base.py

from __future__ import unicode_literals

import copy
import inspect
import warnings
from itertools import chain

from django.apps import apps
from django.conf import settings
from django.core import checks
from django.core.exceptions import (
    NON_FIELD_ERRORS, FieldDoesNotExist, FieldError, MultipleObjectsReturned,
    ObjectDoesNotExist, ValidationError,
)
from django.db import (
    DEFAULT_DB_ALIAS, DJANGO_VERSION_PICKLE_KEY, DatabaseError, connections,
    router, transaction,
)
from django.db.models import signals
from django.db.models.constants import LOOKUP_SEP
from django.db.models.deletion import CASCADE, Collector
from django.db.models.fields import AutoField
from django.db.models.fields.related import (
    ForeignObjectRel, ManyToOneRel, OneToOneField, lazy_related_operation,
    resolve_relation,
)
from django.db.models.manager import Manager
from django.db.models.options import Options
from django.db.models.query import Q
from django.db.models.utils import make_model_tuple
from django.utils import six
from django.utils.deprecation import RemovedInDjango20Warning
from django.utils.encoding import (
    force_str, force_text, python_2_unicode_compatible,
)
from django.utils.functional import curry
from django.utils.six.moves import zip
from django.utils.text import capfirst, get_text_list
from django.utils.translation import ugettext_lazy as _
from django.utils.version import get_version


@python_2_unicode_compatible
class Deferred(object):
    def __repr__(self):
        return str('<Deferred field>')

    def __str__(self):
        return str('<Deferred field>')

DEFERRED = Deferred()


def subclass_exception(name, parents, module, attached_to=None):
    """
    Create exception subclass. Used by ModelBase below.

    If 'attached_to' is supplied, the exception will be created in a way that
    allows it to be pickled, assuming the returned exception class will be added
    as an attribute to the 'attached_to' class.
    """
    class_dict = {'__module__': module}
    if attached_to is not None:
        def __reduce__(self):
            # Exceptions are special - they've got state that isn't
            # in self.__dict__. We assume it is all in self.args.
            return (unpickle_inner_exception, (attached_to, name), self.args)

        def __setstate__(self, args):
            self.args = args

        class_dict['__reduce__'] = __reduce__
        class_dict['__setstate__'] = __setstate__

    return type(name, parents, class_dict)


class ModelBase(type):
    """
    Metaclass for all models.
    """
    def __new__(cls, name, bases, attrs):
        super_new = super(ModelBase, cls).__new__

        # Also ensure initialization is only performed for subclasses of Model
        # (excluding Model class itself).
        parents = [b for b in bases if isinstance(b, ModelBase)]
        if not parents:
            return super_new(cls, name, bases, attrs)

        # Create the class.
        module = attrs.pop('__module__')
        new_class = super_new(cls, name, bases, {'__module__': module})
        attr_meta = attrs.pop('Meta', None)
        abstract = getattr(attr_meta, 'abstract', False)
        if not attr_meta:
            meta = getattr(new_class, 'Meta', None)
        else:
            meta = attr_meta
        base_meta = getattr(new_class, '_meta', None)

        app_label = None

        # Look for an application configuration to attach the model to.
        app_config = apps.get_containing_app_config(module)

        if getattr(meta, 'app_label', None) is None:
            if app_config is None:
                if not abstract:
                    raise RuntimeError(
                        "Model class %s.%s doesn't declare an explicit "
                        "app_label and isn't in an application in "
                        "INSTALLED_APPS." % (module, name)
                    )

            else:
                app_label = app_config.label

        new_class.add_to_class('_meta', Options(meta, app_label))
        if not abstract:
            new_class.add_to_class(
                'DoesNotExist',
                subclass_exception(
                    str('DoesNotExist'),
                    tuple(
                        x.DoesNotExist for x in parents if hasattr(x, '_meta') and not x._meta.abstract
                    ) or (ObjectDoesNotExist,),
                    module,
                    attached_to=new_class))
            new_class.add_to_class(
                'MultipleObjectsReturned',
                subclass_exception(
                    str('MultipleObjectsReturned'),
                    tuple(
                        x.MultipleObjectsReturned for x in parents if hasattr(x, '_meta') and not x._meta.abstract
                    ) or (MultipleObjectsReturned,),
                    module,
                    attached_to=new_class))
            if base_meta and not base_meta.abstract:
                # Non-abstract child classes inherit some attributes from their
                # non-abstract parent (unless an ABC comes before it in the
                # method resolution order).
                if not hasattr(meta, 'ordering'):
                    new_class._meta.ordering = base_meta.ordering
                if not hasattr(meta, 'get_latest_by'):
                    new_class._meta.get_latest_by = base_meta.get_latest_by

        is_proxy = new_class._meta.proxy

        # If the model is a proxy, ensure that the base class
        # hasn't been swapped out.
        if is_proxy and base_meta and base_meta.swapped:
            raise TypeError("%s cannot proxy the swapped model '%s'." % (name, base_meta.swapped))

        # Add all attributes to the class.
        for obj_name, obj in attrs.items():
            new_class.add_to_class(obj_name, obj)

        # All the fields of any type declared on this model
        new_fields = chain(
            new_class._meta.local_fields,
            new_class._meta.local_many_to_many,
            new_class._meta.private_fields
        )
        field_names = {f.name for f in new_fields}

        # Basic setup for proxy models.
        if is_proxy:
            base = None
            for parent in [kls for kls in parents if hasattr(kls, '_meta')]:
                if parent._meta.abstract:
                    if parent._meta.fields:
                        raise TypeError(
                            "Abstract base class containing model fields not "
                            "permitted for proxy model '%s'." % name
                        )
                    else:
                        continue
                if base is None:
                    base = parent
                elif parent._meta.concrete_model is not base._meta.concrete_model:
                    raise TypeError("Proxy model '%s' has more than one non-abstract model base class." % name)
            if base is None:
                raise TypeError("Proxy model '%s' has no non-abstract model base class." % name)
            new_class._meta.setup_proxy(base)
            new_class._meta.concrete_model = base._meta.concrete_model
        else:
            new_class._meta.concrete_model = new_class

        # Collect the parent links for multi-table inheritance.
        parent_links = {}
        for base in reversed([new_class] + parents):
            # Conceptually equivalent to `if base is Model`.
            if not hasattr(base, '_meta'):
                continue
            # Skip concrete parent classes.
            if base != new_class and not base._meta.abstract:
                continue
            # Locate OneToOneField instances.
            for field in base._meta.local_fields:
                if isinstance(field, OneToOneField):
                    related = resolve_relation(new_class, field.remote_field.model)
                    parent_links[make_model_tuple(related)] = field

        # Track fields inherited from base models.
        inherited_attributes = set()
        # Do the appropriate setup for any model parents.
        for base in new_class.mro():
            if base not in parents or not hasattr(base, '_meta'):
                # Things without _meta aren't functional models, so they're
                # uninteresting parents.
                inherited_attributes |= set(base.__dict__.keys())
                continue

            parent_fields = base._meta.local_fields + base._meta.local_many_to_many
            if not base._meta.abstract:
                # Check for clashes between locally declared fields and those
                # on the base classes.
                for field in parent_fields:
                    if field.name in field_names:
                        raise FieldError(
                            'Local field %r in class %r clashes with field of '
                            'the same name from base class %r.' % (
                                field.name,
                                name,
                                base.__name__,
                            )
                        )
                    else:
                        inherited_attributes.add(field.name)

                # Concrete classes...
                base = base._meta.concrete_model
                base_key = make_model_tuple(base)
                if base_key in parent_links:
                    field = parent_links[base_key]
                elif not is_proxy:
                    attr_name = '%s_ptr' % base._meta.model_name
                    field = OneToOneField(
                        base,
                        on_delete=CASCADE,
                        name=attr_name,
                        auto_created=True,
                        parent_link=True,
                    )

                    if attr_name in field_names:
                        raise FieldError(
                            "Auto-generated field '%s' in class %r for "
                            "parent_link to base class %r clashes with "
                            "declared field of the same name." % (
                                attr_name,
                                name,
                                base.__name__,
                            )
                        )

                    # Only add the ptr field if it's not already present;
                    # e.g. migrations will already have it specified
                    if not hasattr(new_class, attr_name):
                        new_class.add_to_class(attr_name, field)
                else:
                    field = None
                new_class._meta.parents[base] = field
            else:
                base_parents = base._meta.parents.copy()

                # Add fields from abstract base class if it wasn't overridden.
                for field in parent_fields:
                    if (field.name not in field_names and
                            field.name not in new_class.__dict__ and
                            field.name not in inherited_attributes):
                        new_field = copy.deepcopy(field)
                        new_class.add_to_class(field.name, new_field)
                        # Replace parent links defined on this base by the new
                        # field. It will be appropriately resolved if required.
                        if field.one_to_one:
                            for parent, parent_link in base_parents.items():
                                if field == parent_link:
                                    base_parents[parent] = new_field

                # Pass any non-abstract parent classes onto child.
                new_class._meta.parents.update(base_parents)

            # Inherit private fields (like GenericForeignKey) from the parent
            # class
            for field in base._meta.private_fields:
                if field.name in field_names:
                    if not base._meta.abstract:
                        raise FieldError(
                            'Local field %r in class %r clashes with field of '
                            'the same name from base class %r.' % (
                                field.name,
                                name,
                                base.__name__,
                            )
                        )
                else:
                    new_class.add_to_class(field.name, copy.deepcopy(field))

        if abstract:
            # Abstract base models can't be instantiated and don't appear in
            # the list of models for an app. We do the final setup for them a
            # little differently from normal models.
            attr_meta.abstract = False
            new_class.Meta = attr_meta
            return new_class

        new_class._prepare()
        new_class._meta.apps.register_model(new_class._meta.app_label, new_class)
        return new_class

    def add_to_class(cls, name, value):
        # We should call the contribute_to_class method only if it's bound
        if not inspect.isclass(value) and hasattr(value, 'contribute_to_class'):
            value.contribute_to_class(cls, name)
        else:
            setattr(cls, name, value)

    def _prepare(cls):
        """
        Creates some methods once self._meta has been populated.
        """
        opts = cls._meta
        opts._prepare(cls)

        if opts.order_with_respect_to:
            cls.get_next_in_order = curry(cls._get_next_or_previous_in_order, is_next=True)
            cls.get_previous_in_order = curry(cls._get_next_or_previous_in_order, is_next=False)

            # Defer creating accessors on the foreign class until it has been
            # created and registered. If remote_field is None, we're ordering
            # with respect to a GenericForeignKey and don't know what the
            # foreign class is - we'll add those accessors later in
            # contribute_to_class().
            if opts.order_with_respect_to.remote_field:
                wrt = opts.order_with_respect_to
                remote = wrt.remote_field.model
                lazy_related_operation(make_foreign_order_accessors, cls, remote)

        # Give the class a docstring -- its definition.
        if cls.__doc__ is None:
            cls.__doc__ = "%s(%s)" % (cls.__name__, ", ".join(f.name for f in opts.fields))

        get_absolute_url_override = settings.ABSOLUTE_URL_OVERRIDES.get(opts.label_lower)
        if get_absolute_url_override:
            setattr(cls, 'get_absolute_url', get_absolute_url_override)

        if not opts.managers or cls._requires_legacy_default_manager():
            if any(f.name == 'objects' for f in opts.fields):
                raise ValueError(
                    "Model %s must specify a custom Manager, because it has a "
                    "field named 'objects'." % cls.__name__
                )
            manager = Manager()
            manager.auto_created = True
            cls.add_to_class('objects', manager)

        signals.class_prepared.send(sender=cls)

    def _requires_legacy_default_manager(cls):  # RemovedInDjango20Warning
        opts = cls._meta

        if opts.manager_inheritance_from_future:
            return False

        future_default_manager = opts.default_manager

        # Step 1: Locate a manager that would have been promoted
        # to default manager with the legacy system.
        for manager in opts.managers:
            originating_model = manager._originating_model
            if (cls is originating_model or cls._meta.proxy or
                    originating_model._meta.abstract):

                if manager is not cls._default_manager and not opts.default_manager_name:
                    warnings.warn(
                        "Managers from concrete parents will soon qualify as default "
                        "managers if they appear before any other managers in the "
                        "MRO. As a result, '{legacy_default_manager}' declared on "
                        "'{legacy_default_manager_model}' will no longer be the "
                        "default manager for '{model}' in favor of "
                        "'{future_default_manager}' declared on "
                        "'{future_default_manager_model}'. "
                        "You can redeclare '{legacy_default_manager}' on '{cls}' "
                        "to keep things the way they are or you can switch to the new "
                        "behavior right away by setting "
                        "`Meta.manager_inheritance_from_future` to `True`.".format(
                            cls=cls.__name__,
                            model=opts.label,
                            legacy_default_manager=manager.name,
                            legacy_default_manager_model=manager._originating_model._meta.label,
                            future_default_manager=future_default_manager.name,
                            future_default_manager_model=future_default_manager._originating_model._meta.label,
                        ),
                        RemovedInDjango20Warning, 2
                    )

                    opts.default_manager_name = manager.name
                    opts._expire_cache()

                break

        # Step 2: Since there are managers but none of them qualified as
        # default managers under the legacy system (meaning that there are
        # managers from concrete parents that would be promoted under the
        # new system), we need to create a new Manager instance for the
        # 'objects' attribute as a deprecation shim.
        else:
            # If the "future" default manager was auto created there is no
            # point warning the user since it's basically the same manager.
            if not future_default_manager.auto_created:
                warnings.warn(
                    "Managers from concrete parents will soon qualify as "
                    "default managers. As a result, the 'objects' manager "
                    "won't be created (or recreated) automatically "
                    "anymore on '{model}' and '{future_default_manager}' "
                    "declared on '{future_default_manager_model}' will be "
                    "promoted to default manager. You can declare "
                    "explicitly `objects = models.Manager()` on '{cls}' "
                    "to keep things the way they are or you can switch "
                    "to the new behavior right away by setting "
                    "`Meta.manager_inheritance_from_future` to `True`.".format(
                        cls=cls.__name__,
                        model=opts.label,
                        future_default_manager=future_default_manager.name,
                        future_default_manager_model=future_default_manager._originating_model._meta.label,
                    ),
                    RemovedInDjango20Warning, 2
                )

            return True

    @property
    def _base_manager(cls):
        return cls._meta.base_manager

    @property
    def _default_manager(cls):
        return cls._meta.default_manager


class ModelState(object):
    """
    A class for storing instance state
    """
    def __init__(self, db=None):
        self.db = db
        # If true, uniqueness validation checks will consider this a new, as-yet-unsaved object.
        # Necessary for correct validation of new instances of objects with explicit (non-auto) PKs.
        # This impacts validation only; it has no effect on the actual save.
        self.adding = True


class Model(six.with_metaclass(ModelBase)):

    def __init__(self, *args, **kwargs):
        signals.pre_init.send(sender=self.__class__, args=args, kwargs=kwargs)

        # Set up the storage for instance state
        self._state = ModelState()

        # There is a rather weird disparity here; if kwargs, it's set, then args
        # overrides it. It should be one or the other; don't duplicate the work
        # The reason for the kwargs check is that standard iterator passes in by
        # args, and instantiation for iteration is 33% faster.
        args_len = len(args)
        if args_len > len(self._meta.concrete_fields):
            # Daft, but matches old exception sans the err msg.
            raise IndexError("Number of args exceeds number of fields")

        if not kwargs:
            fields_iter = iter(self._meta.concrete_fields)
            # The ordering of the zip calls matter - zip throws StopIteration
            # when an iter throws it. So if the first iter throws it, the second
            # is *not* consumed. We rely on this, so don't change the order
            # without changing the logic.
            for val, field in zip(args, fields_iter):
                if val is DEFERRED:
                    continue
                setattr(self, field.attname, val)
        else:
            # Slower, kwargs-ready version.
            fields_iter = iter(self._meta.fields)
            for val, field in zip(args, fields_iter):
                if val is DEFERRED:
                    continue
                setattr(self, field.attname, val)
                kwargs.pop(field.name, None)
                # Maintain compatibility with existing calls.
                if isinstance(field.remote_field, ManyToOneRel):
                    kwargs.pop(field.attname, None)

        # Now we're left with the unprocessed fields that *must* come from
        # keywords, or default.

        for field in fields_iter:
            is_related_object = False
            # Virtual field
            if field.attname not in kwargs and field.column is None:
                continue
            if kwargs:
                if isinstance(field.remote_field, ForeignObjectRel):
                    try:
                        # Assume object instance was passed in.
                        rel_obj = kwargs.pop(field.name)
                        is_related_object = True
                    except KeyError:
                        try:
                            # Object instance wasn't passed in -- must be an ID.
                            val = kwargs.pop(field.attname)
                        except KeyError:
                            val = field.get_default()
                    else:
                        # Object instance was passed in. Special case: You can
                        # pass in "None" for related objects if it's allowed.
                        if rel_obj is None and field.null:
                            val = None
                else:
                    try:
                        val = kwargs.pop(field.attname)
                    except KeyError:
                        # This is done with an exception rather than the
                        # default argument on pop because we don't want
                        # get_default() to be evaluated, and then not used.
                        # Refs #12057.
                        val = field.get_default()
            else:
                val = field.get_default()

            if is_related_object:
                # If we are passed a related instance, set it using the
                # field.name instead of field.attname (e.g. "user" instead of
                # "user_id") so that the object gets properly cached (and type
                # checked) by the RelatedObjectDescriptor.
                if rel_obj is not DEFERRED:
                    setattr(self, field.name, rel_obj)
            else:
                if val is not DEFERRED:
                    setattr(self, field.attname, val)

        if kwargs:
            for prop in list(kwargs):
                try:
                    # Any remaining kwargs must correspond to properties or
                    # virtual fields.
                    if (isinstance(getattr(self.__class__, prop), property) or
                            self._meta.get_field(prop)):
                        if kwargs[prop] is not DEFERRED:
                            setattr(self, prop, kwargs[prop])
                        del kwargs[prop]
                except (AttributeError, FieldDoesNotExist):
                    pass
            if kwargs:
                raise TypeError("'%s' is an invalid keyword argument for this function" % list(kwargs)[0])
        super(Model, self).__init__()
        signals.post_init.send(sender=self.__class__, instance=self)

    @classmethod
    def from_db(cls, db, field_names, values):
        if len(values) != len(cls._meta.concrete_fields):
            values = list(values)
            values.reverse()
            values = [values.pop() if f.attname in field_names else DEFERRED for f in cls._meta.concrete_fields]
        new = cls(*values)
        new._state.adding = False
        new._state.db = db
        return new

    def __repr__(self):
        try:
            u = six.text_type(self)
        except (UnicodeEncodeError, UnicodeDecodeError):
            u = '[Bad Unicode data]'
        return force_str('<%s: %s>' % (self.__class__.__name__, u))

    def __str__(self):
        if six.PY2 and hasattr(self, '__unicode__'):
            return force_text(self).encode('utf-8')
        return str('%s object' % self.__class__.__name__)

    def __eq__(self, other):
        if not isinstance(other, Model):
            return False
        if self._meta.concrete_model != other._meta.concrete_model:
            return False
        my_pk = self._get_pk_val()
        if my_pk is None:
            return self is other
        return my_pk == other._get_pk_val()

    def __ne__(self, other):
        return not self.__eq__(other)

    def __hash__(self):
        if self._get_pk_val() is None:
            raise TypeError("Model instances without primary key value are unhashable")
        return hash(self._get_pk_val())

    def __reduce__(self):
        """
        Provides pickling support. Normally, this just dispatches to Python's
        standard handling. However, for models with deferred field loading, we
        need to do things manually, as they're dynamically created classes and
        only module-level classes can be pickled by the default path.
        """
        data = self.__dict__
        data[DJANGO_VERSION_PICKLE_KEY] = get_version()
        class_id = self._meta.app_label, self._meta.object_name
        return model_unpickle, (class_id,), data

    def __setstate__(self, state):
        msg = None
        pickled_version = state.get(DJANGO_VERSION_PICKLE_KEY)
        if pickled_version:
            current_version = get_version()
            if current_version != pickled_version:
                msg = (
                    "Pickled model instance's Django version %s does not match "
                    "the current version %s." % (pickled_version, current_version)
                )
        else:
            msg = "Pickled model instance's Django version is not specified."

        if msg:
            warnings.warn(msg, RuntimeWarning, stacklevel=2)

        self.__dict__.update(state)

    def _get_pk_val(self, meta=None):
        if not meta:
            meta = self._meta
        return getattr(self, meta.pk.attname)

    def _set_pk_val(self, value):
        return setattr(self, self._meta.pk.attname, value)

    pk = property(_get_pk_val, _set_pk_val)

    def get_deferred_fields(self):
        """
        Returns a set containing names of deferred fields for this instance.
        """
        return {
            f.attname for f in self._meta.concrete_fields
            if f.attname not in self.__dict__
        }

    def refresh_from_db(self, using=None, fields=None, **kwargs):
        """
        Reloads field values from the database.

        By default, the reloading happens from the database this instance was
        loaded from, or by the read router if this instance wasn't loaded from
        any database. The using parameter will override the default.

        Fields can be used to specify which fields to reload. The fields
        should be an iterable of field attnames. If fields is None, then
        all non-deferred fields are reloaded.

        When accessing deferred fields of an instance, the deferred loading
        of the field will call this method.
        """
        if fields is not None:
            if len(fields) == 0:
                return
            if any(LOOKUP_SEP in f for f in fields):
                raise ValueError(
                    'Found "%s" in fields argument. Relations and transforms '
                    'are not allowed in fields.' % LOOKUP_SEP)

        db = using if using is not None else self._state.db
        db_instance_qs = self.__class__._default_manager.using(db).filter(pk=self.pk)

        # Use provided fields, if not set then reload all non-deferred fields.
        deferred_fields = self.get_deferred_fields()
        if fields is not None:
            fields = list(fields)
            db_instance_qs = db_instance_qs.only(*fields)
        elif deferred_fields:
            fields = [f.attname for f in self._meta.concrete_fields
                      if f.attname not in deferred_fields]
            db_instance_qs = db_instance_qs.only(*fields)

        db_instance = db_instance_qs.get()
        non_loaded_fields = db_instance.get_deferred_fields()
        for field in self._meta.concrete_fields:
            if field.attname in non_loaded_fields:
                # This field wasn't refreshed - skip ahead.
                continue
            setattr(self, field.attname, getattr(db_instance, field.attname))
            # Throw away stale foreign key references.
            if field.is_relation and field.get_cache_name() in self.__dict__:
                rel_instance = getattr(self, field.get_cache_name())
                local_val = getattr(db_instance, field.attname)
                related_val = None if rel_instance is None else getattr(rel_instance, field.target_field.attname)
                if local_val != related_val or (local_val is None and related_val is None):
                    del self.__dict__[field.get_cache_name()]
        self._state.db = db_instance._state.db

    def serializable_value(self, field_name):
        """
        Returns the value of the field name for this instance. If the field is
        a foreign key, returns the id value, instead of the object. If there's
        no Field object with this name on the model, the model attribute's
        value is returned directly.

        Used to serialize a field's value (in the serializer, or form output,
        for example). Normally, you would just access the attribute directly
        and not use this method.
        """
        try:
            field = self._meta.get_field(field_name)
        except FieldDoesNotExist:
            return getattr(self, field_name)
        return getattr(self, field.attname)

    def save(self, force_insert=False, force_update=False, using=None,
             update_fields=None):
        """
        Saves the current instance. Override this in a subclass if you want to
        control the saving process.

        The 'force_insert' and 'force_update' parameters can be used to insist
        that the "save" must be an SQL insert or update (or equivalent for
        non-SQL backends), respectively. Normally, they should not be set.
        """
        # Ensure that a model instance without a PK hasn't been assigned to
        # a ForeignKey or OneToOneField on this model. If the field is
        # nullable, allowing the save() would result in silent data loss.
        for field in self._meta.concrete_fields:
            if field.is_relation:
                # If the related field isn't cached, then an instance hasn't
                # been assigned and there's no need to worry about this check.
                try:
                    getattr(self, field.get_cache_name())
                except AttributeError:
                    continue
                obj = getattr(self, field.name, None)
                # A pk may have been assigned manually to a model instance not
                # saved to the database (or auto-generated in a case like
                # UUIDField), but we allow the save to proceed and rely on the
                # database to raise an IntegrityError if applicable. If
                # constraints aren't supported by the database, there's the
                # unavoidable risk of data corruption.
                if obj and obj.pk is None:
                    # Remove the object from a related instance cache.
                    if not field.remote_field.multiple:
                        delattr(obj, field.remote_field.get_cache_name())
                    raise ValueError(
                        "save() prohibited to prevent data loss due to "
                        "unsaved related object '%s'." % field.name
                    )

        using = using or router.db_for_write(self.__class__, instance=self)
        if force_insert and (force_update or update_fields):
            raise ValueError("Cannot force both insert and updating in model saving.")

        deferred_fields = self.get_deferred_fields()
        if update_fields is not None:
            # If update_fields is empty, skip the save. We do also check for
            # no-op saves later on for inheritance cases. This bailout is
            # still needed for skipping signal sending.
            if len(update_fields) == 0:
                return

            update_fields = frozenset(update_fields)
            field_names = set()

            for field in self._meta.fields:
                if not field.primary_key:
                    field_names.add(field.name)

                    if field.name != field.attname:
                        field_names.add(field.attname)

            non_model_fields = update_fields.difference(field_names)

            if non_model_fields:
                raise ValueError("The following fields do not exist in this "
                                 "model or are m2m fields: %s"
                                 % ', '.join(non_model_fields))

        # If saving to the same database, and this model is deferred, then
        # automatically do a "update_fields" save on the loaded fields.
        elif not force_insert and deferred_fields and using == self._state.db:
            field_names = set()
            for field in self._meta.concrete_fields:
                if not field.primary_key and not hasattr(field, 'through'):
                    field_names.add(field.attname)
            loaded_fields = field_names.difference(deferred_fields)
            if loaded_fields:
                update_fields = frozenset(loaded_fields)

        self.save_base(using=using, force_insert=force_insert,
                       force_update=force_update, update_fields=update_fields)
    save.alters_data = True

    def save_base(self, raw=False, force_insert=False,
                  force_update=False, using=None, update_fields=None):
        """
        Handles the parts of saving which should be done only once per save,
        yet need to be done in raw saves, too. This includes some sanity
        checks and signal sending.

        The 'raw' argument is telling save_base not to save any parent
        models and not to do any changes to the values before save. This
        is used by fixture loading.
        """
        using = using or router.db_for_write(self.__class__, instance=self)
        assert not (force_insert and (force_update or update_fields))
        assert update_fields is None or len(update_fields) > 0
        cls = origin = self.__class__
        # Skip proxies, but keep the origin as the proxy model.
        if cls._meta.proxy:
            cls = cls._meta.concrete_model
        meta = cls._meta
        if not meta.auto_created:
            signals.pre_save.send(sender=origin, instance=self, raw=raw, using=using,
                                  update_fields=update_fields)
        with transaction.atomic(using=using, savepoint=False):
            if not raw:
                self._save_parents(cls, using, update_fields)
            updated = self._save_table(raw, cls, force_insert, force_update, using, update_fields)
        # Store the database on which the object was saved
        self._state.db = using
        # Once saved, this is no longer a to-be-added instance.
        self._state.adding = False

        # Signal that the save is complete
        if not meta.auto_created:
            signals.post_save.send(sender=origin, instance=self, created=(not updated),
                                   update_fields=update_fields, raw=raw, using=using)

    save_base.alters_data = True

    def _save_parents(self, cls, using, update_fields):
        """
        Saves all the parents of cls using values from self.
        """
        meta = cls._meta
        for parent, field in meta.parents.items():
            # Make sure the link fields are synced between parent and self.
            if (field and getattr(self, parent._meta.pk.attname) is None and
                    getattr(self, field.attname) is not None):
                setattr(self, parent._meta.pk.attname, getattr(self, field.attname))
            self._save_parents(cls=parent, using=using, update_fields=update_fields)
            self._save_table(cls=parent, using=using, update_fields=update_fields)
            # Set the parent's PK value to self.
            if field:
                setattr(self, field.attname, self._get_pk_val(parent._meta))
                # Since we didn't have an instance of the parent handy set
                # attname directly, bypassing the descriptor. Invalidate
                # the related object cache, in case it's been accidentally
                # populated. A fresh instance will be re-built from the
                # database if necessary.
                cache_name = field.get_cache_name()
                if hasattr(self, cache_name):
                    delattr(self, cache_name)

    def _save_table(self, raw=False, cls=None, force_insert=False,
                    force_update=False, using=None, update_fields=None):
        """
        Does the heavy-lifting involved in saving. Updates or inserts the data
        for a single table.
        """
        meta = cls._meta
        non_pks = [f for f in meta.local_concrete_fields if not f.primary_key]

        if update_fields:
            non_pks = [f for f in non_pks
                       if f.name in update_fields or f.attname in update_fields]

        pk_val = self._get_pk_val(meta)
        if pk_val is None:
            pk_val = meta.pk.get_pk_value_on_save(self)
            setattr(self, meta.pk.attname, pk_val)
        pk_set = pk_val is not None
        if not pk_set and (force_update or update_fields):
            raise ValueError("Cannot force an update in save() with no primary key.")
        updated = False
        # If possible, try an UPDATE. If that doesn't update anything, do an INSERT.
        if pk_set and not force_insert:
            base_qs = cls._base_manager.using(using)
            values = [(f, None, (getattr(self, f.attname) if raw else f.pre_save(self, False)))
                      for f in non_pks]
            forced_update = update_fields or force_update
            updated = self._do_update(base_qs, using, pk_val, values, update_fields,
                                      forced_update)
            if force_update and not updated:
                raise DatabaseError("Forced update did not affect any rows.")
            if update_fields and not updated:
                raise DatabaseError("Save with update_fields did not affect any rows.")
        if not updated:
            if meta.order_with_respect_to:
                # If this is a model with an order_with_respect_to
                # autopopulate the _order field
                field = meta.order_with_respect_to
                filter_args = field.get_filter_kwargs_for_object(self)
                order_value = cls._base_manager.using(using).filter(**filter_args).count()
                self._order = order_value

            fields = meta.local_concrete_fields
            if not pk_set:
                fields = [f for f in fields if not isinstance(f, AutoField)]

            update_pk = bool(meta.has_auto_field and not pk_set)
            result = self._do_insert(cls._base_manager, using, fields, update_pk, raw)
            if update_pk:
                setattr(self, meta.pk.attname, result)
        return updated

    def _do_update(self, base_qs, using, pk_val, values, update_fields, forced_update):
        """
        This method will try to update the model. If the model was updated (in
        the sense that an update query was done and a matching row was found
        from the DB) the method will return True.
        """
        filtered = base_qs.filter(pk=pk_val)
        if not values:
            # We can end up here when saving a model in inheritance chain where
            # update_fields doesn't target any field in current model. In that
            # case we just say the update succeeded. Another case ending up here
            # is a model with just PK - in that case check that the PK still
            # exists.
            return update_fields is not None or filtered.exists()
        if self._meta.select_on_save and not forced_update:
            if filtered.exists():
                # It may happen that the object is deleted from the DB right after
                # this check, causing the subsequent UPDATE to return zero matching
                # rows. The same result can occur in some rare cases when the
                # database returns zero despite the UPDATE being executed
                # successfully (a row is matched and updated). In order to
                # distinguish these two cases, the object's existence in the
                # database is again checked for if the UPDATE query returns 0.
                return filtered._update(values) > 0 or filtered.exists()
            else:
                return False
        return filtered._update(values) > 0

    def _do_insert(self, manager, using, fields, update_pk, raw):
        """
        Do an INSERT. If update_pk is defined then this method should return
        the new pk for the model.
        """
        return manager._insert([self], fields=fields, return_id=update_pk,
                               using=using, raw=raw)

    def delete(self, using=None, keep_parents=False):
        using = using or router.db_for_write(self.__class__, instance=self)
        assert self._get_pk_val() is not None, (
            "%s object can't be deleted because its %s attribute is set to None." %
            (self._meta.object_name, self._meta.pk.attname)
        )

        collector = Collector(using=using)
        collector.collect([self], keep_parents=keep_parents)
        return collector.delete()

    delete.alters_data = True

    def _get_FIELD_display(self, field):
        value = getattr(self, field.attname)
        return force_text(dict(field.flatchoices).get(value, value), strings_only=True)

    def _get_next_or_previous_by_FIELD(self, field, is_next, **kwargs):
        if not self.pk:
            raise ValueError("get_next/get_previous cannot be used on unsaved objects.")
        op = 'gt' if is_next else 'lt'
        order = '' if is_next else '-'
        param = force_text(getattr(self, field.attname))
        q = Q(**{'%s__%s' % (field.name, op): param})
        q = q | Q(**{field.name: param, 'pk__%s' % op: self.pk})
        qs = self.__class__._default_manager.using(self._state.db).filter(**kwargs).filter(q).order_by(
            '%s%s' % (order, field.name), '%spk' % order
        )
        try:
            return qs[0]
        except IndexError:
            raise self.DoesNotExist("%s matching query does not exist." % self.__class__._meta.object_name)

    def _get_next_or_previous_in_order(self, is_next):
        cachename = "__%s_order_cache" % is_next
        if not hasattr(self, cachename):
            op = 'gt' if is_next else 'lt'
            order = '_order' if is_next else '-_order'
            order_field = self._meta.order_with_respect_to
            filter_args = order_field.get_filter_kwargs_for_object(self)
            obj = self.__class__._default_manager.filter(**filter_args).filter(**{
                '_order__%s' % op: self.__class__._default_manager.values('_order').filter(**{
                    self._meta.pk.name: self.pk
                })
            }).order_by(order)[:1].get()
            setattr(self, cachename, obj)
        return getattr(self, cachename)

    def prepare_database_save(self, field):
        if self.pk is None:
            raise ValueError("Unsaved model instance %r cannot be used in an ORM query." % self)
        return getattr(self, field.remote_field.get_related_field().attname)

    def clean(self):
        """
        Hook for doing any extra model-wide validation after clean() has been
        called on every field by self.clean_fields. Any ValidationError raised
        by this method will not be associated with a particular field; it will
        have a special-case association with the field defined by NON_FIELD_ERRORS.
        """
        pass

    def validate_unique(self, exclude=None):
        """
        Checks unique constraints on the model and raises ``ValidationError``
        if any failed.
        """
        unique_checks, date_checks = self._get_unique_checks(exclude=exclude)

        errors = self._perform_unique_checks(unique_checks)
        date_errors = self._perform_date_checks(date_checks)

        for k, v in date_errors.items():
            errors.setdefault(k, []).extend(v)

        if errors:
            raise ValidationError(errors)

    def _get_unique_checks(self, exclude=None):
        """
        Gather a list of checks to perform. Since validate_unique could be
        called from a ModelForm, some fields may have been excluded; we can't
        perform a unique check on a model that is missing fields involved
        in that check.
        Fields that did not validate should also be excluded, but they need
        to be passed in via the exclude argument.
        """
        if exclude is None:
            exclude = []
        unique_checks = []

        unique_togethers = [(self.__class__, self._meta.unique_together)]
        for parent_class in self._meta.get_parent_list():
            if parent_class._meta.unique_together:
                unique_togethers.append((parent_class, parent_class._meta.unique_together))

        for model_class, unique_together in unique_togethers:
            for check in unique_together:
                for name in check:
                    # If this is an excluded field, don't add this check.
                    if name in exclude:
                        break
                else:
                    unique_checks.append((model_class, tuple(check)))

        # These are checks for the unique_for_<date/year/month>.
        date_checks = []

        # Gather a list of checks for fields declared as unique and add them to
        # the list of checks.

        fields_with_class = [(self.__class__, self._meta.local_fields)]
        for parent_class in self._meta.get_parent_list():
            fields_with_class.append((parent_class, parent_class._meta.local_fields))

        for model_class, fields in fields_with_class:
            for f in fields:
                name = f.name
                if name in exclude:
                    continue
                if f.unique:
                    unique_checks.append((model_class, (name,)))
                if f.unique_for_date and f.unique_for_date not in exclude:
                    date_checks.append((model_class, 'date', name, f.unique_for_date))
                if f.unique_for_year and f.unique_for_year not in exclude:
                    date_checks.append((model_class, 'year', name, f.unique_for_year))
                if f.unique_for_month and f.unique_for_month not in exclude:
                    date_checks.append((model_class, 'month', name, f.unique_for_month))
        return unique_checks, date_checks

    def _perform_unique_checks(self, unique_checks):
        errors = {}

        for model_class, unique_check in unique_checks:
            # Try to look up an existing object with the same values as this
            # object's values for all the unique field.

            lookup_kwargs = {}
            for field_name in unique_check:
                f = self._meta.get_field(field_name)
                lookup_value = getattr(self, f.attname)
                if lookup_value is None:
                    # no value, skip the lookup
                    continue
                if f.primary_key and not self._state.adding:
                    # no need to check for unique primary key when editing
                    continue
                lookup_kwargs[str(field_name)] = lookup_value

            # some fields were skipped, no reason to do the check
            if len(unique_check) != len(lookup_kwargs):
                continue

            qs = model_class._default_manager.filter(**lookup_kwargs)

            # Exclude the current object from the query if we are editing an
            # instance (as opposed to creating a new one)
            # Note that we need to use the pk as defined by model_class, not
            # self.pk. These can be different fields because model inheritance
            # allows single model to have effectively multiple primary keys.
            # Refs #17615.
            model_class_pk = self._get_pk_val(model_class._meta)
            if not self._state.adding and model_class_pk is not None:
                qs = qs.exclude(pk=model_class_pk)
            if qs.exists():
                if len(unique_check) == 1:
                    key = unique_check[0]
                else:
                    key = NON_FIELD_ERRORS
                errors.setdefault(key, []).append(self.unique_error_message(model_class, unique_check))

        return errors

    def _perform_date_checks(self, date_checks):
        errors = {}
        for model_class, lookup_type, field, unique_for in date_checks:
            lookup_kwargs = {}
            # there's a ticket to add a date lookup, we can remove this special
            # case if that makes it's way in
            date = getattr(self, unique_for)
            if date is None:
                continue
            if lookup_type == 'date':
                lookup_kwargs['%s__day' % unique_for] = date.day
                lookup_kwargs['%s__month' % unique_for] = date.month
                lookup_kwargs['%s__year' % unique_for] = date.year
            else:
                lookup_kwargs['%s__%s' % (unique_for, lookup_type)] = getattr(date, lookup_type)
            lookup_kwargs[field] = getattr(self, field)

            qs = model_class._default_manager.filter(**lookup_kwargs)
            # Exclude the current object from the query if we are editing an
            # instance (as opposed to creating a new one)
            if not self._state.adding and self.pk is not None:
                qs = qs.exclude(pk=self.pk)

            if qs.exists():
                errors.setdefault(field, []).append(
                    self.date_error_message(lookup_type, field, unique_for)
                )
        return errors

    def date_error_message(self, lookup_type, field_name, unique_for):
        opts = self._meta
        field = opts.get_field(field_name)
        return ValidationError(
            message=field.error_messages['unique_for_date'],
            code='unique_for_date',
            params={
                'model': self,
                'model_name': six.text_type(capfirst(opts.verbose_name)),
                'lookup_type': lookup_type,
                'field': field_name,
                'field_label': six.text_type(capfirst(field.verbose_name)),
                'date_field': unique_for,
                'date_field_label': six.text_type(capfirst(opts.get_field(unique_for).verbose_name)),
            }
        )

    def unique_error_message(self, model_class, unique_check):
        opts = model_class._meta

        params = {
            'model': self,
            'model_class': model_class,
            'model_name': six.text_type(capfirst(opts.verbose_name)),
            'unique_check': unique_check,
        }

        # A unique field
        if len(unique_check) == 1:
            field = opts.get_field(unique_check[0])
            params['field_label'] = six.text_type(capfirst(field.verbose_name))
            return ValidationError(
                message=field.error_messages['unique'],
                code='unique',
                params=params,
            )

        # unique_together
        else:
            field_labels = [capfirst(opts.get_field(f).verbose_name) for f in unique_check]
            params['field_labels'] = six.text_type(get_text_list(field_labels, _('and')))
            return ValidationError(
                message=_("%(model_name)s with this %(field_labels)s already exists."),
                code='unique_together',
                params=params,
            )

    def full_clean(self, exclude=None, validate_unique=True):
        """
        Calls clean_fields, clean, and validate_unique, on the model,
        and raises a ``ValidationError`` for any errors that occurred.
        """
        errors = {}
        if exclude is None:
            exclude = []
        else:
            exclude = list(exclude)

        try:
            self.clean_fields(exclude=exclude)
        except ValidationError as e:
            errors = e.update_error_dict(errors)

        # Form.clean() is run even if other validation fails, so do the
        # same with Model.clean() for consistency.
        try:
            self.clean()
        except ValidationError as e:
            errors = e.update_error_dict(errors)

        # Run unique checks, but only for fields that passed validation.
        if validate_unique:
            for name in errors.keys():
                if name != NON_FIELD_ERRORS and name not in exclude:
                    exclude.append(name)
            try:
                self.validate_unique(exclude=exclude)
            except ValidationError as e:
                errors = e.update_error_dict(errors)

        if errors:
            raise ValidationError(errors)

    def clean_fields(self, exclude=None):
        """
        Cleans all fields and raises a ValidationError containing a dict
        of all validation errors if any occur.
        """
        if exclude is None:
            exclude = []

        errors = {}
        for f in self._meta.fields:
            if f.name in exclude:
                continue
            # Skip validation for empty fields with blank=True. The developer
            # is responsible for making sure they have a valid value.
            raw_value = getattr(self, f.attname)
            if f.blank and raw_value in f.empty_values:
                continue
            try:
                setattr(self, f.attname, f.clean(raw_value, self))
            except ValidationError as e:
                errors[f.name] = e.error_list

        if errors:
            raise ValidationError(errors)

    @classmethod
    def check(cls, **kwargs):
        errors = []
        errors.extend(cls._check_swappable())
        errors.extend(cls._check_model())
        errors.extend(cls._check_managers(**kwargs))
        if not cls._meta.swapped:
            errors.extend(cls._check_fields(**kwargs))
            errors.extend(cls._check_m2m_through_same_relationship())
            errors.extend(cls._check_long_column_names())
            clash_errors = cls._check_id_field() + cls._check_field_name_clashes()
            errors.extend(clash_errors)
            # If there are field name clashes, hide consequent column name
            # clashes.
            if not clash_errors:
                errors.extend(cls._check_column_name_clashes())
            errors.extend(cls._check_index_together())
            errors.extend(cls._check_unique_together())
            errors.extend(cls._check_ordering())

        return errors

    @classmethod
    def _check_swappable(cls):
        """ Check if the swapped model exists. """

        errors = []
        if cls._meta.swapped:
            try:
                apps.get_model(cls._meta.swapped)
            except ValueError:
                errors.append(
                    checks.Error(
                        "'%s' is not of the form 'app_label.app_name'." % cls._meta.swappable,
                        id='models.E001',
                    )
                )
            except LookupError:
                app_label, model_name = cls._meta.swapped.split('.')
                errors.append(
                    checks.Error(
                        "'%s' references '%s.%s', which has not been "
                        "installed, or is abstract." % (
                            cls._meta.swappable, app_label, model_name
                        ),
                        id='models.E002',
                    )
                )
        return errors

    @classmethod
    def _check_model(cls):
        errors = []
        if cls._meta.proxy:
            if cls._meta.local_fields or cls._meta.local_many_to_many:
                errors.append(
                    checks.Error(
                        "Proxy model '%s' contains model fields." % cls.__name__,
                        id='models.E017',
                    )
                )
        return errors

    @classmethod
    def _check_managers(cls, **kwargs):
        """ Perform all manager checks. """

        errors = []
        for manager in cls._meta.managers:
            errors.extend(manager.check(**kwargs))
        return errors

    @classmethod
    def _check_fields(cls, **kwargs):
        """ Perform all field checks. """

        errors = []
        for field in cls._meta.local_fields:
            errors.extend(field.check(**kwargs))
        for field in cls._meta.local_many_to_many:
            errors.extend(field.check(from_model=cls, **kwargs))
        return errors

    @classmethod
    def _check_m2m_through_same_relationship(cls):
        """ Check if no relationship model is used by more than one m2m field.
        """

        errors = []
        seen_intermediary_signatures = []

        fields = cls._meta.local_many_to_many

        # Skip when the target model wasn't found.
        fields = (f for f in fields if isinstance(f.remote_field.model, ModelBase))

        # Skip when the relationship model wasn't found.
        fields = (f for f in fields if isinstance(f.remote_field.through, ModelBase))

        for f in fields:
            signature = (f.remote_field.model, cls, f.remote_field.through)
            if signature in seen_intermediary_signatures:
                errors.append(
                    checks.Error(
                        "The model has two many-to-many relations through "
                        "the intermediate model '%s'." % f.remote_field.through._meta.label,
                        obj=cls,
                        id='models.E003',
                    )
                )
            else:
                seen_intermediary_signatures.append(signature)
        return errors

    @classmethod
    def _check_id_field(cls):
        """ Check if `id` field is a primary key. """
        fields = list(f for f in cls._meta.local_fields if f.name == 'id' and f != cls._meta.pk)
        # fields is empty or consists of the invalid "id" field
        if fields and not fields[0].primary_key and cls._meta.pk.name == 'id':
            return [
                checks.Error(
                    "'id' can only be used as a field name if the field also "
                    "sets 'primary_key=True'.",
                    obj=cls,
                    id='models.E004',
                )
            ]
        else:
            return []

    @classmethod
    def _check_field_name_clashes(cls):
        """ Ref #17673. """

        errors = []
        used_fields = {}  # name or attname -> field

        # Check that multi-inheritance doesn't cause field name shadowing.
        for parent in cls._meta.get_parent_list():
            for f in parent._meta.local_fields:
                clash = used_fields.get(f.name) or used_fields.get(f.attname) or None
                if clash:
                    errors.append(
                        checks.Error(
                            "The field '%s' from parent model "
                            "'%s' clashes with the field '%s' "
                            "from parent model '%s'." % (
                                clash.name, clash.model._meta,
                                f.name, f.model._meta
                            ),
                            obj=cls,
                            id='models.E005',
                        )
                    )
                used_fields[f.name] = f
                used_fields[f.attname] = f

        # Check that fields defined in the model don't clash with fields from
        # parents, including auto-generated fields like multi-table inheritance
        # child accessors.
        for parent in cls._meta.get_parent_list():
            for f in parent._meta.get_fields():
                if f not in used_fields:
                    used_fields[f.name] = f

        for f in cls._meta.local_fields:
            clash = used_fields.get(f.name) or used_fields.get(f.attname) or None
            # Note that we may detect clash between user-defined non-unique
            # field "id" and automatically added unique field "id", both
            # defined at the same model. This special case is considered in
            # _check_id_field and here we ignore it.
            id_conflict = f.name == "id" and clash and clash.name == "id" and clash.model == cls
            if clash and not id_conflict:
                errors.append(
                    checks.Error(
                        "The field '%s' clashes with the field '%s' "
                        "from model '%s'." % (
                            f.name, clash.name, clash.model._meta
                        ),
                        obj=f,
                        id='models.E006',
                    )
                )
            used_fields[f.name] = f
            used_fields[f.attname] = f

        return errors

    @classmethod
    def _check_column_name_clashes(cls):
        # Store a list of column names which have already been used by other fields.
        used_column_names = []
        errors = []

        for f in cls._meta.local_fields:
            _, column_name = f.get_attname_column()

            # Ensure the column name is not already in use.
            if column_name and column_name in used_column_names:
                errors.append(
                    checks.Error(
                        "Field '%s' has column name '%s' that is used by "
                        "another field." % (f.name, column_name),
                        hint="Specify a 'db_column' for the field.",
                        obj=cls,
                        id='models.E007'
                    )
                )
            else:
                used_column_names.append(column_name)

        return errors

    @classmethod
    def _check_index_together(cls):
        """ Check the value of "index_together" option. """
        if not isinstance(cls._meta.index_together, (tuple, list)):
            return [
                checks.Error(
                    "'index_together' must be a list or tuple.",
                    obj=cls,
                    id='models.E008',
                )
            ]

        elif any(not isinstance(fields, (tuple, list)) for fields in cls._meta.index_together):
            return [
                checks.Error(
                    "All 'index_together' elements must be lists or tuples.",
                    obj=cls,
                    id='models.E009',
                )
            ]

        else:
            errors = []
            for fields in cls._meta.index_together:
                errors.extend(cls._check_local_fields(fields, "index_together"))
            return errors

    @classmethod
    def _check_unique_together(cls):
        """ Check the value of "unique_together" option. """
        if not isinstance(cls._meta.unique_together, (tuple, list)):
            return [
                checks.Error(
                    "'unique_together' must be a list or tuple.",
                    obj=cls,
                    id='models.E010',
                )
            ]

        elif any(not isinstance(fields, (tuple, list)) for fields in cls._meta.unique_together):
            return [
                checks.Error(
                    "All 'unique_together' elements must be lists or tuples.",
                    obj=cls,
                    id='models.E011',
                )
            ]

        else:
            errors = []
            for fields in cls._meta.unique_together:
                errors.extend(cls._check_local_fields(fields, "unique_together"))
            return errors

    @classmethod
    def _check_local_fields(cls, fields, option):
        from django.db import models

        # In order to avoid hitting the relation tree prematurely, we use our
        # own fields_map instead of using get_field()
        forward_fields_map = {
            field.name: field for field in cls._meta._get_fields(reverse=False)
        }

        errors = []
        for field_name in fields:
            try:
                field = forward_fields_map[field_name]
            except KeyError:
                errors.append(
                    checks.Error(
                        "'%s' refers to the non-existent field '%s'." % (
                            option, field_name,
                        ),
                        obj=cls,
                        id='models.E012',
                    )
                )
            else:
                if isinstance(field.remote_field, models.ManyToManyRel):
                    errors.append(
                        checks.Error(
                            "'%s' refers to a ManyToManyField '%s', but "
                            "ManyToManyFields are not permitted in '%s'." % (
                                option, field_name, option,
                            ),
                            obj=cls,
                            id='models.E013',
                        )
                    )
                elif field not in cls._meta.local_fields:
                    errors.append(
                        checks.Error(
                            "'%s' refers to field '%s' which is not local to model '%s'."
                            % (option, field_name, cls._meta.object_name),
                            hint="This issue may be caused by multi-table inheritance.",
                            obj=cls,
                            id='models.E016',
                        )
                    )
        return errors

    @classmethod
    def _check_ordering(cls):
        """ Check "ordering" option -- is it a list of strings and do all fields
        exist? """
        if cls._meta._ordering_clash:
            return [
                checks.Error(
                    "'ordering' and 'order_with_respect_to' cannot be used together.",
                    obj=cls,
                    id='models.E021',
                ),
            ]

        if cls._meta.order_with_respect_to or not cls._meta.ordering:
            return []

        if not isinstance(cls._meta.ordering, (list, tuple)):
            return [
                checks.Error(
                    "'ordering' must be a tuple or list (even if you want to order by only one field).",
                    obj=cls,
                    id='models.E014',
                )
            ]

        errors = []
        fields = cls._meta.ordering

        # Skip '?' fields.
        fields = (f for f in fields if f != '?')

        # Convert "-field" to "field".
        fields = ((f[1:] if f.startswith('-') else f) for f in fields)

        # Skip ordering in the format field1__field2 (FIXME: checking
        # this format would be nice, but it's a little fiddly).
        fields = (f for f in fields if '__' not in f)

        # Skip ordering on pk. This is always a valid order_by field
        # but is an alias and therefore won't be found by opts.get_field.
        fields = {f for f in fields if f != 'pk'}

        # Check for invalid or non-existent fields in ordering.
        invalid_fields = []

        # Any field name that is not present in field_names does not exist.
        # Also, ordering by m2m fields is not allowed.
        opts = cls._meta
        valid_fields = set(chain.from_iterable(
            (f.name, f.attname) if not (f.auto_created and not f.concrete) else (f.field.related_query_name(),)
            for f in chain(opts.fields, opts.related_objects)
        ))

        invalid_fields.extend(fields - valid_fields)

        for invalid_field in invalid_fields:
            errors.append(
                checks.Error(
                    "'ordering' refers to the non-existent field '%s'." % invalid_field,
                    obj=cls,
                    id='models.E015',
                )
            )
        return errors

    @classmethod
    def _check_long_column_names(cls):
        """
        Check that any auto-generated column names are shorter than the limits
        for each database in which the model will be created.
        """
        errors = []
        allowed_len = None
        db_alias = None

        # Find the minimum max allowed length among all specified db_aliases.
        for db in settings.DATABASES.keys():
            # skip databases where the model won't be created
            if not router.allow_migrate_model(db, cls):
                continue
            connection = connections[db]
            max_name_length = connection.ops.max_name_length()
            if max_name_length is None or connection.features.truncates_names:
                continue
            else:
                if allowed_len is None:
                    allowed_len = max_name_length
                    db_alias = db
                elif max_name_length < allowed_len:
                    allowed_len = max_name_length
                    db_alias = db

        if allowed_len is None:
            return errors

        for f in cls._meta.local_fields:
            _, column_name = f.get_attname_column()

            # Check if auto-generated name for the field is too long
            # for the database.
            if f.db_column is None and column_name is not None and len(column_name) > allowed_len:
                errors.append(
                    checks.Error(
                        'Autogenerated column name too long for field "%s". '
                        'Maximum length is "%s" for database "%s".'
                        % (column_name, allowed_len, db_alias),
                        hint="Set the column name manually using 'db_column'.",
                        obj=cls,
                        id='models.E018',
                    )
                )

        for f in cls._meta.local_many_to_many:
            # Check if auto-generated name for the M2M field is too long
            # for the database.
            for m2m in f.remote_field.through._meta.local_fields:
                _, rel_name = m2m.get_attname_column()
                if m2m.db_column is None and rel_name is not None and len(rel_name) > allowed_len:
                    errors.append(
                        checks.Error(
                            'Autogenerated column name too long for M2M field '
                            '"%s". Maximum length is "%s" for database "%s".'
                            % (rel_name, allowed_len, db_alias),
                            hint=(
                                "Use 'through' to create a separate model for "
                                "M2M and then set column_name using 'db_column'."
                            ),
                            obj=cls,
                            id='models.E019',
                        )
                    )

        return errors


############################################
# HELPER FUNCTIONS (CURRIED MODEL METHODS) #
############################################

# ORDERING METHODS #########################

def method_set_order(ordered_obj, self, id_list, using=None):
    if using is None:
        using = DEFAULT_DB_ALIAS
    order_wrt = ordered_obj._meta.order_with_respect_to
    filter_args = order_wrt.get_forward_related_filter(self)
    # FIXME: It would be nice if there was an "update many" version of update
    # for situations like this.
    with transaction.atomic(using=using, savepoint=False):
        for i, j in enumerate(id_list):
            ordered_obj.objects.filter(pk=j, **filter_args).update(_order=i)


def method_get_order(ordered_obj, self):
    order_wrt = ordered_obj._meta.order_with_respect_to
    filter_args = order_wrt.get_forward_related_filter(self)
    pk_name = ordered_obj._meta.pk.name
    return ordered_obj.objects.filter(**filter_args).values_list(pk_name, flat=True)


def make_foreign_order_accessors(model, related_model):
    setattr(
        related_model,
        'get_%s_order' % model.__name__.lower(),
        curry(method_get_order, model)
    )
    setattr(
        related_model,
        'set_%s_order' % model.__name__.lower(),
        curry(method_set_order, model)
    )

########
# MISC #
########


def model_unpickle(model_id):
    """
    Used to unpickle Model subclasses with deferred fields.
    """
    if isinstance(model_id, tuple):
        model = apps.get_model(*model_id)
    else:
        # Backwards compat - the model was cached directly in earlier versions.
        model = model_id
    return model.__new__(model)
model_unpickle.__safe_for_unpickle__ = True


def unpickle_inner_exception(klass, exception_name):
    # Get the exception class from the class it is attached to:
    exception = getattr(klass, exception_name)
    return exception.__new__(exception)