jsonscreamer 0.5.0

Json Screamer is a fast JSON Schema validation library built with a few goals in mind:

1. fast - up to 10x faster than the de-facto standard jsonschema library
2. correct - full compliance with the json schema test suite, except some gnarly $ref edge cases
3. easy to maintain - pure python code, regular function calls etc.
4. iteration over all errors for a given schema

Currently it only handles the Draft 7 spec. If you want a more battle-tested and robust implementation, use jsonschema. If you want an even faster implementation use fastjsonschema (up to 2x quicker). The jsonscreamer library sits somewhere in between: more correct than fastjsonschema (e.g. counting [0, False] as having unique items) and faster than jsonschema.

Our primary motivations for not using fastjsonschema were correctness, security and ability to customise by writing regular python code. If the idea of dynamically creating source code and calling exec on it makes you (or your security team) uncomfortable that's probably a big reason not to use fastjsonschema.

Benchmarks

The small benchmark from our test suite gives the following numbers (under python3.11):

In real-world usage with large schemas we've seen an ~14x speedup over jsonschema. For these benchmarks, since fastjsonschema accepts nonsense datetimes like 2025-19-39T29:00:00Z as valid by default, it made sense to explicitly set all 3 validators' date-time format checkers as the same (we'd need to do this extra check anyway if using fastjsonschema):

Usage

For good performance, create a single Validator instance and call its methods many times:

>>> from jsonscreamer import Validator
>>> from jsonscreamer.format import is_date_time_iso
>>> Validator.check_schema({"type": "string"})
>>> val = Validator({"type": "string"})
>>> val.is_valid(1)
False
>>> val.is_valid("1")
True
>>> list(val.iter_errors("1"))
[]
>>> val.validate(1)
Traceback (most recent call last):
  ...
jsonscreamer.types.ValidationError: ((), "1 is not of type 'string'", 'type')

>>> # Compliant format checkers are run by default, but this one is faster
>>> # if you're OK with python's default ISO8601 instead of RFC3339:
>>> schema = {"type": "string", "format": "date-time"}
>>> val = Validator(schema, formats={"date-time": is_date_time_iso})
>>> val.is_valid("2020-01-01 01:02:03")
True

The Validator class has 3 primary methods:

• is_valid(instance) which returns True or False
• validate(instance) which will raise a ValidationError if any error is encountered
• iter_errors(instance) which gives an iterator of ValidationErrors for the instance

and the ValidationError class itself has the following properties:

• absolute_path: the path to the error within the item (e.g. ("spam", 0, "eggs"))
• message: a human-readable error message
• type: the type of validation error

Custom formats

By default jsonscreamer installs with only a subset of format validators (those which require no external dependencies), you can install all format checkers via pip install jsonscreamer[all-formats]. The Validator class also takes a formats parameter which you can set to False to disable all formats, or provide a dictionary of your own formats to override or add to the existing formats.

When providing a dictionary of custom formats, the the keys are the format names (like "email") and the values are validators which recieve a string and return either True or False, depending on if it conforms to the format.

For example:

def check_contains_spam(item: str) -> bool:
    return "spam" in item

spammy_validator = Validator(
    {"type": "string", "format": "spam"},
    formats={"spam": check_contians_spam},
)

print(spammy_validator.is_valid("spam spam spam spam"))  # True
print(spammy_validator.is_valid("eggs"))  # False

or, disabling the format checks completely:

>>> from jsonscreamer import Validator
>>> val = Validator({"type": "string", "format": "date-time"}, formats=False)
>>> val.is_valid("not a date")
True

Custom ref handlers

The Validator class takes a handlers parameter which can be used to specify how to handle $ref URIs. By default we will attempt to download any schema with a URI starting with http or https - to everride this behaviour you must provide a dictionary where the keys are the URI scheme and the values are functions which take a URI and provide a json schema.

To forbid remote refs you can write something like the following:

def raising_handler(uri: str) -> None:
    raise ValueError(f"cannot resolve remote ref: {uri}")

local_validator = Validator(
    ..., handlers={"http": raising_handler, "https": raising_handler},
)

or if you have already downloaded all of the remote schemas you could just add a local lookup function of your choosing.

Test suite compliance

For the Draft 7 schema test suite, we pass 210 out of 212 tests. We consider the two failures to be very niche cases to do with relative $ref resolution in the "definitions" section. We are currently more compliant than fastjsonschema, and for almost all real-world schemas this should be considered complete.

Roadmap

Resolver: currently we are using a subclass of fastjsonschema's resolver for ref resolution. We've added a few compatibility hacks to pass more of the json schema test site. We'd like to move to something more robust.

2019 Draft: the 2019 draft is on our roadmap once ref resolution is sorted.

2020 Draft: the 2020 draft is on our roadmap after the 2019 draft is sorted.

Earlier Drafts: we might consider this if there is demand.