Disable uri-reference format check in jsonsschema (#2771)

* Disable uri-reference format check. Consistently use same validator across codebase

* Remove validation in SchemaInfo as not used anywhere and it referenced the wrong jsonschema draft

* Add compatibility for older jsonschema versions

* Improve comments

* Simplify validate_jsonschema

altair/utils/display.py

@@ -4,10 +4,9 @@
 from typing import Callable, Dict
 import uuid
 
-from jsonschema import validate
-
 from .plugin_registry import PluginRegistry
 from .mimebundle import spec_to_mimebundle
+from .schemapi import validate_jsonschema
 
 
 # ==============================================================================
@@ -126,7 +125,7 @@ def _validate(self):
         # type: () -> None
         """Validate the spec against the schema."""
         schema_dict = json.loads(pkgutil.get_data(*self.schema_path).decode("utf-8"))
-        validate(self.spec, schema_dict)
+        validate_jsonschema(self.spec, schema_dict)
 
     def _repr_mimebundle_(self, include=None, exclude=None):
         """Return a MIME bundle for display in Jupyter frontends."""

altair/utils/schemapi.py

@@ -8,12 +8,13 @@
 from typing import Any
 
 import jsonschema
+import jsonschema.exceptions
 import numpy as np
 import pandas as pd
 
 from altair import vegalite
 
-
+JSONSCHEMA_VALIDATOR = jsonschema.Draft7Validator
 # If DEBUG_MODE is True, then schema objects are converted to dict and
 # validated at creation time. This slows things down, particularly for
 # larger specs, but leads to much more useful tracebacks for the user.
@@ -43,6 +44,24 @@ def debug_mode(arg):
         DEBUG_MODE = original
 
 
+def validate_jsonschema(spec, schema, resolver=None):
+    # We don't use jsonschema.validate as this would validate the schema itself.
+    # Instead, we pass the schema directly to the validator class. This is done for
+    # two reasons: The schema comes from Vega-Lite and is not based on the user
+    # input, therefore there is no need to validate it in the first place. Furthermore,
+    # the "uri-reference" format checker fails for some of the references as URIs in
+    # "$ref" are not encoded,
+    # e.g. '#/definitions/ValueDefWithCondition<MarkPropFieldOrDatumDef,
+    # (Gradient|string|null)>' would be a valid $ref in a Vega-Lite schema but
+    # it is not a valid URI reference due to the characters such as '<'.
+    validator = JSONSCHEMA_VALIDATOR(
+        schema, format_checker=JSONSCHEMA_VALIDATOR.FORMAT_CHECKER, resolver=resolver
+    )
+    error = jsonschema.exceptions.best_match(validator.iter_errors(spec))
+    if error is not None:
+        raise error
+
+
 def _subclasses(cls):
     """Breadth-first sequence of all classes which inherit from cls."""
     seen = set()
@@ -158,7 +177,7 @@ class SchemaBase(object):
     _schema = None
     _rootschema = None
     _class_is_valid_at_instantiation = True
-    _validator = jsonschema.Draft7Validator
+    _validator = JSONSCHEMA_VALIDATOR
 
     def __init__(self, *args, **kwds):
         # Two valid options for initialization, which should be handled by
@@ -448,9 +467,7 @@ def validate(cls, instance, schema=None):
         if schema is None:
             schema = cls._schema
         resolver = jsonschema.RefResolver.from_schema(cls._rootschema or cls._schema)
-        return jsonschema.validate(
-            instance, schema, cls=cls._validator, resolver=resolver
-        )
+        return validate_jsonschema(instance, schema, resolver=resolver)
 
     @classmethod
     def resolve_references(cls, schema=None):
@@ -469,7 +486,7 @@ def validate_property(cls, name, value, schema=None):
         value = _todict(value, validate=False, context={})
         props = cls.resolve_references(schema or cls._schema).get("properties", {})
         resolver = jsonschema.RefResolver.from_schema(cls._rootschema or cls._schema)
-        return jsonschema.validate(value, props.get(name, {}), resolver=resolver)
+        return validate_jsonschema(value, props.get(name, {}), resolver=resolver)
 
     def __dir__(self):
         return list(self._kwds.keys())
@@ -563,7 +580,7 @@ def from_dict(
             for possible_schema in schemas:
                 resolver = jsonschema.RefResolver.from_schema(rootschema)
                 try:
-                    jsonschema.validate(dct, possible_schema, resolver=resolver)
+                    validate_jsonschema(dct, possible_schema, resolver=resolver)
                 except jsonschema.ValidationError:
                     continue
                 else:

tools/schemapi/schemapi.py

@@ -6,12 +6,13 @@
 from typing import Any
 
 import jsonschema
+import jsonschema.exceptions
 import numpy as np
 import pandas as pd
 
 from altair import vegalite
 
-
+JSONSCHEMA_VALIDATOR = jsonschema.Draft7Validator
 # If DEBUG_MODE is True, then schema objects are converted to dict and
 # validated at creation time. This slows things down, particularly for
 # larger specs, but leads to much more useful tracebacks for the user.
@@ -41,6 +42,24 @@ def debug_mode(arg):
         DEBUG_MODE = original
 
 
+def validate_jsonschema(spec, schema, resolver=None):
+    # We don't use jsonschema.validate as this would validate the schema itself.
+    # Instead, we pass the schema directly to the validator class. This is done for
+    # two reasons: The schema comes from Vega-Lite and is not based on the user
+    # input, therefore there is no need to validate it in the first place. Furthermore,
+    # the "uri-reference" format checker fails for some of the references as URIs in
+    # "$ref" are not encoded,
+    # e.g. '#/definitions/ValueDefWithCondition<MarkPropFieldOrDatumDef,
+    # (Gradient|string|null)>' would be a valid $ref in a Vega-Lite schema but
+    # it is not a valid URI reference due to the characters such as '<'.
+    validator = JSONSCHEMA_VALIDATOR(
+        schema, format_checker=JSONSCHEMA_VALIDATOR.FORMAT_CHECKER, resolver=resolver
+    )
+    error = jsonschema.exceptions.best_match(validator.iter_errors(spec))
+    if error is not None:
+        raise error
+
+
 def _subclasses(cls):
     """Breadth-first sequence of all classes which inherit from cls."""
     seen = set()
@@ -156,7 +175,7 @@ class SchemaBase(object):
     _schema = None
     _rootschema = None
     _class_is_valid_at_instantiation = True
-    _validator = jsonschema.Draft7Validator
+    _validator = JSONSCHEMA_VALIDATOR
 
     def __init__(self, *args, **kwds):
         # Two valid options for initialization, which should be handled by
@@ -446,9 +465,7 @@ def validate(cls, instance, schema=None):
         if schema is None:
             schema = cls._schema
         resolver = jsonschema.RefResolver.from_schema(cls._rootschema or cls._schema)
-        return jsonschema.validate(
-            instance, schema, cls=cls._validator, resolver=resolver
-        )
+        return validate_jsonschema(instance, schema, resolver=resolver)
 
     @classmethod
     def resolve_references(cls, schema=None):
@@ -467,7 +484,7 @@ def validate_property(cls, name, value, schema=None):
         value = _todict(value, validate=False, context={})
         props = cls.resolve_references(schema or cls._schema).get("properties", {})
         resolver = jsonschema.RefResolver.from_schema(cls._rootschema or cls._schema)
-        return jsonschema.validate(value, props.get(name, {}), resolver=resolver)
+        return validate_jsonschema(value, props.get(name, {}), resolver=resolver)
 
     def __dir__(self):
         return list(self._kwds.keys())
@@ -561,7 +578,7 @@ def from_dict(
             for possible_schema in schemas:
                 resolver = jsonschema.RefResolver.from_schema(rootschema)
                 try:
-                    jsonschema.validate(dct, possible_schema, resolver=resolver)
+                    validate_jsonschema(dct, possible_schema, resolver=resolver)
                 except jsonschema.ValidationError:
                     continue
                 else:

tools/schemapi/utils.py

@@ -1,8 +1,6 @@
 """Utilities for working with schemas"""
 
-import json
 import keyword
-import pkgutil
 import re
 import textwrap
 import urllib
@@ -13,12 +11,6 @@
 EXCLUDE_KEYS = ("definitions", "title", "description", "$schema", "id")
 
 
-def load_metaschema():
-    schema = pkgutil.get_data("schemapi", "jsonschema-draft04.json")
-    schema = schema.decode()
-    return json.loads(schema)
-
-
 def resolve_references(schema, root=None):
     """Resolve References within a JSON schema"""
     resolver = jsonschema.RefResolver.from_schema(root or schema)
@@ -144,18 +136,14 @@ def values(self):
 class SchemaInfo(object):
     """A wrapper for inspecting a JSON schema"""
 
-    def __init__(self, schema, rootschema=None, validate=False):
+    def __init__(self, schema, rootschema=None):
         if hasattr(schema, "_schema"):
             if hasattr(schema, "_rootschema"):
                 schema, rootschema = schema._schema, schema._rootschema
             else:
                 schema, rootschema = schema._schema, schema._schema
         elif not rootschema:
             rootschema = schema
-        if validate:
-            metaschema = load_metaschema()
-            jsonschema.validate(schema, metaschema)
-            jsonschema.validate(rootschema, metaschema)
         self.raw_schema = schema
         self.rootschema = rootschema
         self.schema = resolve_references(schema, rootschema)