path: root/contrib/python/zope.interface/py2/zope/interface/declarations.py

##############################################################################
# Copyright (c) 2003 Zope Foundation and Contributors.
# All Rights Reserved.
#
# This software is subject to the provisions of the Zope Public License,
# Version 2.1 (ZPL).  A copy of the ZPL should accompany this distribution.
# THIS SOFTWARE IS PROVIDED "AS IS" AND ANY AND ALL EXPRESS OR IMPLIED
# WARRANTIES ARE DISCLAIMED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
# WARRANTIES OF TITLE, MERCHANTABILITY, AGAINST INFRINGEMENT, AND FITNESS
# FOR A PARTICULAR PURPOSE.
##############################################################################
"""Implementation of interface declarations

There are three flavors of declarations:

  - Declarations are used to simply name declared interfaces.

  - ImplementsDeclarations are used to express the interfaces that a
    class implements (that instances of the class provides).

    Implements specifications support inheriting interfaces.

  - ProvidesDeclarations are used to express interfaces directly
    provided by objects.

"""
__docformat__ = 'restructuredtext'

import sys
from types import FunctionType
from types import MethodType
from types import ModuleType
import weakref

from zope.interface.advice import addClassAdvisor
from zope.interface.interface import Interface
from zope.interface.interface import InterfaceClass
from zope.interface.interface import SpecificationBase
from zope.interface.interface import Specification
from zope.interface.interface import NameAndModuleComparisonMixin
from zope.interface._compat import CLASS_TYPES as DescriptorAwareMetaClasses
from zope.interface._compat import PYTHON3
from zope.interface._compat import _use_c_impl

__all__ = [
    # None. The public APIs of this module are
    # re-exported from zope.interface directly.
]

# pylint:disable=too-many-lines

# Registry of class-implementation specifications
BuiltinImplementationSpecifications = {}

_ADVICE_ERROR = ('Class advice impossible in Python3.  '
                 'Use the @%s class decorator instead.')

_ADVICE_WARNING = ('The %s API is deprecated, and will not work in Python3  '
                   'Use the @%s class decorator instead.')

def _next_super_class(ob):
    # When ``ob`` is an instance of ``super``, return
    # the next class in the MRO that we should actually be
    # looking at. Watch out for diamond inheritance!
    self_class = ob.__self_class__
    class_that_invoked_super = ob.__thisclass__
    complete_mro = self_class.__mro__
    next_class = complete_mro[complete_mro.index(class_that_invoked_super) + 1]
    return next_class

class named(object):

    def __init__(self, name):
        self.name = name

    def __call__(self, ob):
        ob.__component_name__ = self.name
        return ob


class Declaration(Specification):
    """Interface declarations"""

    __slots__ = ()

    def __init__(self, *bases):
        Specification.__init__(self, _normalizeargs(bases))

    def __contains__(self, interface):
        """Test whether an interface is in the specification
        """

        return self.extends(interface) and interface in self.interfaces()

    def __iter__(self):
        """Return an iterator for the interfaces in the specification
        """
        return self.interfaces()

    def flattened(self):
        """Return an iterator of all included and extended interfaces
        """
        return iter(self.__iro__)

    def __sub__(self, other):
        """Remove interfaces from a specification
        """
        return Declaration(*[
            i for i in self.interfaces()
            if not [
                j
                for j in other.interfaces()
                if i.extends(j, 0) # non-strict extends
            ]
        ])

    def __add__(self, other):
        """
        Add two specifications or a specification and an interface
        and produce a new declaration.

        .. versionchanged:: 5.4.0
           Now tries to preserve a consistent resolution order. Interfaces
           being added to this object are added to the front of the resulting resolution
           order if they already extend an interface in this object. Previously,
           they were always added to the end of the order, which easily resulted in
           invalid orders.
        """
        before = []
        result = list(self.interfaces())
        seen = set(result)
        for i in other.interfaces():
            if i in seen:
                continue
            seen.add(i)
            if any(i.extends(x) for x in result):
                # It already extends us, e.g., is a subclass,
                # so it needs to go at the front of the RO.
                before.append(i)
            else:
                result.append(i)
        return Declaration(*(before + result))

    # XXX: Is __radd__ needed? No tests break if it's removed.
    # If it is needed, does it need to handle the C3 ordering differently?
    # I (JAM) don't *think* it does.
    __radd__ = __add__

    @staticmethod
    def _add_interfaces_to_cls(interfaces, cls):
        # Strip redundant interfaces already provided
        # by the cls so we don't produce invalid
        # resolution orders.
        implemented_by_cls = implementedBy(cls)
        interfaces = tuple([
            iface
            for iface in interfaces
            if not implemented_by_cls.isOrExtends(iface)
        ])
        return interfaces + (implemented_by_cls,)

    @staticmethod
    def _argument_names_for_repr(interfaces):
        # These don't actually have to be interfaces, they could be other
        # Specification objects like Implements. Also, the first
        # one is typically/nominally the cls.
        ordered_names = []
        names = set()
        for iface in interfaces:
            duplicate_transform = repr
            if isinstance(iface, InterfaceClass):
                # Special case to get 'foo.bar.IFace'
                # instead of '<InterfaceClass foo.bar.IFace>'
                this_name = iface.__name__
                duplicate_transform = str
            elif isinstance(iface, type):
                # Likewise for types. (Ignoring legacy old-style
                # classes.)
                this_name = iface.__name__
                duplicate_transform = _implements_name
            elif (isinstance(iface, Implements)
                  and not iface.declared
                  and iface.inherit in interfaces):
                # If nothing is declared, there's no need to even print this;
                # it would just show as ``classImplements(Class)``, and the
                # ``Class`` has typically already.
                continue
            else:
                this_name = repr(iface)

            already_seen = this_name in names
            names.add(this_name)
            if already_seen:
                this_name = duplicate_transform(iface)

            ordered_names.append(this_name)
        return ', '.join(ordered_names)


class _ImmutableDeclaration(Declaration):
    # A Declaration that is immutable. Used as a singleton to
    # return empty answers for things like ``implementedBy``.
    # We have to define the actual singleton after normalizeargs
    # is defined, and that in turn is defined after InterfaceClass and
    # Implements.

    __slots__ = ()

    __instance = None

    def __new__(cls):
        if _ImmutableDeclaration.__instance is None:
            _ImmutableDeclaration.__instance = object.__new__(cls)
        return _ImmutableDeclaration.__instance

    def __reduce__(self):
        return "_empty"

    @property
    def __bases__(self):
        return ()

    @__bases__.setter
    def __bases__(self, new_bases):
        # We expect the superclass constructor to set ``self.__bases__ = ()``.
        # Rather than attempt to special case that in the constructor and allow
        # setting __bases__ only at that time, it's easier to just allow setting
        # the empty tuple at any time. That makes ``x.__bases__ = x.__bases__`` a nice
        # no-op too. (Skipping the superclass constructor altogether is a recipe
        # for maintenance headaches.)
        if new_bases != ():
            raise TypeError("Cannot set non-empty bases on shared empty Declaration.")

    # As the immutable empty declaration, we cannot be changed.
    # This means there's no logical reason for us to have dependents
    # or subscriptions: we'll never notify them. So there's no need for
    # us to keep track of any of that.
    @property
    def dependents(self):
        return {}

    changed = subscribe = unsubscribe = lambda self, _ignored: None

    def interfaces(self):
        # An empty iterator
        return iter(())

    def extends(self, interface, strict=True):
        return interface is self._ROOT

    def get(self, name, default=None):
        return default

    def weakref(self, callback=None):
        # We're a singleton, we never go away. So there's no need to return
        # distinct weakref objects here; their callbacks will never
        # be called. Instead, we only need to return a callable that
        # returns ourself. The easiest one is to return _ImmutableDeclaration
        # itself; testing on Python 3.8 shows that's faster than a function that
        # returns _empty. (Remember, one goal is to avoid allocating any
        # object, and that includes a method.)
        return _ImmutableDeclaration

    @property
    def _v_attrs(self):
        # _v_attrs is not a public, documented property, but some client
        # code uses it anyway as a convenient place to cache things. To keep
        # the empty declaration truly immutable, we must ignore that. That includes
        # ignoring assignments as well.
        return {}

    @_v_attrs.setter
    def _v_attrs(self, new_attrs):
        pass


##############################################################################
#
# Implementation specifications
#
# These specify interfaces implemented by instances of classes

class Implements(NameAndModuleComparisonMixin,
                 Declaration):
    # Inherit from NameAndModuleComparisonMixin to be
    # mutually comparable with InterfaceClass objects.
    # (The two must be mutually comparable to be able to work in e.g., BTrees.)
    # Instances of this class generally don't have a __module__ other than
    # `zope.interface.declarations`, whereas they *do* have a __name__ that is the
    # fully qualified name of the object they are representing.

    # Note, though, that equality and hashing are still identity based. This
    # accounts for things like nested objects that have the same name (typically
    # only in tests) and is consistent with pickling. As far as comparisons to InterfaceClass
    # goes, we'll never have equal name and module to those, so we're still consistent there.
    # Instances of this class are essentially intended to be unique and are
    # heavily cached (note how our __reduce__ handles this) so having identity
    # based hash and eq should also work.

    # We want equality and hashing to be based on identity. However, we can't actually
    # implement __eq__/__ne__ to do this because sometimes we get wrapped in a proxy.
    # We need to let the proxy types implement these methods so they can handle unwrapping
    # and then rely on: (1) the interpreter automatically changing `implements == proxy` into
    # `proxy == implements` (which will call proxy.__eq__ to do the unwrapping) and then
    # (2) the default equality and hashing semantics being identity based.

    # class whose specification should be used as additional base
    inherit = None

    # interfaces actually declared for a class
    declared = ()

    # Weak cache of {class: <implements>} for super objects.
    # Created on demand. These are rare, as of 5.0 anyway. Using a class
    # level default doesn't take space in instances. Using _v_attrs would be
    # another place to store this without taking space unless needed.
    _super_cache = None

    __name__ = '?'

    @classmethod
    def named(cls, name, *bases):
        # Implementation method: Produce an Implements interface with
        # a fully fleshed out __name__ before calling the constructor, which
        # sets bases to the given interfaces and which may pass this object to
        # other objects (e.g., to adjust dependents). If they're sorting or comparing
        # by name, this needs to be set.
        inst = cls.__new__(cls)
        inst.__name__ = name
        inst.__init__(*bases)
        return inst

    def changed(self, originally_changed):
        try:
            del self._super_cache
        except AttributeError:
            pass
        return super(Implements, self).changed(originally_changed)

    def __repr__(self):
        if self.inherit:
            name = getattr(self.inherit, '__name__', None) or _implements_name(self.inherit)
        else:
            name = self.__name__
        declared_names = self._argument_names_for_repr(self.declared)
        if declared_names:
            declared_names = ', ' + declared_names
        return 'classImplements(%s%s)' % (name, declared_names)

    def __reduce__(self):
        return implementedBy, (self.inherit, )


def _implements_name(ob):
    # Return the __name__ attribute to be used by its __implemented__
    # property.
    # This must be stable for the "same" object across processes
    # because it is used for sorting. It needn't be unique, though, in cases
    # like nested classes named Foo created by different functions, because
    # equality and hashing is still based on identity.
    # It might be nice to use __qualname__ on Python 3, but that would produce
    # different values between Py2 and Py3.
    return (getattr(ob, '__module__', '?') or '?') + \
        '.' + (getattr(ob, '__name__', '?') or '?')


def _implementedBy_super(sup):
    # TODO: This is now simple enough we could probably implement
    # in C if needed.

    # If the class MRO is strictly linear, we could just
    # follow the normal algorithm for the next class in the
    # search order (e.g., just return
    # ``implemented_by_next``). But when diamond inheritance
    # or mixins + interface declarations are present, we have
    # to consider the whole MRO and compute a new Implements
    # that excludes the classes being skipped over but
    # includes everything else.
    implemented_by_self = implementedBy(sup.__self_class__)
    cache = implemented_by_self._super_cache # pylint:disable=protected-access
    if cache is None:
        cache = implemented_by_self._super_cache = weakref.WeakKeyDictionary()

    key = sup.__thisclass__
    try:
        return cache[key]
    except KeyError:
        pass

    next_cls = _next_super_class(sup)
    # For ``implementedBy(cls)``:
    # .__bases__ is .declared + [implementedBy(b) for b in cls.__bases__]
    # .inherit is cls

    implemented_by_next = implementedBy(next_cls)
    mro = sup.__self_class__.__mro__
    ix_next_cls = mro.index(next_cls)
    classes_to_keep = mro[ix_next_cls:]
    new_bases = [implementedBy(c) for c in classes_to_keep]

    new = Implements.named(
        implemented_by_self.__name__ + ':' + implemented_by_next.__name__,
        *new_bases
    )
    new.inherit = implemented_by_next.inherit
    new.declared = implemented_by_next.declared
    # I don't *think* that new needs to subscribe to ``implemented_by_self``;
    # it auto-subscribed to its bases, and that should be good enough.
    cache[key] = new

    return new


@_use_c_impl
def implementedBy(cls): # pylint:disable=too-many-return-statements,too-many-branches
    """Return the interfaces implemented for a class' instances

      The value returned is an `~zope.interface.interfaces.IDeclaration`.
    """
    try:
        if isinstance(cls, super):
            # Yes, this needs to be inside the try: block. Some objects
            # like security proxies even break isinstance.
            return _implementedBy_super(cls)

        spec = cls.__dict__.get('__implemented__')
    except AttributeError:

        # we can't get the class dict. This is probably due to a
        # security proxy.  If this is the case, then probably no
        # descriptor was installed for the class.

        # We don't want to depend directly on zope.security in
        # zope.interface, but we'll try to make reasonable
        # accommodations in an indirect way.

        # We'll check to see if there's an implements:

        spec = getattr(cls, '__implemented__', None)
        if spec is None:
            # There's no spec stred in the class. Maybe its a builtin:
            spec = BuiltinImplementationSpecifications.get(cls)
            if spec is not None:
                return spec
            return _empty

        if spec.__class__ == Implements:
            # we defaulted to _empty or there was a spec. Good enough.
            # Return it.
            return spec

        # TODO: need old style __implements__ compatibility?
        # Hm, there's an __implemented__, but it's not a spec. Must be
        # an old-style declaration. Just compute a spec for it
        return Declaration(*_normalizeargs((spec, )))

    if isinstance(spec, Implements):
        return spec

    if spec is None:
        spec = BuiltinImplementationSpecifications.get(cls)
        if spec is not None:
            return spec

    # TODO: need old style __implements__ compatibility?
    spec_name = _implements_name(cls)
    if spec is not None:
        # old-style __implemented__ = foo declaration
        spec = (spec, ) # tuplefy, as it might be just an int
        spec = Implements.named(spec_name, *_normalizeargs(spec))
        spec.inherit = None    # old-style implies no inherit
        del cls.__implemented__ # get rid of the old-style declaration
    else:
        try:
            bases = cls.__bases__
        except AttributeError:
            if not callable(cls):
                raise TypeError("ImplementedBy called for non-factory", cls)
            bases = ()

        spec = Implements.named(spec_name, *[implementedBy(c) for c in bases])
        spec.inherit = cls

    try:
        cls.__implemented__ = spec
        if not hasattr(cls, '__providedBy__'):
            cls.__providedBy__ = objectSpecificationDescriptor

        if (isinstance(cls, DescriptorAwareMetaClasses)
                and '__provides__' not in cls.__dict__):
            # Make sure we get a __provides__ descriptor
            cls.__provides__ = ClassProvides(
                cls,
                getattr(cls, '__class__', type(cls)),
                )

    except TypeError:
        if not isinstance(cls, type):
            raise TypeError("ImplementedBy called for non-type", cls)
        BuiltinImplementationSpecifications[cls] = spec

    return spec


def classImplementsOnly(cls, *interfaces):
    """
    Declare the only interfaces implemented by instances of a class

    The arguments after the class are one or more interfaces or interface
    specifications (`~zope.interface.interfaces.IDeclaration` objects).

    The interfaces given (including the interfaces in the specifications)
    replace any previous declarations, *including* inherited definitions. If you
    wish to preserve inherited declarations, you can pass ``implementedBy(cls)``
    in *interfaces*. This can be used to alter the interface resolution order.
    """
    spec = implementedBy(cls)
    # Clear out everything inherited. It's important to
    # also clear the bases right now so that we don't improperly discard
    # interfaces that are already implemented by *old* bases that we're
    # about to get rid of.
    spec.declared = ()
    spec.inherit = None
    spec.__bases__ = ()
    _classImplements_ordered(spec, interfaces, ())


def classImplements(cls, *interfaces):
    """
    Declare additional interfaces implemented for instances of a class

    The arguments after the class are one or more interfaces or
    interface specifications (`~zope.interface.interfaces.IDeclaration` objects).

    The interfaces given (including the interfaces in the specifications)
    are added to any interfaces previously declared. An effort is made to
    keep a consistent C3 resolution order, but this cannot be guaranteed.

    .. versionchanged:: 5.0.0
       Each individual interface in *interfaces* may be added to either the
       beginning or end of the list of interfaces declared for *cls*,
       based on inheritance, in order to try to maintain a consistent
       resolution order. Previously, all interfaces were added to the end.
    .. versionchanged:: 5.1.0
       If *cls* is already declared to implement an interface (or derived interface)
       in *interfaces* through inheritance, the interface is ignored. Previously, it
       would redundantly be made direct base of *cls*, which often produced inconsistent
       interface resolution orders. Now, the order will be consistent, but may change.
       Also, if the ``__bases__`` of the *cls* are later changed, the *cls* will no
       longer be considered to implement such an interface (changing the ``__bases__`` of *cls*
       has never been supported).
    """
    spec = implementedBy(cls)
    interfaces = tuple(_normalizeargs(interfaces))

    before = []
    after = []

    # Take steps to try to avoid producing an invalid resolution
    # order, while still allowing for BWC (in the past, we always
    # appended)
    for iface in interfaces:
        for b in spec.declared:
            if iface.extends(b):
                before.append(iface)
                break
        else:
            after.append(iface)
    _classImplements_ordered(spec, tuple(before), tuple(after))


def classImplementsFirst(cls, iface):
    """
    Declare that instances of *cls* additionally provide *iface*.

    The second argument is an interface or interface specification.
    It is added as the highest priority (first in the IRO) interface;
    no attempt is made to keep a consistent resolution order.

    .. versionadded:: 5.0.0
    """
    spec = implementedBy(cls)
    _classImplements_ordered(spec, (iface,), ())


def _classImplements_ordered(spec, before=(), after=()):
    # Elide everything already inherited.
    # Except, if it is the root, and we don't already declare anything else
    # that would imply it, allow the root through. (TODO: When we disallow non-strict
    # IRO, this part of the check can be removed because it's not possible to re-declare
    # like that.)
    before = [
        x
        for x in before
        if not spec.isOrExtends(x) or (x is Interface and not spec.declared)
    ]
    after = [
        x
        for x in after
        if not spec.isOrExtends(x) or (x is Interface and not spec.declared)
    ]

    # eliminate duplicates
    new_declared = []
    seen = set()
    for l in before, spec.declared, after:
        for b in l:
            if b not in seen:
                new_declared.append(b)
                seen.add(b)

    spec.declared = tuple(new_declared)

    # compute the bases
    bases = new_declared # guaranteed no dupes

    if spec.inherit is not None:
        for c in spec.inherit.__bases__:
            b = implementedBy(c)
            if b not in seen:
                seen.add(b)
                bases.append(b)

    spec.__bases__ = tuple(bases)


def _implements_advice(cls):
    interfaces, do_classImplements = cls.__dict__['__implements_advice_data__']
    del cls.__implements_advice_data__
    do_classImplements(cls, *interfaces)
    return cls


class implementer(object):
    """
    Declare the interfaces implemented by instances of a class.

    This function is called as a class decorator.

    The arguments are one or more interfaces or interface
    specifications (`~zope.interface.interfaces.IDeclaration`
    objects).

    The interfaces given (including the interfaces in the
    specifications) are added to any interfaces previously declared,
    unless the interface is already implemented.

    Previous declarations include declarations for base classes unless
    implementsOnly was used.

    This function is provided for convenience. It provides a more
    convenient way to call `classImplements`. For example::

        @implementer(I1)
        class C(object):
            pass

    is equivalent to calling::

        classImplements(C, I1)

    after the class has been created.

    .. seealso:: `classImplements`
       The change history provided there applies to this function too.
    """
    __slots__ = ('interfaces',)

    def __init__(self, *interfaces):
        self.interfaces = interfaces

    def __call__(self, ob):
        if isinstance(ob, DescriptorAwareMetaClasses):
            # This is the common branch for new-style (object) and
            # on Python 2 old-style classes.
            classImplements(ob, *self.interfaces)
            return ob

        spec_name = _implements_name(ob)
        spec = Implements.named(spec_name, *self.interfaces)
        try:
            ob.__implemented__ = spec
        except AttributeError:
            raise TypeError("Can't declare implements", ob)
        return ob

class implementer_only(object):
    """Declare the only interfaces implemented by instances of a class

      This function is called as a class decorator.

      The arguments are one or more interfaces or interface
      specifications (`~zope.interface.interfaces.IDeclaration` objects).

      Previous declarations including declarations for base classes
      are overridden.

      This function is provided for convenience. It provides a more
      convenient way to call `classImplementsOnly`. For example::

        @implementer_only(I1)
        class C(object): pass

      is equivalent to calling::

        classImplementsOnly(I1)

      after the class has been created.
      """

    def __init__(self, *interfaces):
        self.interfaces = interfaces

    def __call__(self, ob):
        if isinstance(ob, (FunctionType, MethodType)):
            # XXX Does this decorator make sense for anything but classes?
            # I don't think so. There can be no inheritance of interfaces
            # on a method or function....
            raise ValueError('The implementer_only decorator is not '
                             'supported for methods or functions.')

        # Assume it's a class:
        classImplementsOnly(ob, *self.interfaces)
        return ob

def _implements(name, interfaces, do_classImplements):
    # This entire approach is invalid under Py3K.  Don't even try to fix
    # the coverage for this block there. :(
    frame = sys._getframe(2) # pylint:disable=protected-access
    locals = frame.f_locals # pylint:disable=redefined-builtin

    # Try to make sure we were called from a class def. In 2.2.0 we can't
    # check for __module__ since it doesn't seem to be added to the locals
    # until later on.
    if locals is frame.f_globals or '__module__' not in locals:
        raise TypeError(name+" can be used only from a class definition.")

    if '__implements_advice_data__' in locals:
        raise TypeError(name+" can be used only once in a class definition.")

    locals['__implements_advice_data__'] = interfaces, do_classImplements
    addClassAdvisor(_implements_advice, depth=3)

def implements(*interfaces):
    """
    Declare interfaces implemented by instances of a class.

    .. deprecated:: 5.0
        This only works for Python 2. The `implementer` decorator
        is preferred for all versions.

    This function is called in a class definition.

    The arguments are one or more interfaces or interface
    specifications (`~zope.interface.interfaces.IDeclaration`
    objects).

    The interfaces given (including the interfaces in the
    specifications) are added to any interfaces previously declared.

    Previous declarations include declarations for base classes unless
    `implementsOnly` was used.

    This function is provided for convenience. It provides a more
    convenient way to call `classImplements`. For example::

        implements(I1)

    is equivalent to calling::

        classImplements(C, I1)

    after the class has been created.
    """
    # This entire approach is invalid under Py3K.  Don't even try to fix
    # the coverage for this block there. :(
    if PYTHON3:
        raise TypeError(_ADVICE_ERROR % 'implementer')
    _implements("implements", interfaces, classImplements)

def implementsOnly(*interfaces):
    """Declare the only interfaces implemented by instances of a class

      This function is called in a class definition.

      The arguments are one or more interfaces or interface
      specifications (`~zope.interface.interfaces.IDeclaration` objects).

      Previous declarations including declarations for base classes
      are overridden.

      This function is provided for convenience. It provides a more
      convenient way to call `classImplementsOnly`. For example::

        implementsOnly(I1)

      is equivalent to calling::

        classImplementsOnly(I1)

      after the class has been created.
    """
    # This entire approach is invalid under Py3K.  Don't even try to fix
    # the coverage for this block there. :(
    if PYTHON3:
        raise TypeError(_ADVICE_ERROR % 'implementer_only')
    _implements("implementsOnly", interfaces, classImplementsOnly)

##############################################################################
#
# Instance declarations

class Provides(Declaration):  # Really named ProvidesClass
    """Implement ``__provides__``, the instance-specific specification

    When an object is pickled, we pickle the interfaces that it implements.
    """

    def __init__(self, cls, *interfaces):
        self.__args = (cls, ) + interfaces
        self._cls = cls
        Declaration.__init__(self, *self._add_interfaces_to_cls(interfaces, cls))

    # Added to by ``moduleProvides``, et al
    _v_module_names = ()

    def __repr__(self):
        # The typical way to create instances of this
        # object is via calling ``directlyProvides(...)`` or ``alsoProvides()``,
        # but that's not the only way. Proxies, for example,
        # directly use the ``Provides(...)`` function (which is the
        # more generic method, and what we pickle as). We're after the most
        # readable, useful repr in the common case, so we use the most
        # common name.
        #
        # We also cooperate with ``moduleProvides`` to attempt to do the
        # right thing for that API. See it for details.
        function_name = 'directlyProvides'
        if self._cls is ModuleType and self._v_module_names:
            # See notes in ``moduleProvides``/``directlyProvides``
            providing_on_module = True
            interfaces = self.__args[1:]
        else:
            providing_on_module = False
            interfaces = (self._cls,) + self.__bases__
        ordered_names = self._argument_names_for_repr(interfaces)
        if providing_on_module:
            mod_names = self._v_module_names
            if len(mod_names) == 1:
                mod_names = "sys.modules[%r]" % mod_names[0]
            ordered_names = (
                '%s, ' % (mod_names,)
            ) + ordered_names
        return "%s(%s)" % (
            function_name,
            ordered_names,
        )

    def __reduce__(self):
        # This reduces to the Provides *function*, not
        # this class.
        return Provides, self.__args

    __module__ = 'zope.interface'

    def __get__(self, inst, cls):
        """Make sure that a class __provides__ doesn't leak to an instance
        """
        if inst is None and cls is self._cls:
            # We were accessed through a class, so we are the class'
            # provides spec. Just return this object, but only if we are
            # being called on the same class that we were defined for:
            return self

        raise AttributeError('__provides__')

ProvidesClass = Provides

# Registry of instance declarations
# This is a memory optimization to allow objects to share specifications.
InstanceDeclarations = weakref.WeakValueDictionary()

def Provides(*interfaces): # pylint:disable=function-redefined
    """Cache instance declarations

      Instance declarations are shared among instances that have the same
      declaration. The declarations are cached in a weak value dictionary.
    """
    spec = InstanceDeclarations.get(interfaces)
    if spec is None:
        spec = ProvidesClass(*interfaces)
        InstanceDeclarations[interfaces] = spec

    return spec

Provides.__safe_for_unpickling__ = True


def directlyProvides(object, *interfaces): # pylint:disable=redefined-builtin
    """Declare interfaces declared directly for an object

      The arguments after the object are one or more interfaces or interface
      specifications (`~zope.interface.interfaces.IDeclaration` objects).

      The interfaces given (including the interfaces in the specifications)
      replace interfaces previously declared for the object.
    """
    cls = getattr(object, '__class__', None)
    if cls is not None and getattr(cls, '__class__', None) is cls:
        # It's a meta class (well, at least it it could be an extension class)
        # Note that we can't get here from Py3k tests:  there is no normal
        # class which isn't descriptor aware.
        if not isinstance(object,
                          DescriptorAwareMetaClasses):
            raise TypeError("Attempt to make an interface declaration on a "
                            "non-descriptor-aware class")

    interfaces = _normalizeargs(interfaces)
    if cls is None:
        cls = type(object)

    issub = False
    for damc in DescriptorAwareMetaClasses:
        if issubclass(cls, damc):
            issub = True
            break
    if issub:
        # we have a class or type.  We'll use a special descriptor
        # that provides some extra caching
        object.__provides__ = ClassProvides(object, cls, *interfaces)
    else:
        provides = object.__provides__ = Provides(cls, *interfaces)
        # See notes in ``moduleProvides``.
        if issubclass(cls, ModuleType) and hasattr(object, '__name__'):
            provides._v_module_names += (object.__name__,)



def alsoProvides(object, *interfaces): # pylint:disable=redefined-builtin
    """Declare interfaces declared directly for an object

    The arguments after the object are one or more interfaces or interface
    specifications (`~zope.interface.interfaces.IDeclaration` objects).

    The interfaces given (including the interfaces in the specifications) are
    added to the interfaces previously declared for the object.
    """
    directlyProvides(object, directlyProvidedBy(object), *interfaces)


def noLongerProvides(object, interface): # pylint:disable=redefined-builtin
    """ Removes a directly provided interface from an object.
    """
    directlyProvides(object, directlyProvidedBy(object) - interface)
    if interface.providedBy(object):
        raise ValueError("Can only remove directly provided interfaces.")


@_use_c_impl
class ClassProvidesBase(SpecificationBase):

    __slots__ = (
        '_cls',
        '_implements',
    )

    def __get__(self, inst, cls):
        # member slots are set by subclass
        # pylint:disable=no-member
        if cls is self._cls:
            # We only work if called on the class we were defined for

            if inst is None:
                # We were accessed through a class, so we are the class'
                # provides spec. Just return this object as is:
                return self

            return self._implements

        raise AttributeError('__provides__')


class ClassProvides(Declaration, ClassProvidesBase):
    """Special descriptor for class ``__provides__``

    The descriptor caches the implementedBy info, so that
    we can get declarations for objects without instance-specific
    interfaces a bit quicker.
    """

    __slots__ = (
        '__args',
    )

    def __init__(self, cls, metacls, *interfaces):
        self._cls = cls
        self._implements = implementedBy(cls)
        self.__args = (cls, metacls, ) + interfaces
        Declaration.__init__(self, *self._add_interfaces_to_cls(interfaces, metacls))

    def __repr__(self):
        # There are two common ways to get instances of this object:
        # The most interesting way is calling ``@provider(..)`` as a decorator
        # of a class; this is the same as calling ``directlyProvides(cls, ...)``.
        #
        # The other way is by default: anything that invokes ``implementedBy(x)``
        # will wind up putting an instance in ``type(x).__provides__``; this includes
        # the ``@implementer(...)`` decorator. Those instances won't have any
        # interfaces.
        #
        # Thus, as our repr, we go with the ``directlyProvides()`` syntax.
        interfaces = (self._cls, ) + self.__args[2:]
        ordered_names = self._argument_names_for_repr(interfaces)
        return "directlyProvides(%s)" % (ordered_names,)

    def __reduce__(self):
        return self.__class__, self.__args

    # Copy base-class method for speed
    __get__ = ClassProvidesBase.__get__


def directlyProvidedBy(object): # pylint:disable=redefined-builtin
    """Return the interfaces directly provided by the given object

    The value returned is an `~zope.interface.interfaces.IDeclaration`.
    """
    provides = getattr(object, "__provides__", None)
    if (
            provides is None # no spec
            # We might have gotten the implements spec, as an
            # optimization. If so, it's like having only one base, that we
            # lop off to exclude class-supplied declarations:
            or isinstance(provides, Implements)
    ):
        return _empty

    # Strip off the class part of the spec:
    return Declaration(provides.__bases__[:-1])


def classProvides(*interfaces):
    """Declare interfaces provided directly by a class

      This function is called in a class definition.

      The arguments are one or more interfaces or interface specifications
      (`~zope.interface.interfaces.IDeclaration` objects).

      The given interfaces (including the interfaces in the specifications)
      are used to create the class's direct-object interface specification.
      An error will be raised if the module class has an direct interface
      specification. In other words, it is an error to call this function more
      than once in a class definition.

      Note that the given interfaces have nothing to do with the interfaces
      implemented by instances of the class.

      This function is provided for convenience. It provides a more convenient
      way to call `directlyProvides` for a class. For example::

        classProvides(I1)

      is equivalent to calling::

        directlyProvides(theclass, I1)

      after the class has been created.
    """
    # This entire approach is invalid under Py3K.  Don't even try to fix
    # the coverage for this block there. :(

    if PYTHON3:
        raise TypeError(_ADVICE_ERROR % 'provider')

    frame = sys._getframe(1) # pylint:disable=protected-access
    locals = frame.f_locals # pylint:disable=redefined-builtin

    # Try to make sure we were called from a class def
    if (locals is frame.f_globals) or ('__module__' not in locals):
        raise TypeError("classProvides can be used only from a "
                        "class definition.")

    if '__provides__' in locals:
        raise TypeError(
            "classProvides can only be used once in a class definition.")

    locals["__provides__"] = _normalizeargs(interfaces)

    addClassAdvisor(_classProvides_advice, depth=2)

def _classProvides_advice(cls):
    # This entire approach is invalid under Py3K.  Don't even try to fix
    # the coverage for this block there. :(
    interfaces = cls.__dict__['__provides__']
    del cls.__provides__
    directlyProvides(cls, *interfaces)
    return cls


class provider(object):
    """Class decorator version of classProvides"""

    def __init__(self, *interfaces):
        self.interfaces = interfaces

    def __call__(self, ob):
        directlyProvides(ob, *self.interfaces)
        return ob


def moduleProvides(*interfaces):
    """Declare interfaces provided by a module

    This function is used in a module definition.

    The arguments are one or more interfaces or interface specifications
    (`~zope.interface.interfaces.IDeclaration` objects).

    The given interfaces (including the interfaces in the specifications) are
    used to create the module's direct-object interface specification.  An
    error will be raised if the module already has an interface specification.
    In other words, it is an error to call this function more than once in a
    module definition.

    This function is provided for convenience. It provides a more convenient
    way to call directlyProvides. For example::

      moduleProvides(I1)

    is equivalent to::

      directlyProvides(sys.modules[__name__], I1)
    """
    frame = sys._getframe(1) # pylint:disable=protected-access
    locals = frame.f_locals # pylint:disable=redefined-builtin

    # Try to make sure we were called from a module body
    if (locals is not frame.f_globals) or ('__name__' not in locals):
        raise TypeError(
            "moduleProvides can only be used from a module definition.")

    if '__provides__' in locals:
        raise TypeError(
            "moduleProvides can only be used once in a module definition.")

    # Note: This is cached based on the key ``(ModuleType, *interfaces)``;
    # One consequence is that any module that provides the same interfaces
    # gets the same ``__repr__``, meaning that you can't tell what module
    # such a declaration came from. Adding the module name to ``_v_module_names``
    # attempts to correct for this; it works in some common situations, but fails
    # (1) after pickling (the data is lost) and (2) if declarations are
    # actually shared and (3) if the alternate spelling of ``directlyProvides()``
    # is used. Problem (3)  is fixed by cooperating with ``directlyProvides``
    # to maintain this information, and problem (2) is worked around by
    # printing all the names, but (1) is unsolvable without introducing
    # new classes or changing the stored data...but it doesn't actually matter,
    # because ``ModuleType`` can't be pickled!
    p = locals["__provides__"] = Provides(ModuleType,
                                          *_normalizeargs(interfaces))
    p._v_module_names += (locals['__name__'],)


##############################################################################
#
# Declaration querying support

# XXX:  is this a fossil?  Nobody calls it, no unit tests exercise it, no
#       doctests import it, and the package __init__ doesn't import it.
#       (Answer: Versions of zope.container prior to 4.4.0 called this,
#        and zope.proxy.decorator up through at least 4.3.5 called this.)
def ObjectSpecification(direct, cls):
    """Provide object specifications

    These combine information for the object and for it's classes.
    """
    return Provides(cls, direct) # pragma: no cover fossil

@_use_c_impl
def getObjectSpecification(ob):
    try:
        provides = ob.__provides__
    except AttributeError:
        provides = None

    if provides is not None:
        if isinstance(provides, SpecificationBase):
            return provides

    try:
        cls = ob.__class__
    except AttributeError:
        # We can't get the class, so just consider provides
        return _empty
    return implementedBy(cls)


@_use_c_impl
def providedBy(ob):
    """
    Return the interfaces provided by *ob*.

    If *ob* is a :class:`super` object, then only interfaces implemented
    by the remainder of the classes in the method resolution order are
    considered. Interfaces directly provided by the object underlying *ob*
    are not.
    """
    # Here we have either a special object, an old-style declaration
    # or a descriptor

    # Try to get __providedBy__
    try:
        if isinstance(ob, super): # Some objects raise errors on isinstance()
            return implementedBy(ob)

        r = ob.__providedBy__
    except AttributeError:
        # Not set yet. Fall back to lower-level thing that computes it
        return getObjectSpecification(ob)

    try:
        # We might have gotten a descriptor from an instance of a
        # class (like an ExtensionClass) that doesn't support
        # descriptors.  We'll make sure we got one by trying to get
        # the only attribute, which all specs have.
        r.extends
    except AttributeError:

        # The object's class doesn't understand descriptors.
        # Sigh. We need to get an object descriptor, but we have to be
        # careful.  We want to use the instance's __provides__, if
        # there is one, but only if it didn't come from the class.

        try:
            r = ob.__provides__
        except AttributeError:
            # No __provides__, so just fall back to implementedBy
            return implementedBy(ob.__class__)

        # We need to make sure we got the __provides__ from the
        # instance. We'll do this by making sure we don't get the same
        # thing from the class:

        try:
            cp = ob.__class__.__provides__
        except AttributeError:
            # The ob doesn't have a class or the class has no
            # provides, assume we're done:
            return r

        if r is cp:
            # Oops, we got the provides from the class. This means
            # the object doesn't have it's own. We should use implementedBy
            return implementedBy(ob.__class__)

    return r


@_use_c_impl
class ObjectSpecificationDescriptor(object):
    """Implement the ``__providedBy__`` attribute

    The ``__providedBy__`` attribute computes the interfaces provided by
    an object. If an object has an ``__provides__`` attribute, that is returned.
    Otherwise, `implementedBy` the *cls* is returned.

    .. versionchanged:: 5.4.0
       Both the default (C) implementation and the Python implementation
       now let exceptions raised by accessing ``__provides__`` propagate.
       Previously, the C version ignored all exceptions.
    .. versionchanged:: 5.4.0
       The Python implementation now matches the C implementation and lets
       a ``__provides__`` of ``None`` override what the class is declared to
       implement.
    """

    def __get__(self, inst, cls):
        """Get an object specification for an object
        """
        if inst is None:
            return getObjectSpecification(cls)

        try:
            return inst.__provides__
        except AttributeError:
            return implementedBy(cls)


##############################################################################

def _normalizeargs(sequence, output=None):
    """Normalize declaration arguments

    Normalization arguments might contain Declarions, tuples, or single
    interfaces.

    Anything but individual interfaces or implements specs will be expanded.
    """
    if output is None:
        output = []

    cls = sequence.__class__
    if InterfaceClass in cls.__mro__ or Implements in cls.__mro__:
        output.append(sequence)
    else:
        for v in sequence:
            _normalizeargs(v, output)

    return output

_empty = _ImmutableDeclaration()

objectSpecificationDescriptor = ObjectSpecificationDescriptor()