State OAS binary data semantics

README.rst

@@ -128,52 +128,66 @@ unresolved-metaschema fallback warnings.
    assert validator_for(schema32) is OAS32Validator
 
 
-Strict vs Pragmatic Validators
-==============================
+Binary Data Semantics
+=====================
 
-OpenAPI 3.0 has two validator variants with different behaviors for binary format:
+The handling of binary-like payloads differs between OpenAPI versions.
 
-**OAS30Validator (default - pragmatic)**
-   - Accepts Python ``bytes`` for ``type: string`` with ``format: binary``
-   - More lenient for Python use cases where binary data is common
-   - Use when validating Python objects directly
+OpenAPI 3.0
+-----------
+
+OpenAPI 3.0 keeps historical ``format: binary`` / ``format: byte`` usage on
+``type: string``.
+
+**OAS30Validator (default - compatibility behavior)**
+   - ``type: string`` accepts ``str``
+   - ``type: string, format: binary`` accepts Python ``bytes`` and strings
+   - useful when validating Python-native runtime data
 
 **OAS30StrictValidator**
-   - Follows OAS spec strictly: only accepts ``str`` for ``type: string``
-   - For ``format: binary``, only accepts base64-encoded strings
-   - Use when strict spec compliance is required
+   - ``type: string`` accepts ``str`` only
+   - ``type: string, format: binary`` uses strict format validation
+   - use when you want strict, spec-oriented behavior for 3.0 schemas
+
+OpenAPI 3.1+
+------------
+
+OpenAPI 3.1+ follows JSON Schema semantics for string typing in this library.
 
-Comparison Matrix
------------------
+- ``type: string`` accepts ``str`` only (not ``bytes``)
+- ``format: binary`` and ``format: byte`` are not treated as built-in formats
+- for base64-in-JSON, model with ``contentEncoding: base64`` (optionally
+  ``contentMediaType``)
+- for raw binary payloads, model via media type (for example
+  ``application/octet-stream``) rather than schema string formats
+
+Quick Reference
+---------------
 
 .. list-table::
    :header-rows: 1
-   :widths: 35 20 22 23
-
-   * - Schema
-     - Value
-     - OAS30Validator (default)
-     - OAS30StrictValidator
-   * - ``type: string``
-     - ``"test"`` (str)
-     - Pass
-     - Pass
-   * - ``type: string``
-     - ``b"test"`` (bytes)
-     - **Fail**
-     - **Fail**
-   * - ``type: string, format: binary``
-     - ``b"test"`` (bytes)
+   :widths: 28 24 24 24
+
+   * - Context
+     - ``"text"`` (str)
+     - ``b"text"`` (bytes)
+     - Notes
+   * - OAS 3.0 + ``OAS30Validator``
      - Pass
-     - **Fail**
-   * - ``type: string, format: binary``
-     - ``"dGVzdA=="`` (base64)
+     - Pass for ``format: binary``
+     - Compatibility behavior for Python runtime payloads
+   * - OAS 3.0 + ``OAS30StrictValidator``
      - Pass
+     - Fail
+     - Strict 3.0 validation mode
+   * - OAS 3.1 + ``OAS31Validator``
      - Pass
-   * - ``type: string, format: binary``
-     - ``"test"`` (plain str)
+     - Fail
+     - Use ``contentEncoding``/``contentMediaType`` and media types
+   * - OAS 3.2 + ``OAS32Validator``
      - Pass
-     - **Fail**
+     - Fail
+     - Same semantics as OAS 3.1
 
 For more details read about `Validation <https://openapi-schema-validator.readthedocs.io/en/latest/validation.html>`__.
 

docs/validation.rst

@@ -166,54 +166,68 @@ OpenAPI 3.0 schema comes with ``readOnly`` and ``writeOnly`` keywords. In order
        ...
    ValidationError: Tried to write read-only property with 23
 
-Strict vs Pragmatic Validators
-------------------------------
+Binary Data Semantics
+---------------------
 
-OpenAPI 3.0 has two validator variants with different behaviors for binary format:
+The handling of binary-like payloads differs between OpenAPI versions.
 
-**OAS30Validator (default - pragmatic)**
+OpenAPI 3.0
+~~~~~~~~~~~
 
-- Accepts Python ``bytes`` for ``type: string`` with ``format: binary``
-- More lenient for Python use cases where binary data is common
-- Use when validating Python objects directly
+OpenAPI 3.0 keeps historical ``format: binary`` / ``format: byte`` usage on
+``type: string``.
+
+**OAS30Validator (default - compatibility behavior)**
+
+- ``type: string`` accepts ``str``
+- ``type: string, format: binary`` accepts Python ``bytes`` and strings
+- useful when validating Python-native runtime data
 
 **OAS30StrictValidator**
 
-- Follows OAS spec strictly: only accepts ``str`` for ``type: string``
-- For ``format: binary``, only accepts base64-encoded strings
-- Use when strict spec compliance is required
+- ``type: string`` accepts ``str`` only
+- ``type: string, format: binary`` uses strict format validation
+- use when you want strict, spec-oriented behavior for 3.0 schemas
+
+OpenAPI 3.1+
+~~~~~~~~~~~~
+
+OpenAPI 3.1+ follows JSON Schema semantics for string typing in this library.
 
-Comparison Matrix
-~~~~~~~~~~~~~~~~~
+- ``type: string`` accepts ``str`` only (not ``bytes``)
+- ``format: binary`` and ``format: byte`` are not treated as built-in formats
+- for base64-in-JSON, model with ``contentEncoding: base64`` (optionally
+  ``contentMediaType``)
+- for raw binary payloads, model via media type (for example
+  ``application/octet-stream``) rather than schema string formats
+
+Quick Reference
+~~~~~~~~~~~~~~~
 
 .. list-table::
    :header-rows: 1
-   :widths: 35 20 22 23
-
-   * - Schema
-     - Value
-     - OAS30Validator (default)
-     - OAS30StrictValidator
-   * - ``type: string``
-     - ``"test"`` (str)
-     - Pass
-     - Pass
-   * - ``type: string``
-     - ``b"test"`` (bytes)
-     - **Fail**
-     - **Fail**
-   * - ``type: string, format: binary``
-     - ``b"test"`` (bytes)
+   :widths: 28 24 24 24
+
+   * - Context
+     - ``"text"`` (str)
+     - ``b"text"`` (bytes)
+     - Notes
+   * - OAS 3.0 + ``OAS30Validator``
      - Pass
-     - **Fail**
-   * - ``type: string, format: binary``
-     - ``"dGVzdA=="`` (base64)
+     - Pass for ``format: binary``
+     - Compatibility behavior for Python runtime payloads
+   * - OAS 3.0 + ``OAS30StrictValidator``
      - Pass
+     - Fail
+     - Strict 3.0 validation mode
+   * - OAS 3.1 + ``OAS31Validator``
      - Pass
-   * - ``type: string, format: binary``
-     - ``"test"`` (plain str)
+     - Fail
+     - Use ``contentEncoding``/``contentMediaType`` and media types
+   * - OAS 3.2 + ``OAS32Validator``
      - Pass
-     - **Fail**
+     - Fail
+     - Same semantics as OAS 3.1
 
 Example usage:
 

tests/integration/test_validators.py

@@ -851,6 +851,26 @@ def test_string_disallow_binary(self, validator_class, value):
         with pytest.raises(ValidationError):
             validator.validate(value)
 
+    @pytest.mark.parametrize("value", [b"test"])
+    def test_string_binary_rejects_bytes(
+        self, validator_class, format_checker, value
+    ):
+        schema = {"type": "string", "format": "binary"}
+        validator = validator_class(schema, format_checker=format_checker)
+
+        with pytest.raises(ValidationError):
+            validator.validate(value)
+
+    @pytest.mark.parametrize("value", [True, 3, 3.12, None])
+    def test_string_binary_invalid(
+        self, validator_class, format_checker, value
+    ):
+        schema = {"type": "string", "format": "binary"}
+        validator = validator_class(schema, format_checker=format_checker)
+
+        with pytest.raises(ValidationError):
+            validator.validate(value)
+
     @pytest.mark.parametrize(
         "schema_type",
         [