'minimum': -7})
>>> builder.to_schema()
{'$schema': 'http://json-schema.org/schema#', 'type': 'integer', 'minimum': -7}

Note that the exclusive builder is much more particular.

>>> builder = MinNumberSchemaBuilder()
>>> picky_builder = ExclusiveMinNumberSchemaBuilder()
>>> picky_builder.add_object(5)
>>> picky_builder.to_schema()
{'$schema': 'http://json-schema.org/schema#', 'type': 'integer', 'minimum': 5}
>>> builder.add_object(None) # this is fine
>>> picky_builder.add_object(None) # this fails
genson.schema.node.SchemaGenerationError: Could not find matching schema type for object: None

Contributing

When contributing, please follow these steps:

1. Clone the repo and make your changes.
2. Make sure your code has test cases written against it.
3. Lint your code with Flake8.
4. Run tox to make sure the test suite passes.
5. Ensure the docs are accurate.
6. Add your name to the list of contributers.
7. Submit a Pull Request.

Tests

Tests are written in unittest and are run using tox and `nose`_. Tox will run all tests with coverage against each supported Python version that is installed on your machine.

$ tox

You should always run tox before submitting a PR, but it takes some time, so when developing, it may be faster just to run nosetests in your own environment. e.g.

$ nosetests
$ nosetests --with-coverate --cover-package=genson
$ nosetests test.test_gen_single:TestBasicTypes.test_number

Integration

When you submit a PR, Travis CI performs the following steps:

1. Lints the code with Flake8
2. Runs the entire test suite against each supported Python version.
3. Ensures that test coverage is at least 90%

If any of these steps fail, your PR cannot be merged until it is fixed.

Potential Future Features

The following are extra features under consideration.

• recognize every validation keyword and ignore any that don't apply
• option to set error level
• custom serializer plugins
• logical support for more keywords:
  • enum
  • minimum/maximum
  • minLength/maxLength
  • minItems/maxItems
  • minProperties/maxProperties
  • additionalItems
  • additionalProperties
  • format & pattern
  • $ref & id