hypothesis-jsonschema

• never generate trailing newlines for regex patterns ending in $ (allowed by Python, but not by JSON Schema)

• reduced filtering for object keys (#88)
• updated to jsonschema >= 4.0.0 (#89); though support for Draft 2019-09 and 2020-12 will take longer
• requires Python 3.7+, a few months ahead of the 3.6 end of life date

• improved handling for fractional multipleOf values

• Allow custom_formats for known string formats (#83)

• Fix support date and time formats (#79)

• PyPy support (#77)

• Generate empty lists when maxItems > 0 but no elements are allowed (#75)
• Correct handling of regex patterns which are invalid in Python (#75)

• Remove internal caching due to hash collisions (#71)
• Improve performance for conditional keywords

• Canonicalise anyOf special cases when all subschemas have only the type keyword

• Performance improvements from careful caching (#62)
• Use a validator that corresponds to the input schema draft version (#66)

• fixed string schemas with different format keywords (#63)

• improved handling of overlapping items keywords (#58)

• improved handling of overlapping dependencies keywords (#57)

• fixed an internal bug where results incorrectly depended on iteration order (#59)

• Adds a custom_formats keyword argument to from_schema(), so that you can specify a strategy to generate strings for custom formats like credit card numbers. Thanks to Dmitry Dygalo, whose sponsorship motivated me to add the feature!

• Substantial performance gains for some schemas, via improved handling of the contains, not, anyOf, and if/then/else keywords

• Performance improvement for object schemas with additionalProperties: false (issue #55)

• Performance improvement for schemas with non-validation keys (such as description)
• Errors from e.g. invalid schemas are deferred from import time to become failing tests
• Improved handling for some schemas with overlapping non-integer multipleOf keys

• Significantly improved efficiency of certain patternProperties schemas.

• Fixed several bugs related to interactions between properties, patternProperties, and additionalProperties. As a result some strategies will be more efficient than before and others less; and further gains seem likely.

• Improved strategy for json-pointer and relative-json-pointer string formats
• Improved generation of arrays with rarely-satisfied contains constraints
• Improved canonicalisation and merging of allOf schemas

• Performance improvement in calculating schema intersections

• Improved canonicalisation of uniqueItems: false case
• Improved canonicalisation of numeric schemas
• Reuse jsonschema validators during canonicalisation (performance improvement)

• Added a strategy for the "color" format
• Only apply string length filter when needed (small performance improvement)

• Fixed error in resolution of certain $ref\ s
• Improved canonicalisation of anyOf and contains keys

• Requires Hypothesis >= 5.3.0, for improved IP address strategies
• Better canoncialisation of array schemata

• Resolve local, non-recursive references via the $ref keyword.

This is the largest feature in a while, and for some schemata it is a breaking change. It's also a fundamental part of the spec, so I'm OK with that!

Note that hypothesis-jsonschema will raise an explicit error rather than fetching a remote resource via URI. If think your tests really should hit the network, get in touch and we can discuss adding an off-by-default option for this.

• Improved canonicalisation of conflicting minProperties and maxProperties
• Explictly reject draft-03 schemata, which are not supported
• More specific type annotations for from_schema
• Better performance for certain object schemata

• enum schema now shrink to a minimal example rather than the first value listed. This also makes the internals more efficient in certain rare cases.
• Improved handling of bounded numeric sub-schemas

• Improved handling of non-integer numeric schemas with multipleOf
• Improved handling of not in many common cases
• Improved handling of object schemas with dependencies on required keys
• Fixed cases where propertyNames bans a required key

• Improved handling of numeric schemas, especially integer schemas with multipleOf.
• Bump the minimum version of Hypothesis to ensure that all users have the unique array bugfix from 0.9.12.
• Fixed a bug where array schemas with an array of items, additionalItems: false, and a maxItems larger than the number of allowed items would resolve to an invalid strategy.

• Improved internal handling of schemas for arrays which must always be length-zero.

• Fixed RFC 3339 strings generation. Thanks to Dmitry Dygalo for the patch!
• Fixed a bug where equal floats and ints could be generated in a unique array, even though JSONSchema considers 0 === 0.0 (though this may also require an upstream fix to work...)

• Fixed a bug where objects which could have either zero or one properties would always be generated with zero.

• Updated project metadata and development tooling
• Supported and tested on Python 3.8

• Correct handling of {"items": [...], "uniqueItems": true"} schemas

• Corrected handling of the "format" keyword with unknown values - custom values are allowed by the spec and should be treated as annotations (i.e. ignored).

• Improved canonicalisation, especially for deeply nested schemas.

• A performance optimisation for null and boolean schema, which relies on a bugfix in jsonschema >= 3.0.2.

• Improved handling of the contains keyword for arrays

• Improved canonicalisation and merging for a wide range of schemas, which as usual unlocks significant optimisations and performance improvements for cases where they apply.

• Future-proofed canonicalisation of type key.

• Better internal canonicalization, which makes current and future optimisations more widely applicable.
• Yet another fix, this time for negative zero and numeric bouds as floats with sub-integer precision. IEEE 754 is tricky, even with Hypothesis!
• Fixes handling of enum with elements disallowed by base schema, handling of if-then-else with a base schema, and handling of regex patterns that are invalid in Python.

• Fix the fix for numeric schemas with multipleOf and exclusive bounds.

• Supports merging schemas for overlapping patternProperties, a significant performance improvement in most cases.
• If the "type" key is missing, it is now inferred from other keys rather than always defaulting to "object".
• Fixed handling of complicated numeric bounds.

• Improve performance for object schemas where the min and max size can be further constrained from properties and propertyNames attributes.

• Supports draft-04 schemata with the latest version of jsonschema

• Further improved support for allOf, oneOf, and anyOf with base schemata
• Added support for dependencies
• Handles overlapping patternProperties

• Now requires jsonschema >= 3.0
• Improved support for allOf, oneOf, and propertyNames
• Supports schemata with "type": [an array of types]
• Warning-free on Hypothesis 4.11

• Fix continuous delivery configuration (before the latent bug manifested)

• Support for conditional subschemata, i.e. the if, then, else keywords, and the anyOf, allOf, oneOf, and not keywords.

• Works with jsonschema 3.0 pre-release
• Initial support for draft06 and draft07

• Dropped dependency on canonicaljson
• Less warnings on Python 3.7

• Relicensed under the more permissive Mozilla Public License, like Hypothesis
• Requires Hypothesis version 4.0 or later
• Fixed an array bounds bug with maxItems and contains keywords

Supports string formats (email, datetime, etc) and simple use of the "contains" keyword for arrays.

Good support for all basic types. MVP.

Inference for null, boolean, string, and numeric types.