[takayamaki]

Part of #215 (item 9).

A schema without type skipped every keyword.
For example, this schema accepted { "count": "x" } and {}:

required: [count]
properties:
  count:
    type: integer

In JSON Schema, which OpenAPI 3.1 follows, a keyword applies when the value has the type the keyword is for.
type does not need to be set.
So properties and required apply to an object, and maxLength applies to a string.

Changes

• When type is not set, the validator picks the type from the value, and then validates the value in the same way as a schema with that type.
  It reuses pick_array_type, renamed to pick_type, which already does this for array-form type.
• A keyword for another type is still ignored.
  For example, properties does not reject a string.
• A value of any other class, like a Date, is still accepted without checks.

I checked the same cases with three well-known JSON Schema validators.
They all give the same results, for both Draft 2020-12 and Draft 4:
ajv 8.20.0, python-jsonschema 4.26.0 and json_schemer 2.5.0.

schema	value	the three validators	before	after
type: object with properties	1	invalid	invalid	invalid
properties without type	1	valid	valid	valid
properties without type	{ "count": 1 }	valid	valid	valid
properties without type	{ "count": "x" }	invalid	valid	invalid

properties here is { count: { type: integer } }.
Only the last row differed from the three validators before this change.

I did not check the declared version.
The OpenAPI 3.0 Schema Object also follows JSON Schema here, so the same fix is right for 3.0 documents.

Compatibility

This makes validation stricter.
A value that breaks a keyword in a schema without type used to pass, and now fails.
I added a note in the CHANGELOG.
I did not add a config option, but I can add one if you prefer.

The committee test suite passes with this branch.

null for a schema without type is not changed here.
It is still rejected, as described in #215 item 5.

[geemus]

lgtm, thanks!

[geemus]

lgtm, thanks!