Fixed #12990, Refs #27694 -- Added JSONField model field.

Thanks to Adam Johnson, Carlton Gibson, Mariusz Felisiak, and Raphael
Michel for mentoring this Google Summer of Code 2019 project and
everyone else who helped with the patch.

Special thanks to Mads Jensen, Nick Pope, and Simon Charette for
extensive reviews.

Co-authored-by: Mariusz Felisiak <felisiak.mariusz@gmail.com>

django/db/backends/base/features.py

@@ -300,6 +300,15 @@ class BaseDatabaseFeatures:
     # Does the backend support boolean expressions in the SELECT clause?
     supports_boolean_expr_in_select_clause = True
 
+    # Does the backend support JSONField?
+    supports_json_field = True
+    # Can the backend introspect a JSONField?
+    can_introspect_json_field = True
+    # Does the backend support primitives in JSONField?
+    supports_primitives_in_json_field = True
+    # Is there a true datatype for JSON?
+    has_native_json_field = False
+
     def __init__(self, connection):
         self.connection = connection
 

django/db/backends/base/operations.py

@@ -159,6 +159,13 @@ class BaseDatabaseOperations:
         """
         return self.date_extract_sql(lookup_type, field_name)
 
+    def json_cast_text_sql(self, field_name):
+        """Return the SQL to cast a JSON value to text value."""
+        raise NotImplementedError(
+            'subclasses of BaseDatabaseOperations may require a '
+            'json_cast_text_sql() method'
+        )
+
     def deferrable_sql(self):
         """
         Return the SQL to make a constraint "initially deferred" during a

django/db/backends/mysql/base.py

@@ -118,6 +118,7 @@ class DatabaseWrapper(BaseDatabaseWrapper):
         'BigIntegerField': 'bigint',
         'IPAddressField': 'char(15)',
         'GenericIPAddressField': 'char(39)',
+        'JSONField': 'json',
         'NullBooleanField': 'bool',
         'OneToOneField': 'integer',
         'PositiveBigIntegerField': 'bigint UNSIGNED',
@@ -341,11 +342,16 @@ class DatabaseWrapper(BaseDatabaseWrapper):
     @cached_property
     def data_type_check_constraints(self):
         if self.features.supports_column_check_constraints:
-            return {
+            check_constraints = {
                 'PositiveBigIntegerField': '`%(column)s` >= 0',
                 'PositiveIntegerField': '`%(column)s` >= 0',
                 'PositiveSmallIntegerField': '`%(column)s` >= 0',
             }
+            if self.mysql_is_mariadb and self.mysql_version < (10, 4, 3):
+                # MariaDB < 10.4.3 doesn't automatically use the JSON_VALID as
+                # a check constraint.
+                check_constraints['JSONField'] = 'JSON_VALID(`%(column)s`)'
+            return check_constraints
         return {}
 
     @cached_property

django/db/backends/mysql/features.py

@@ -160,3 +160,15 @@ class DatabaseFeatures(BaseDatabaseFeatures):
     def supports_default_in_lead_lag(self):
         # To be added in https://jira.mariadb.org/browse/MDEV-12981.
         return not self.connection.mysql_is_mariadb
+
+    @cached_property
+    def supports_json_field(self):
+        if self.connection.mysql_is_mariadb:
+            return self.connection.mysql_version >= (10, 2, 7)
+        return self.connection.mysql_version >= (5, 7, 8)
+
+    @cached_property
+    def can_introspect_json_field(self):
+        if self.connection.mysql_is_mariadb:
+            return self.supports_json_field and self.can_introspect_check_constraints
+        return self.supports_json_field

django/db/backends/mysql/introspection.py

@@ -9,7 +9,7 @@ from django.db.backends.base.introspection import (
 from django.db.models import Index
 from django.utils.datastructures import OrderedSet
 
-FieldInfo = namedtuple('FieldInfo', BaseFieldInfo._fields + ('extra', 'is_unsigned'))
+FieldInfo = namedtuple('FieldInfo', BaseFieldInfo._fields + ('extra', 'is_unsigned', 'has_json_constraint'))
 InfoLine = namedtuple('InfoLine', 'col_name data_type max_len num_prec num_scale extra column_default is_unsigned')
 
 
@@ -24,6 +24,7 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
         FIELD_TYPE.DOUBLE: 'FloatField',
         FIELD_TYPE.FLOAT: 'FloatField',
         FIELD_TYPE.INT24: 'IntegerField',
+        FIELD_TYPE.JSON: 'JSONField',
         FIELD_TYPE.LONG: 'IntegerField',
         FIELD_TYPE.LONGLONG: 'BigIntegerField',
         FIELD_TYPE.SHORT: 'SmallIntegerField',
@@ -53,6 +54,10 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
                 return 'PositiveIntegerField'
             elif field_type == 'SmallIntegerField':
                 return 'PositiveSmallIntegerField'
+        # JSON data type is an alias for LONGTEXT in MariaDB, use check
+        # constraints clauses to introspect JSONField.
+        if description.has_json_constraint:
+            return 'JSONField'
         return field_type
 
     def get_table_list(self, cursor):
@@ -66,6 +71,19 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
         Return a description of the table with the DB-API cursor.description
         interface."
         """
+        json_constraints = {}
+        if self.connection.mysql_is_mariadb and self.connection.features.can_introspect_json_field:
+            # JSON data type is an alias for LONGTEXT in MariaDB, select
+            # JSON_VALID() constraints to introspect JSONField.
+            cursor.execute("""
+                SELECT c.constraint_name AS column_name
+                FROM information_schema.check_constraints AS c
+                WHERE
+                    c.table_name = %s AND
+                    LOWER(c.check_clause) = 'json_valid(`' + LOWER(c.constraint_name) + '`)' AND
+                    c.constraint_schema = DATABASE()
+            """, [table_name])
+            json_constraints = {row[0] for row in cursor.fetchall()}
         # information_schema database gives more accurate results for some figures:
         # - varchar length returned by cursor.description is an internal length,
         #   not visible length (#5725)
@@ -100,6 +118,7 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
                 info.column_default,
                 info.extra,
                 info.is_unsigned,
+                line[0] in json_constraints,
             ))
         return fields
 

django/db/backends/mysql/operations.py

@@ -368,3 +368,13 @@ class DatabaseOperations(BaseDatabaseOperations):
 
     def insert_statement(self, ignore_conflicts=False):
         return 'INSERT IGNORE INTO' if ignore_conflicts else super().insert_statement(ignore_conflicts)
+
+    def lookup_cast(self, lookup_type, internal_type=None):
+        lookup = '%s'
+        if internal_type == 'JSONField':
+            if self.connection.mysql_is_mariadb or lookup_type in (
+                'iexact', 'contains', 'icontains', 'startswith', 'istartswith',
+                'endswith', 'iendswith', 'regex', 'iregex',
+            ):
+                lookup = 'JSON_UNQUOTE(%s)'
+        return lookup

django/db/backends/oracle/base.py

@@ -123,6 +123,7 @@ class DatabaseWrapper(BaseDatabaseWrapper):
         'FilePathField': 'NVARCHAR2(%(max_length)s)',
         'FloatField': 'DOUBLE PRECISION',
         'IntegerField': 'NUMBER(11)',
+        'JSONField': 'NCLOB',
         'BigIntegerField': 'NUMBER(19)',
         'IPAddressField': 'VARCHAR2(15)',
         'GenericIPAddressField': 'VARCHAR2(39)',
@@ -141,6 +142,7 @@ class DatabaseWrapper(BaseDatabaseWrapper):
     }
     data_type_check_constraints = {
         'BooleanField': '%(qn_column)s IN (0,1)',
+        'JSONField': '%(qn_column)s IS JSON',
         'NullBooleanField': '%(qn_column)s IN (0,1)',
         'PositiveBigIntegerField': '%(qn_column)s >= 0',
         'PositiveIntegerField': '%(qn_column)s >= 0',

django/db/backends/oracle/features.py

@@ -60,3 +60,4 @@ class DatabaseFeatures(BaseDatabaseFeatures):
     supports_slicing_ordering_in_compound = True
     allows_multiple_constraints_on_same_fields = False
     supports_boolean_expr_in_select_clause = False
+    supports_primitives_in_json_field = False

django/db/backends/oracle/introspection.py

@@ -7,7 +7,7 @@ from django.db.backends.base.introspection import (
     BaseDatabaseIntrospection, FieldInfo as BaseFieldInfo, TableInfo,
 )
 
-FieldInfo = namedtuple('FieldInfo', BaseFieldInfo._fields + ('is_autofield',))
+FieldInfo = namedtuple('FieldInfo', BaseFieldInfo._fields + ('is_autofield', 'is_json'))
 
 
 class DatabaseIntrospection(BaseDatabaseIntrospection):
@@ -45,6 +45,8 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
                     return 'IntegerField'
             elif scale == -127:
                 return 'FloatField'
+        elif data_type == cx_Oracle.NCLOB and description.is_json:
+            return 'JSONField'
 
         return super().get_field_type(data_type, description)
 
@@ -83,12 +85,23 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
                 CASE
                     WHEN identity_column = 'YES' THEN 1
                     ELSE 0
-                END as is_autofield
+                END as is_autofield,
+                CASE
+                    WHEN EXISTS (
+                        SELECT  1
+                        FROM user_json_columns
+                        WHERE
+                            user_json_columns.table_name = user_tab_cols.table_name AND
+                            user_json_columns.column_name = user_tab_cols.column_name
+                    )
+                    THEN 1
+                    ELSE 0
+                END as is_json
             FROM user_tab_cols
             WHERE table_name = UPPER(%s)""", [table_name])
         field_map = {
-            column: (internal_size, default if default != 'NULL' else None, is_autofield)
-            for column, default, internal_size, is_autofield in cursor.fetchall()
+            column: (internal_size, default if default != 'NULL' else None, is_autofield, is_json)
+            for column, default, internal_size, is_autofield, is_json in cursor.fetchall()
         }
         self.cache_bust_counter += 1
         cursor.execute("SELECT * FROM {} WHERE ROWNUM < 2 AND {} > 0".format(
@@ -97,11 +110,11 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
         description = []
         for desc in cursor.description:
             name = desc[0]
-            internal_size, default, is_autofield = field_map[name]
+            internal_size, default, is_autofield, is_json = field_map[name]
             name = name % {}  # cx_Oracle, for some reason, doubles percent signs.
             description.append(FieldInfo(
                 self.identifier_converter(name), *desc[1:3], internal_size, desc[4] or 0,
-                desc[5] or 0, *desc[6:], default, is_autofield,
+                desc[5] or 0, *desc[6:], default, is_autofield, is_json,
             ))
         return description
 

django/db/backends/oracle/operations.py

@@ -176,7 +176,7 @@ END;
     def get_db_converters(self, expression):
         converters = super().get_db_converters(expression)
         internal_type = expression.output_field.get_internal_type()
-        if internal_type == 'TextField':
+        if internal_type in ['JSONField', 'TextField']:
             converters.append(self.convert_textfield_value)
         elif internal_type == 'BinaryField':
             converters.append(self.convert_binaryfield_value)
@@ -269,7 +269,7 @@ END;
         return tuple(columns)
 
     def field_cast_sql(self, db_type, internal_type):
-        if db_type and db_type.endswith('LOB'):
+        if db_type and db_type.endswith('LOB') and internal_type != 'JSONField':
             return "DBMS_LOB.SUBSTR(%s)"
         else:
             return "%s"
@@ -307,6 +307,8 @@ END;
     def lookup_cast(self, lookup_type, internal_type=None):
         if lookup_type in ('iexact', 'icontains', 'istartswith', 'iendswith'):
             return "UPPER(%s)"
+        if internal_type == 'JSONField' and lookup_type == 'exact':
+            return 'DBMS_LOB.SUBSTR(%s)'
         return "%s"
 
     def max_in_list_size(self):

django/db/backends/postgresql/base.py

@@ -86,6 +86,7 @@ class DatabaseWrapper(BaseDatabaseWrapper):
         'BigIntegerField': 'bigint',
         'IPAddressField': 'inet',
         'GenericIPAddressField': 'inet',
+        'JSONField': 'jsonb',
         'NullBooleanField': 'boolean',
         'OneToOneField': 'integer',
         'PositiveBigIntegerField': 'bigint',

django/db/backends/postgresql/features.py

@@ -12,6 +12,7 @@ class DatabaseFeatures(BaseDatabaseFeatures):
     has_real_datatype = True
     has_native_uuid_field = True
     has_native_duration_field = True
+    has_native_json_field = True
     can_defer_constraint_checks = True
     has_select_for_update = True
     has_select_for_update_nowait = True

django/db/backends/postgresql/introspection.py

@@ -26,6 +26,7 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
         1266: 'TimeField',
         1700: 'DecimalField',
         2950: 'UUIDField',
+        3802: 'JSONField',
     }
 
     ignored_tables = []

django/db/backends/postgresql/operations.py

@@ -74,6 +74,9 @@ class DatabaseOperations(BaseDatabaseOperations):
     def time_trunc_sql(self, lookup_type, field_name):
         return "DATE_TRUNC('%s', %s)::time" % (lookup_type, field_name)
 
+    def json_cast_text_sql(self, field_name):
+        return '(%s)::text' % field_name
+
     def deferrable_sql(self):
         return " DEFERRABLE INITIALLY DEFERRED"
 

django/db/backends/sqlite3/base.py

@@ -5,6 +5,7 @@ import datetime
 import decimal
 import functools
 import hashlib
+import json
 import math
 import operator
 import re
@@ -101,6 +102,7 @@ class DatabaseWrapper(BaseDatabaseWrapper):
         'BigIntegerField': 'bigint',
         'IPAddressField': 'char(15)',
         'GenericIPAddressField': 'char(39)',
+        'JSONField': 'text',
         'NullBooleanField': 'bool',
         'OneToOneField': 'integer',
         'PositiveBigIntegerField': 'bigint unsigned',
@@ -115,6 +117,7 @@ class DatabaseWrapper(BaseDatabaseWrapper):
     }
     data_type_check_constraints = {
         'PositiveBigIntegerField': '"%(column)s" >= 0',
+        'JSONField': '(JSON_VALID("%(column)s") OR "%(column)s" IS NULL)',
         'PositiveIntegerField': '"%(column)s" >= 0',
         'PositiveSmallIntegerField': '"%(column)s" >= 0',
     }
@@ -233,6 +236,7 @@ class DatabaseWrapper(BaseDatabaseWrapper):
         create_deterministic_function('DEGREES', 1, none_guard(math.degrees))
         create_deterministic_function('EXP', 1, none_guard(math.exp))
         create_deterministic_function('FLOOR', 1, none_guard(math.floor))
+        create_deterministic_function('JSON_CONTAINS', 2, _sqlite_json_contains)
         create_deterministic_function('LN', 1, none_guard(math.log))
         create_deterministic_function('LOG', 2, none_guard(lambda x, y: math.log(y, x)))
         create_deterministic_function('LPAD', 3, _sqlite_lpad)
@@ -598,3 +602,11 @@ def _sqlite_lpad(text, length, fill_text):
 @none_guard
 def _sqlite_rpad(text, length, fill_text):
     return (text + fill_text * length)[:length]
+
+
+@none_guard
+def _sqlite_json_contains(haystack, needle):
+    target, candidate = json.loads(haystack), json.loads(needle)
+    if isinstance(target, dict) and isinstance(candidate, dict):
+        return target.items() >= candidate.items()
+    return target == candidate

django/db/backends/sqlite3/features.py

@@ -1,4 +1,9 @@
+import operator
+
+from django.db import transaction
 from django.db.backends.base.features import BaseDatabaseFeatures
+from django.db.utils import OperationalError
+from django.utils.functional import cached_property
 
 from .base import Database
 
@@ -45,3 +50,14 @@ class DatabaseFeatures(BaseDatabaseFeatures):
     supports_aggregate_filter_clause = Database.sqlite_version_info >= (3, 30, 1)
     supports_order_by_nulls_modifier = Database.sqlite_version_info >= (3, 30, 0)
     order_by_nulls_first = True
+
+    @cached_property
+    def supports_json_field(self):
+        try:
+            with self.connection.cursor() as cursor, transaction.atomic():
+                cursor.execute('SELECT JSON(\'{"a": "b"}\')')
+        except OperationalError:
+            return False
+        return True
+
+    can_introspect_json_field = property(operator.attrgetter('supports_json_field'))

django/db/backends/sqlite3/introspection.py

@@ -9,7 +9,7 @@ from django.db.backends.base.introspection import (
 from django.db.models import Index
 from django.utils.regex_helper import _lazy_re_compile
 
-FieldInfo = namedtuple('FieldInfo', BaseFieldInfo._fields + ('pk',))
+FieldInfo = namedtuple('FieldInfo', BaseFieldInfo._fields + ('pk', 'has_json_constraint'))
 
 field_size_re = _lazy_re_compile(r'^\s*(?:var)?char\s*\(\s*(\d+)\s*\)\s*$')
 
@@ -63,6 +63,8 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
             # No support for BigAutoField or SmallAutoField as SQLite treats
             # all integer primary keys as signed 64-bit integers.
             return 'AutoField'
+        if description.has_json_constraint:
+            return 'JSONField'
         return field_type
 
     def get_table_list(self, cursor):
@@ -81,12 +83,28 @@ class DatabaseIntrospection(BaseDatabaseIntrospection):
         interface.
         """
         cursor.execute('PRAGMA table_info(%s)' % self.connection.ops.quote_name(table_name))
+        table_info = cursor.fetchall()
+        json_columns = set()
+        if self.connection.features.can_introspect_json_field:
+            for line in table_info:
+                column = line[1]
+                json_constraint_sql = '%%json_valid("%s")%%' % column
+                has_json_constraint = cursor.execute("""
+                    SELECT sql
+                    FROM sqlite_master
+                    WHERE
+                        type = 'table' AND
+                        name = %s AND
+                        sql LIKE %s
+                """, [table_name, json_constraint_sql]).fetchone()
+                if has_json_constraint:
+                    json_columns.add(column)
         return [
             FieldInfo(
                 name, data_type, None, get_field_size(data_type), None, None,
-                not notnull, default, pk == 1,
+                not notnull, default, pk == 1, name in json_columns
             )
-            for cid, name, data_type, notnull, default, pk in cursor.fetchall()
+            for cid, name, data_type, notnull, default, pk in table_info
         ]
 
     def get_sequences(self, cursor, table_name, table_fields=()):

django/db/models/__init__.py

@@ -18,6 +18,7 @@ from django.db.models.expressions import (
 from django.db.models.fields import *  # NOQA
 from django.db.models.fields import __all__ as fields_all
 from django.db.models.fields.files import FileField, ImageField
+from django.db.models.fields.json import JSONField
 from django.db.models.fields.proxy import OrderWrt
 from django.db.models.indexes import *  # NOQA
 from django.db.models.indexes import __all__ as indexes_all
@@ -43,9 +44,9 @@ __all__ += [
     'Func', 'OrderBy', 'OuterRef', 'RowRange', 'Subquery', 'Value',
     'ValueRange', 'When',
     'Window', 'WindowFrame',
-    'FileField', 'ImageField', 'OrderWrt', 'Lookup', 'Transform', 'Manager',
-    'Prefetch', 'Q', 'QuerySet', 'prefetch_related_objects', 'DEFERRED', 'Model',
-    'FilteredRelation',
+    'FileField', 'ImageField', 'JSONField', 'OrderWrt', 'Lookup', 'Transform',
+    'Manager', 'Prefetch', 'Q', 'QuerySet', 'prefetch_related_objects',
+    'DEFERRED', 'Model', 'FilteredRelation',
     'ForeignKey', 'ForeignObject', 'OneToOneField', 'ManyToManyField',
     'ForeignObjectRel', 'ManyToOneRel', 'ManyToManyRel', 'OneToOneRel',
 ]

django/db/models/fields/__init__.py

@@ -496,6 +496,8 @@ class Field(RegisterLookupMixin):
             path = path.replace("django.db.models.fields.related", "django.db.models")
         elif path.startswith("django.db.models.fields.files"):
             path = path.replace("django.db.models.fields.files", "django.db.models")
+        elif path.startswith('django.db.models.fields.json'):
+            path = path.replace('django.db.models.fields.json', 'django.db.models')
         elif path.startswith("django.db.models.fields.proxy"):
             path = path.replace("django.db.models.fields.proxy", "django.db.models")
         elif path.startswith("django.db.models.fields"):

django/db/models/fields/json.py

@@ -0,0 +1,525 @@
+import json
+
+from django import forms
+from django.core import checks, exceptions
+from django.db import NotSupportedError, connections, router
+from django.db.models import lookups
+from django.db.models.lookups import PostgresOperatorLookup, Transform
+from django.utils.translation import gettext_lazy as _
+
+from . import Field
+from .mixins import CheckFieldDefaultMixin
+
+__all__ = ['JSONField']
+
+
+class JSONField(CheckFieldDefaultMixin, Field):
+    empty_strings_allowed = False
+    description = _('A JSON object')
+    default_error_messages = {
+        'invalid': _('Value must be valid JSON.'),
+    }
+    _default_hint = ('dict', '{}')
+
+    def __init__(
+        self, verbose_name=None, name=None, encoder=None, decoder=None,
+        **kwargs,
+    ):
+        if encoder and not callable(encoder):
+            raise ValueError('The encoder parameter must be a callable object.')
+        if decoder and not callable(decoder):
+            raise ValueError('The decoder parameter must be a callable object.')
+        self.encoder = encoder
+        self.decoder = decoder
+        super().__init__(verbose_name, name, **kwargs)
+
+    def check(self, **kwargs):
+        errors = super().check(**kwargs)
+        databases = kwargs.get('databases') or []
+        errors.extend(self._check_supported(databases))
+        return errors
+
+    def _check_supported(self, databases):
+        errors = []
+        for db in databases:
+            if not router.allow_migrate_model(db, self.model):
+                continue
+            connection = connections[db]
+            if not (
+                'supports_json_field' in self.model._meta.required_db_features or
+                connection.features.supports_json_field
+            ):
+                errors.append(
+                    checks.Error(
+                        '%s does not support JSONFields.'
+                        % connection.display_name,
+                        obj=self.model,
+                        id='fields.E180',
+                    )
+                )
+        return errors
+
+    def deconstruct(self):
+        name, path, args, kwargs = super().deconstruct()
+        if self.encoder is not None:
+            kwargs['encoder'] = self.encoder
+        if self.decoder is not None:
+            kwargs['decoder'] = self.decoder
+        return name, path, args, kwargs
+
+    def from_db_value(self, value, expression, connection):
+        if value is None:
+            return value
+        if connection.features.has_native_json_field and self.decoder is None:
+            return value
+        try:
+            return json.loads(value, cls=self.decoder)
+        except json.JSONDecodeError:
+            return value
+
+    def get_internal_type(self):
+        return 'JSONField'
+
+    def get_prep_value(self, value):
+        if value is None:
+            return value
+        return json.dumps(value, cls=self.encoder)
+
+    def get_transform(self, name):
+        transform = super().get_transform(name)
+        if transform:
+            return transform
+        return KeyTransformFactory(name)
+
+    def select_format(self, compiler, sql, params):
+        if (
+            compiler.connection.features.has_native_json_field and
+            self.decoder is not None
+        ):
+            return compiler.connection.ops.json_cast_text_sql(sql), params
+        return super().select_format(compiler, sql, params)
+
+    def validate(self, value, model_instance):
+        super().validate(value, model_instance)
+        try:
+            json.dumps(value, cls=self.encoder)
+        except TypeError:
+            raise exceptions.ValidationError(
+                self.error_messages['invalid'],
+                code='invalid',
+                params={'value': value},
+            )
+
+    def value_to_string(self, obj):
+        return self.value_from_object(obj)
+
+    def formfield(self, **kwargs):
+        return super().formfield(**{
+            'form_class': forms.JSONField,
+            'encoder': self.encoder,
+            'decoder': self.decoder,
+            **kwargs,
+        })
+
+
+def compile_json_path(key_transforms, include_root=True):
+    path = ['$'] if include_root else []
+    for key_transform in key_transforms:
+        try:
+            num = int(key_transform)
+        except ValueError:  # non-integer
+            path.append('.')
+            path.append(json.dumps(key_transform))
+        else:
+            path.append('[%s]' % num)
+    return ''.join(path)
+
+
+class DataContains(PostgresOperatorLookup):
+    lookup_name = 'contains'
+    postgres_operator = '@>'
+
+    def as_sql(self, compiler, connection):
+        lhs, lhs_params = self.process_lhs(compiler, connection)
+        rhs, rhs_params = self.process_rhs(compiler, connection)
+        params = tuple(lhs_params) + tuple(rhs_params)
+        return 'JSON_CONTAINS(%s, %s)' % (lhs, rhs), params
+
+    def as_oracle(self, compiler, connection):
+        if isinstance(self.rhs, KeyTransform):
+            return HasKey(self.lhs, self.rhs).as_oracle(compiler, connection)
+        lhs, lhs_params = self.process_lhs(compiler, connection)
+        params = tuple(lhs_params)
+        sql = (
+            "JSON_QUERY(%s, '$%s' WITH WRAPPER) = "
+            "JSON_QUERY('%s', '$.value' WITH WRAPPER)"
+        )
+        rhs = json.loads(self.rhs)
+        if isinstance(rhs, dict):
+            if not rhs:
+                return "DBMS_LOB.SUBSTR(%s) LIKE '{%%%%}'" % lhs, params
+            return ' AND '.join([
+                sql % (
+                    lhs, '.%s' % json.dumps(key), json.dumps({'value': value}),
+                ) for key, value in rhs.items()
+            ]), params
+        return sql % (lhs, '', json.dumps({'value': rhs})), params
+
+
+class ContainedBy(PostgresOperatorLookup):
+    lookup_name = 'contained_by'
+    postgres_operator = '<@'
+
+    def as_sql(self, compiler, connection):
+        lhs, lhs_params = self.process_lhs(compiler, connection)
+        rhs, rhs_params = self.process_rhs(compiler, connection)
+        params = tuple(rhs_params) + tuple(lhs_params)
+        return 'JSON_CONTAINS(%s, %s)' % (rhs, lhs), params
+
+    def as_oracle(self, compiler, connection):
+        raise NotSupportedError('contained_by lookup is not supported on Oracle.')
+
+
+class HasKeyLookup(PostgresOperatorLookup):
+    logical_operator = None
+
+    def as_sql(self, compiler, connection, template=None):
+        # Process JSON path from the left-hand side.
+        if isinstance(self.lhs, KeyTransform):
+            lhs, lhs_params, lhs_key_transforms = self.lhs.preprocess_lhs(compiler, connection)
+            lhs_json_path = compile_json_path(lhs_key_transforms)
+        else:
+            lhs, lhs_params = self.process_lhs(compiler, connection)
+            lhs_json_path = '$'
+        sql = template % lhs
+        # Process JSON path from the right-hand side.
+        rhs = self.rhs
+        rhs_params = []
+        if not isinstance(rhs, (list, tuple)):
+            rhs = [rhs]
+        for key in rhs:
+            if isinstance(key, KeyTransform):
+                *_, rhs_key_transforms = key.preprocess_lhs(compiler, connection)
+            else:
+                rhs_key_transforms = [key]
+            rhs_params.append('%s%s' % (
+                lhs_json_path,
+                compile_json_path(rhs_key_transforms, include_root=False),
+            ))
+        # Add condition for each key.
+        if self.logical_operator:
+            sql = '(%s)' % self.logical_operator.join([sql] * len(rhs_params))
+        return sql, tuple(lhs_params) + tuple(rhs_params)
+
+    def as_mysql(self, compiler, connection):
+        return self.as_sql(compiler, connection, template="JSON_CONTAINS_PATH(%s, 'one', %%s)")
+
+    def as_oracle(self, compiler, connection):
+        sql, params = self.as_sql(compiler, connection, template="JSON_EXISTS(%s, '%%s')")
+        # Add paths directly into SQL because path expressions cannot be passed
+        # as bind variables on Oracle.
+        return sql % tuple(params), []
+
+    def as_postgresql(self, compiler, connection):
+        if isinstance(self.rhs, KeyTransform):
+            *_, rhs_key_transforms = self.rhs.preprocess_lhs(compiler, connection)
+            for key in rhs_key_transforms[:-1]:
+                self.lhs = KeyTransform(key, self.lhs)
+            self.rhs = rhs_key_transforms[-1]
+        return super().as_postgresql(compiler, connection)
+
+    def as_sqlite(self, compiler, connection):
+        return self.as_sql(compiler, connection, template='JSON_TYPE(%s, %%s) IS NOT NULL')
+
+
+class HasKey(HasKeyLookup):
+    lookup_name = 'has_key'
+    postgres_operator = '?'
+    prepare_rhs = False
+
+
+class HasKeys(HasKeyLookup):
+    lookup_name = 'has_keys'
+    postgres_operator = '?&'
+    logical_operator = ' AND '
+
+    def get_prep_lookup(self):
+        return [str(item) for item in self.rhs]
+
+
+class HasAnyKeys(HasKeys):
+    lookup_name = 'has_any_keys'
+    postgres_operator = '?|'
+    logical_operator = ' OR '
+
+
+class JSONExact(lookups.Exact):
+    can_use_none_as_rhs = True
+
+    def process_lhs(self, compiler, connection):
+        lhs, lhs_params = super().process_lhs(compiler, connection)
+        if connection.vendor == 'sqlite':
+            rhs, rhs_params = super().process_rhs(compiler, connection)
+            if rhs == '%s' and rhs_params == [None]:
+                # Use JSON_TYPE instead of JSON_EXTRACT for NULLs.
+                lhs = "JSON_TYPE(%s, '$')" % lhs
+        return lhs, lhs_params
+
+    def process_rhs(self, compiler, connection):
+        rhs, rhs_params = super().process_rhs(compiler, connection)
+        # Treat None lookup values as null.
+        if rhs == '%s' and rhs_params == [None]:
+            rhs_params = ['null']
+        if connection.vendor == 'mysql':
+            func = ["JSON_EXTRACT(%s, '$')"] * len(rhs_params)
+            rhs = rhs % tuple(func)
+        return rhs, rhs_params
+
+
+JSONField.register_lookup(DataContains)
+JSONField.register_lookup(ContainedBy)
+JSONField.register_lookup(HasKey)
+JSONField.register_lookup(HasKeys)
+JSONField.register_lookup(HasAnyKeys)
+JSONField.register_lookup(JSONExact)
+
+
+class KeyTransform(Transform):
+    postgres_operator = '->'
+    postgres_nested_operator = '#>'
+
+    def __init__(self, key_name, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.key_name = str(key_name)
+
+    def preprocess_lhs(self, compiler, connection, lhs_only=False):
+        if not lhs_only:
+            key_transforms = [self.key_name]
+        previous = self.lhs
+        while isinstance(previous, KeyTransform):
+            if not lhs_only:
+                key_transforms.insert(0, previous.key_name)
+            previous = previous.lhs
+        lhs, params = compiler.compile(previous)
+        if connection.vendor == 'oracle':
+            # Escape string-formatting.
+            key_transforms = [key.replace('%', '%%') for key in key_transforms]
+        return (lhs, params, key_transforms) if not lhs_only else (lhs, params)
+
+    def as_mysql(self, compiler, connection):
+        lhs, params, key_transforms = self.preprocess_lhs(compiler, connection)
+        json_path = compile_json_path(key_transforms)
+        return 'JSON_EXTRACT(%s, %%s)' % lhs, tuple(params) + (json_path,)
+
+    def as_oracle(self, compiler, connection):
+        lhs, params, key_transforms = self.preprocess_lhs(compiler, connection)
+        json_path = compile_json_path(key_transforms)
+        return (
+            "COALESCE(JSON_QUERY(%s, '%s'), JSON_VALUE(%s, '%s'))" %
+            ((lhs, json_path) * 2)
+        ), tuple(params) * 2
+
+    def as_postgresql(self, compiler, connection):
+        lhs, params, key_transforms = self.preprocess_lhs(compiler, connection)
+        if len(key_transforms) > 1:
+            return '(%s %s %%s)' % (lhs, self.postgres_nested_operator), params + [key_transforms]
+        try:
+            lookup = int(self.key_name)
+        except ValueError:
+            lookup = self.key_name
+        return '(%s %s %%s)' % (lhs, self.postgres_operator), tuple(params) + (lookup,)
+
+    def as_sqlite(self, compiler, connection):
+        lhs, params, key_transforms = self.preprocess_lhs(compiler, connection)
+        json_path = compile_json_path(key_transforms)
+        return 'JSON_EXTRACT(%s, %%s)' % lhs, tuple(params) + (json_path,)
+
+
+class KeyTextTransform(KeyTransform):
+    postgres_operator = '->>'
+    postgres_nested_operator = '#>>'
+
+
+class KeyTransformTextLookupMixin:
+    """
+    Mixin for combining with a lookup expecting a text lhs from a JSONField
+    key lookup. On PostgreSQL, make use of the ->> operator instead of casting
+    key values to text and performing the lookup on the resulting
+    representation.
+    """
+    def __init__(self, key_transform, *args, **kwargs):
+        if not isinstance(key_transform, KeyTransform):
+            raise TypeError(
+                'Transform should be an instance of KeyTransform in order to '
+                'use this lookup.'
+            )
+        key_text_transform = KeyTextTransform(
+            key_transform.key_name, *key_transform.source_expressions,
+            **key_transform.extra,
+        )
+        super().__init__(key_text_transform, *args, **kwargs)
+
+
+class CaseInsensitiveMixin:
+    """
+    Mixin to allow case-insensitive comparison of JSON values on MySQL.
+    MySQL handles strings used in JSON context using the utf8mb4_bin collation.
+    Because utf8mb4_bin is a binary collation, comparison of JSON values is
+    case-sensitive.
+    """
+    def process_lhs(self, compiler, connection):
+        lhs, lhs_params = super().process_lhs(compiler, connection)
+        if connection.vendor == 'mysql':
+            return 'LOWER(%s)' % lhs, lhs_params
+        return lhs, lhs_params
+
+    def process_rhs(self, compiler, connection):
+        rhs, rhs_params = super().process_rhs(compiler, connection)
+        if connection.vendor == 'mysql':
+            return 'LOWER(%s)' % rhs, rhs_params
+        return rhs, rhs_params
+
+
+class KeyTransformIsNull(lookups.IsNull):
+    # key__isnull=False is the same as has_key='key'
+    def as_oracle(self, compiler, connection):
+        if not self.rhs:
+            return HasKey(self.lhs.lhs, self.lhs.key_name).as_oracle(compiler, connection)
+        return super().as_sql(compiler, connection)
+
+    def as_sqlite(self, compiler, connection):
+        if not self.rhs:
+            return HasKey(self.lhs.lhs, self.lhs.key_name).as_sqlite(compiler, connection)
+        return super().as_sql(compiler, connection)
+
+
+class KeyTransformExact(JSONExact):
+    def process_lhs(self, compiler, connection):
+        lhs, lhs_params = super().process_lhs(compiler, connection)
+        if connection.vendor == 'sqlite':
+            rhs, rhs_params = super().process_rhs(compiler, connection)
+            if rhs == '%s' and rhs_params == ['null']:
+                lhs, _ = self.lhs.preprocess_lhs(compiler, connection, lhs_only=True)
+                lhs = 'JSON_TYPE(%s, %%s)' % lhs
+        return lhs, lhs_params
+
+    def process_rhs(self, compiler, connection):
+        if isinstance(self.rhs, KeyTransform):
+            return super(lookups.Exact, self).process_rhs(compiler, connection)
+        rhs, rhs_params = super().process_rhs(compiler, connection)
+        if connection.vendor == 'oracle':
+            func = []
+            for value in rhs_params:
+                value = json.loads(value)
+                function = 'JSON_QUERY' if isinstance(value, (list, dict)) else 'JSON_VALUE'
+                func.append("%s('%s', '$.value')" % (
+                    function,
+                    json.dumps({'value': value}),
+                ))
+            rhs = rhs % tuple(func)
+            rhs_params = []
+        elif connection.vendor == 'sqlite':
+            func = ["JSON_EXTRACT(%s, '$')" if value != 'null' else '%s' for value in rhs_params]
+            rhs = rhs % tuple(func)
+        return rhs, rhs_params
+
+    def as_oracle(self, compiler, connection):
+        rhs, rhs_params = super().process_rhs(compiler, connection)
+        if rhs_params == ['null']:
+            # Field has key and it's NULL.
+            has_key_expr = HasKey(self.lhs.lhs, self.lhs.key_name)
+            has_key_sql, has_key_params = has_key_expr.as_oracle(compiler, connection)
+            is_null_expr = self.lhs.get_lookup('isnull')(self.lhs, True)
+            is_null_sql, is_null_params = is_null_expr.as_sql(compiler, connection)
+            return (
+                '%s AND %s' % (has_key_sql, is_null_sql),
+                tuple(has_key_params) + tuple(is_null_params),
+            )
+        return super().as_sql(compiler, connection)
+
+
+class KeyTransformIExact(CaseInsensitiveMixin, KeyTransformTextLookupMixin, lookups.IExact):
+    pass
+
+
+class KeyTransformIContains(CaseInsensitiveMixin, KeyTransformTextLookupMixin, lookups.IContains):
+    pass
+
+
+class KeyTransformContains(KeyTransformTextLookupMixin, lookups.Contains):
+    pass
+
+
+class KeyTransformStartsWith(KeyTransformTextLookupMixin, lookups.StartsWith):
+    pass
+
+
+class KeyTransformIStartsWith(CaseInsensitiveMixin, KeyTransformTextLookupMixin, lookups.IStartsWith):
+    pass
+
+
+class KeyTransformEndsWith(KeyTransformTextLookupMixin, lookups.EndsWith):
+    pass
+
+
+class KeyTransformIEndsWith(CaseInsensitiveMixin, KeyTransformTextLookupMixin, lookups.IEndsWith):
+    pass
+
+
+class KeyTransformRegex(KeyTransformTextLookupMixin, lookups.Regex):
+    pass
+
+
+class KeyTransformIRegex(CaseInsensitiveMixin, KeyTransformTextLookupMixin, lookups.IRegex):
+    pass
+
+
+class KeyTransformNumericLookupMixin:
+    def process_rhs(self, compiler, connection):
+        rhs, rhs_params = super().process_rhs(compiler, connection)
+        if not connection.features.has_native_json_field:
+            rhs_params = [json.loads(value) for value in rhs_params]
+        return rhs, rhs_params
+
+
+class KeyTransformLt(KeyTransformNumericLookupMixin, lookups.LessThan):
+    pass
+
+
+class KeyTransformLte(KeyTransformNumericLookupMixin, lookups.LessThanOrEqual):
+    pass
+
+
+class KeyTransformGt(KeyTransformNumericLookupMixin, lookups.GreaterThan):
+    pass
+
+
+class KeyTransformGte(KeyTransformNumericLookupMixin, lookups.GreaterThanOrEqual):
+    pass
+
+
+KeyTransform.register_lookup(KeyTransformExact)
+KeyTransform.register_lookup(KeyTransformIExact)
+KeyTransform.register_lookup(KeyTransformIsNull)
+KeyTransform.register_lookup(KeyTransformContains)
+KeyTransform.register_lookup(KeyTransformIContains)
+KeyTransform.register_lookup(KeyTransformStartsWith)
+KeyTransform.register_lookup(KeyTransformIStartsWith)
+KeyTransform.register_lookup(KeyTransformEndsWith)
+KeyTransform.register_lookup(KeyTransformIEndsWith)
+KeyTransform.register_lookup(KeyTransformRegex)
+KeyTransform.register_lookup(KeyTransformIRegex)
+
+KeyTransform.register_lookup(KeyTransformLt)
+KeyTransform.register_lookup(KeyTransformLte)
+KeyTransform.register_lookup(KeyTransformGt)
+KeyTransform.register_lookup(KeyTransformGte)
+
+
+class KeyTransformFactory:
+
+    def __init__(self, key_name):
+        self.key_name = key_name
+
+    def __call__(self, *args, **kwargs):
+        return KeyTransform(self.key_name, *args, **kwargs)

django/db/models/functions/comparison.py

@@ -29,8 +29,14 @@ class Cast(Func):
         return self.as_sql(compiler, connection, **extra_context)
 
     def as_mysql(self, compiler, connection, **extra_context):
+        template = None
+        output_type = self.output_field.get_internal_type()
         # MySQL doesn't support explicit cast to float.
-        template = '(%(expressions)s + 0.0)' if self.output_field.get_internal_type() == 'FloatField' else None
+        if output_type == 'FloatField':
+            template = '(%(expressions)s + 0.0)'
+        # MariaDB doesn't support explicit cast to JSON.
+        elif output_type == 'JSONField' and connection.mysql_is_mariadb:
+            template = "JSON_EXTRACT(%(expressions)s, '$')"
         return self.as_sql(compiler, connection, template=template, **extra_context)
 
     def as_postgresql(self, compiler, connection, **extra_context):
@@ -39,6 +45,13 @@ class Cast(Func):
         # expression.
         return self.as_sql(compiler, connection, template='(%(expressions)s)::%(db_type)s', **extra_context)
 
+    def as_oracle(self, compiler, connection, **extra_context):
+        if self.output_field.get_internal_type() == 'JSONField':
+            # Oracle doesn't support explicit cast to JSON.
+            template = "JSON_QUERY(%(expressions)s, '$')"
+            return super().as_sql(compiler, connection, template=template, **extra_context)
+        return self.as_sql(compiler, connection, **extra_context)
+
 
 class Coalesce(Func):
     """Return, from left to right, the first non-null expression."""