Interfaces

Interfaces are objects that specify (document) the external behavior of objects that “provide” them. An interface specifies behavior through:

• Informal documentation in a doc string
• Attribute definitions
• Invariants, which are conditions that must hold for objects that provide the interface

Attribute definitions specify specific attributes. They define the attribute name and provide documentation and constraints of attribute values. Attribute definitions can take a number of forms, as we’ll see below.

Defining interfaces

Interfaces are defined using Python class statements:

>>> import zope.interface
>>> class IFoo(zope.interface.Interface):
...    """Foo blah blah"""
...
...    x = zope.interface.Attribute("""X blah blah""")
...
...    def bar(q, r=None):
...        """bar blah blah"""

In the example above, we’ve created an interface, IFoo. We subclassed zope.interface.Interface, which is an ancestor interface for all interfaces, much as object is an ancestor of all new-style classes [1]. The interface is not a class, it’s an Interface, an instance of zope.interface.interface.InterfaceClass:

>>> type(IFoo)
<class 'zope.interface.interface.InterfaceClass'>

We can ask for the interface’s documentation:

>>> IFoo.__doc__
'Foo blah blah'

and its name:

>>> IFoo.__name__
'IFoo'

and even its module:

>>> IFoo.__module__
'builtins'

The interface defined two attributes:

x

This is the simplest form of attribute definition. It has a name and a doc string. It doesn’t formally specify anything else.

bar

This is a method. A method is defined via a function definition. A method is simply an attribute constrained to be a callable with a particular signature, as provided by the function definition.

Note that bar doesn’t take a self argument. Interfaces document how an object is used. When calling instance methods, you don’t pass a self argument, so a self argument isn’t included in the interface signature. The self argument in instance methods is really an implementation detail of Python instances. Other objects, besides instances can provide interfaces and their methods might not be instance methods. For example, modules can provide interfaces and their methods are usually just functions. Even instances can have methods that are not instance methods.

You can access the attributes defined by an interface using mapping syntax:

>>> x = IFoo['x']
>>> type(x)
<class 'zope.interface.interface.Attribute'>
>>> x.__name__
'x'
>>> x.__doc__
'X blah blah'

>>> IFoo.get('x').__name__
'x'

>>> IFoo.get('y')

You can use in to determine if an interface defines a name:

>>> 'x' in IFoo
True

You can iterate over interfaces to get the names they define:

>>> names = list(IFoo)
>>> names.sort()
>>> names
['bar', 'x']

Remember that interfaces aren’t classes. You can’t access attribute definitions as attributes of interfaces:

>>> IFoo.x
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
AttributeError: 'InterfaceClass' object has no attribute 'x'

Methods provide access to the method signature:

>>> bar = IFoo['bar']
>>> bar.getSignatureString()
'(q, r=None)'

TODO

Methods really should have a better API. This is something that needs to be improved.

Declaring interfaces

Having defined interfaces, we can declare that objects provide them. Before we describe the details, lets define some terms:

provide

We say that objects provide interfaces. If an object provides an interface, then the interface specifies the behavior of the object. In other words, interfaces specify the behavior of the objects that provide them.

implement

We normally say that classes implement interfaces. If a class implements an interface, then the instances of the class provide the interface. Objects provide interfaces that their classes implement [2]. (Objects can provide interfaces directly, in addition to what their classes implement.)

It is important to note that classes don’t usually provide the interfaces that they implement.

We can generalize this to factories. For any callable object we can declare that it produces objects that provide some interfaces by saying that the factory implements the interfaces.

Now that we’ve defined these terms, we can talk about the API for declaring interfaces.

Declaring implemented interfaces

The most common way to declare interfaces is using the implementer decorator on a class:

>>> @zope.interface.implementer(IFoo)
... class Foo:
...
...     def __init__(self, x=None):
...         self.x = x
...
...     def bar(self, q, r=None):
...         return q, r, self.x
...
...     def __repr__(self):
...         return "Foo(%s)" % self.x

In this example, we declared that Foo implements IFoo. This means that instances of Foo provide IFoo. Having made this declaration, there are several ways we can introspect the declarations. First, we can ask an interface whether it is implemented by a class:

>>> IFoo.implementedBy(Foo)
True

And we can ask whether an interface is provided by an object:

>>> foo = Foo()
>>> IFoo.providedBy(foo)
True

Of course, Foo doesn’t provide IFoo, it implements it:

>>> IFoo.providedBy(Foo)
False

We can also ask what interfaces are implemented by a class:

>>> list(zope.interface.implementedBy(Foo))
[<InterfaceClass builtins.IFoo>]

It’s an error to ask for interfaces implemented by a non-callable object:

>>> IFoo.implementedBy(foo)
Traceback (most recent call last):
...
TypeError: ('ImplementedBy called for non-factory', Foo(None))

>>> list(zope.interface.implementedBy(foo))
Traceback (most recent call last):
...
TypeError: ('ImplementedBy called for non-factory', Foo(None))

Similarly, we can ask what interfaces are provided by an object:

>>> list(zope.interface.providedBy(foo))
[<InterfaceClass builtins.IFoo>]
>>> list(zope.interface.providedBy(Foo))
[]

We can declare interfaces implemented by other factories (besides classes). We do this using the same implementer decorator.

>>> @zope.interface.implementer(IFoo)
... def yfoo(y):
...     foo = Foo()
...     foo.y = y
...     return foo

>>> list(zope.interface.implementedBy(yfoo))
[<InterfaceClass builtins.IFoo>]

Note that the implementer decorator may modify its argument. Callers should not assume that a new object is created.

Using implementer also works on callable objects. This is used by zope.formlib, as an example:

>>> class yfactory:
...     def __call__(self, y):
...         foo = Foo()
...         foo.y = y
...         return foo
>>> yfoo = yfactory()
>>> yfoo = zope.interface.implementer(IFoo)(yfoo)

>>> list(zope.interface.implementedBy(yfoo))
[<InterfaceClass builtins.IFoo>]

XXX: Double check and update these version numbers:

In zope.interface 3.5.2 and lower, the implementer decorator can not be used for classes, but in 3.6.0 and higher it can:

>>> Foo = zope.interface.implementer(IFoo)(Foo)
>>> list(zope.interface.providedBy(Foo()))
[<InterfaceClass builtins.IFoo>]

Note that class decorators using the @implementer(IFoo) syntax are only supported in Python 2.6 and later.

zope.interface.implementer(*interfaces)

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 (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.

See also

classImplements The change history provided there applies to this function too.

Declaring provided interfaces

We can declare interfaces directly provided by objects. Suppose that we want to document what the __init__ method of the Foo class does. It’s not really part of IFoo. You wouldn’t normally call the __init__ method on Foo instances. Rather, the __init__ method is part of Foo’s __call__ method:

>>> class IFooFactory(zope.interface.Interface):
...     """Create foos"""
...
...     def __call__(x=None):
...         """Create a foo
...
...         The argument provides the initial value for x ...
...         """

It’s the class that provides this interface, so we declare the interface on the class:

>>> zope.interface.directlyProvides(Foo, IFooFactory)

And then, we’ll see that Foo provides some interfaces:

>>> list(zope.interface.providedBy(Foo))
[<InterfaceClass builtins.IFooFactory>]
>>> IFooFactory.providedBy(Foo)
True

Declaring class interfaces is common enough that there’s a special decorator for it, provider:

>>> @zope.interface.implementer(IFoo)
... @zope.interface.provider(IFooFactory)
... class Foo2:
...
...     def __init__(self, x=None):
...         self.x = x
...
...     def bar(self, q, r=None):
...         return q, r, self.x
...
...     def __repr__(self):
...         return "Foo(%s)" % self.x

>>> list(zope.interface.providedBy(Foo2))
[<InterfaceClass builtins.IFooFactory>]
>>> IFooFactory.providedBy(Foo2)
True

There’s a similar function, moduleProvides, that supports interface declarations from within module definitions. For example, see the use of moduleProvides call in zope.interface.__init__, which declares that the package zope.interface provides IInterfaceDeclaration.

Sometimes, we want to declare interfaces on instances, even though those instances get interfaces from their classes. Suppose we create a new interface, ISpecial:

>>> class ISpecial(zope.interface.Interface):
...     reason = zope.interface.Attribute("Reason why we're special")
...     def brag():
...         "Brag about being special"

We can make an existing foo instance special by providing reason and brag attributes:

>>> foo.reason = 'I just am'
>>> def brag():
...      return "I'm special!"
>>> foo.brag = brag
>>> foo.reason
'I just am'
>>> foo.brag()
"I'm special!"

and by declaring the interface:

>>> zope.interface.directlyProvides(foo, ISpecial)

then the new interface is included in the provided interfaces:

>>> ISpecial.providedBy(foo)
True
>>> list(zope.interface.providedBy(foo))
[<InterfaceClass builtins.ISpecial>, <InterfaceClass builtins.IFoo>]

We can find out what interfaces are directly provided by an object:

>>> list(zope.interface.directlyProvidedBy(foo))
[<InterfaceClass builtins.ISpecial>]

>>> newfoo = Foo()
>>> list(zope.interface.directlyProvidedBy(newfoo))
[]

zope.interface.provider(*interfaces)

Class decorator version of classProvides

Inherited declarations

Normally, declarations are inherited:

>>> @zope.interface.implementer(ISpecial)
... class SpecialFoo(Foo):
...     reason = 'I just am'
...     def brag(self):
...         return "I'm special because %s" % self.reason

>>> list(zope.interface.implementedBy(SpecialFoo))
[<InterfaceClass builtins.ISpecial>, <InterfaceClass builtins.IFoo>]

>>> list(zope.interface.providedBy(SpecialFoo()))
[<InterfaceClass builtins.ISpecial>, <InterfaceClass builtins.IFoo>]

Sometimes, you don’t want to inherit declarations. In that case, you can use implementer_only, instead of implementer:

>>> @zope.interface.implementer_only(ISpecial)
... class Special(Foo):
...     reason = 'I just am'
...     def brag(self):
...         return "I'm special because %s" % self.reason

>>> list(zope.interface.implementedBy(Special))
[<InterfaceClass builtins.ISpecial>]

>>> list(zope.interface.providedBy(Special()))
[<InterfaceClass builtins.ISpecial>]

External declarations

Normally, we make implementation declarations as part of a class definition. Sometimes, we may want to make declarations from outside the class definition. For example, we might want to declare interfaces for classes that we didn’t write. The function classImplements can be used for this purpose:

>>> class C:
...     pass

>>> zope.interface.classImplements(C, IFoo)
>>> list(zope.interface.implementedBy(C))
[<InterfaceClass builtins.IFoo>]

zope.interface.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 (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.

Changed in version 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.

Changed in version 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).

We can use classImplementsOnly to exclude inherited interfaces:

>>> class C(Foo):
...     pass

>>> zope.interface.classImplementsOnly(C, ISpecial)
>>> list(zope.interface.implementedBy(C))
[<InterfaceClass builtins.ISpecial>]

zope.interface.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 (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.

Declaration Objects

When we declare interfaces, we create declaration objects. When we query declarations, declaration objects are returned:

>>> type(zope.interface.implementedBy(Special))
<class 'zope.interface.declarations.Implements'>

Declaration objects and interface objects are similar in many ways. In fact, they share a common base class. The important thing to realize about them is that they can be used where interfaces are expected in declarations. Here’s a silly example:

>>> @zope.interface.implementer_only(
...     zope.interface.implementedBy(Foo),
...     ISpecial,
... )
... class Special2(object):
...     reason = 'I just am'
...     def brag(self):
...         return "I'm special because %s" % self.reason

The declaration here is almost the same as zope.interface.implementer(ISpecial), except that the order of interfaces in the resulting declaration is different:

>>> list(zope.interface.implementedBy(Special2))
[<InterfaceClass builtins.IFoo>, <InterfaceClass builtins.ISpecial>]

Interface Inheritance

Interfaces can extend other interfaces. They do this simply by listing the other interfaces as base interfaces:

>>> class IBlat(zope.interface.Interface):
...     """Blat blah blah"""
...
...     y = zope.interface.Attribute("y blah blah")
...     def eek():
...         """eek blah blah"""

>>> IBlat.__bases__
(<InterfaceClass zope.interface.Interface>,)

>>> class IBaz(IFoo, IBlat):
...     """Baz blah"""
...     def eek(a=1):
...         """eek in baz blah"""
...

>>> IBaz.__bases__
(<InterfaceClass builtins.IFoo>, <InterfaceClass builtins.IBlat>)

>>> names = list(IBaz)
>>> names.sort()
>>> names
['bar', 'eek', 'x', 'y']

Note that IBaz overrides eek:

>>> IBlat['eek'].__doc__
'eek blah blah'
>>> IBaz['eek'].__doc__
'eek in baz blah'

We were careful to override eek in a compatible way. When extending an interface, the extending interface should be compatible [3] with the extended interfaces.

We can ask whether one interface extends another:

>>> IBaz.extends(IFoo)
True
>>> IBlat.extends(IFoo)
False

Note that interfaces don’t extend themselves:

>>> IBaz.extends(IBaz)
False

Sometimes we wish they did, but we can instead use isOrExtends:

>>> IBaz.isOrExtends(IBaz)
True
>>> IBaz.isOrExtends(IFoo)
True
>>> IFoo.isOrExtends(IBaz)
False

When we iterate over an interface, we get all of the names it defines, including names defined by base interfaces. Sometimes, we want just the names defined by the interface directly. We can use the names method for that:

>>> list(IBaz.names())
['eek']

Inheritance of attribute specifications

An interface may override attribute definitions from base interfaces. If two base interfaces define the same attribute, the attribute is inherited from the most specific interface. For example, with:

>>> class IBase(zope.interface.Interface):
...
...     def foo():
...         "base foo doc"

>>> class IBase1(IBase):
...     pass

>>> class IBase2(IBase):
...
...     def foo():
...         "base2 foo doc"

>>> class ISub(IBase1, IBase2):
...     pass

ISub’s definition of foo is the one from IBase2, since IBase2 is more specific than IBase:

>>> ISub['foo'].__doc__
'base2 foo doc'

Note that this differs from a depth-first search.

Sometimes, it’s useful to ask whether an interface defines an attribute directly. You can use the direct method to get a directly defined definitions:

>>> IBase.direct('foo').__doc__
'base foo doc'

>>> ISub.direct('foo')

Specifications

Interfaces and declarations are both special cases of specifications. What we described above for interface inheritance applies to both declarations and specifications. Declarations actually extend the interfaces that they declare:

>>> @zope.interface.implementer(IBaz)
... class Baz(object):
...     pass

>>> baz_implements = zope.interface.implementedBy(Baz)
>>> baz_implements.__bases__
(<InterfaceClass builtins.IBaz>, classImplements(object))

>>> baz_implements.extends(IFoo)
True

>>> baz_implements.isOrExtends(IFoo)
True
>>> baz_implements.isOrExtends(baz_implements)
True

Specifications (interfaces and declarations) provide an __sro__ that lists the specification and all of it’s ancestors:

>>> from pprint import pprint
>>> pprint(baz_implements.__sro__)
(classImplements(Baz, IBaz),
 <InterfaceClass builtins.IBaz>,
 <InterfaceClass builtins.IFoo>,
 <InterfaceClass builtins.IBlat>,
 classImplements(object),
 <InterfaceClass zope.interface.Interface>)
>>> class IBiz(zope.interface.Interface):
...    pass
>>> @zope.interface.implementer(IBiz)
... class Biz(Baz):
...    pass
>>> pprint(zope.interface.implementedBy(Biz).__sro__)
(classImplements(Biz, IBiz),
 <InterfaceClass builtins.IBiz>,
 classImplements(Baz, IBaz),
 <InterfaceClass builtins.IBaz>,
 <InterfaceClass builtins.IFoo>,
 <InterfaceClass builtins.IBlat>,
 classImplements(object),
 <InterfaceClass zope.interface.Interface>)

Tagged Values

zope.interface.taggedValue(key, value)

Attaches a tagged value to an interface at definition time.

Interfaces and attribute descriptions support an extension mechanism, borrowed from UML, called “tagged values” that lets us store extra data:

>>> IFoo.setTaggedValue('date-modified', '2004-04-01')
>>> IFoo.setTaggedValue('author', 'Jim Fulton')
>>> IFoo.getTaggedValue('date-modified')
'2004-04-01'
>>> IFoo.queryTaggedValue('date-modified')
'2004-04-01'
>>> IFoo.queryTaggedValue('datemodified')
>>> tags = list(IFoo.getTaggedValueTags())
>>> tags.sort()
>>> tags
['author', 'date-modified']

Function attributes are converted to tagged values when method attribute definitions are created:

>>> class IBazFactory(zope.interface.Interface):
...     def __call__():
...         "create one"
...     __call__.return_type = IBaz

>>> IBazFactory['__call__'].getTaggedValue('return_type')
<InterfaceClass builtins.IBaz>

Tagged values can also be defined from within an interface definition:

>>> class IWithTaggedValues(zope.interface.Interface):
...     zope.interface.taggedValue('squish', 'squash')
>>> IWithTaggedValues.getTaggedValue('squish')
'squash'

Tagged values are inherited in the same way that attribute and method descriptions are. Inheritance can be ignored by using the “direct” versions of functions.

>>> class IExtendsIWithTaggedValues(IWithTaggedValues):
...     zope.interface.taggedValue('child', True)
>>> IExtendsIWithTaggedValues.getTaggedValue('child')
True
>>> IExtendsIWithTaggedValues.getDirectTaggedValue('child')
True
>>> IExtendsIWithTaggedValues.getTaggedValue('squish')
'squash'
>>> print(IExtendsIWithTaggedValues.queryDirectTaggedValue('squish'))
None
>>> IExtendsIWithTaggedValues.setTaggedValue('squish', 'SQUASH')
>>> IExtendsIWithTaggedValues.getTaggedValue('squish')
'SQUASH'
>>> IExtendsIWithTaggedValues.getDirectTaggedValue('squish')
'SQUASH'

Invariants

zope.interface.invariant(call)

Interfaces can express conditions that must hold for objects that provide them. These conditions are expressed using one or more invariants. Invariants are callable objects that will be called with an object that provides an interface. An invariant raises an Invalid exception if the condition doesn’t hold. Here’s an example:

>>> class RangeError(zope.interface.Invalid):
...     """A range has invalid limits"""
...     def __repr__(self):
...         return "RangeError(%r)" % self.args

>>> def range_invariant(ob):
...     if ob.max < ob.min:
...         raise RangeError(ob)

Given this invariant, we can use it in an interface definition:

>>> class IRange(zope.interface.Interface):
...     min = zope.interface.Attribute("Lower bound")
...     max = zope.interface.Attribute("Upper bound")
...
...     zope.interface.invariant(range_invariant)

Interfaces have a method for checking their invariants:

>>> @zope.interface.implementer(IRange)
... class Range(object):
...     def __init__(self, min, max):
...         self.min, self.max = min, max
...
...     def __repr__(self):
...         return "Range(%s, %s)" % (self.min, self.max)

>>> IRange.validateInvariants(Range(1,2))
>>> IRange.validateInvariants(Range(1,1))
>>> IRange.validateInvariants(Range(2,1))
Traceback (most recent call last):
...
RangeError: Range(2, 1)

If you have multiple invariants, you may not want to stop checking after the first error. If you pass a list to validateInvariants, then a single Invalid exception will be raised with the list of exceptions as its argument:

>>> from zope.interface.exceptions import Invalid
>>> errors = []
>>> try:
...     IRange.validateInvariants(Range(2,1), errors)
... except Invalid as e:
...     str(e)
'[RangeError(Range(2, 1))]'

And the list will be filled with the individual exceptions:

>>> errors
[RangeError(Range(2, 1))]

>>> del errors[:]

Adaptation

Interfaces can be called to perform adaptation. Adaptation is the process of converting an object to an object implementing the interface. For example, in mathematics, to represent a point in space or on a graph there’s the familiar Cartesian coordinate system using CartesianPoint(x, y), and there’s also the Polar coordinate system using PolarPoint(r, theta), plus several others (homogeneous, log-polar, etc). Polar points are most convenient for some types of operations, but cartesian points may make more intuitive sense to most people. Before printing an arbitrary point, we might want to adapt it to ICartesianPoint, or before performing some mathematical operation you might want to adapt the arbitrary point to IPolarPoint.

The semantics are based on those of the PEP 246 adapt function.

If an object cannot be adapted, then a TypeError is raised:

>>> class ICartesianPoint(zope.interface.Interface):
...     x = zope.interface.Attribute("Distance from origin along x axis")
...     y = zope.interface.Attribute("Distance from origin along y axis")

>>> ICartesianPoint(0)
Traceback (most recent call last):
...
TypeError: ('Could not adapt', 0, <InterfaceClass builtins.ICartesianPoint>)

unless a default value is provided as a second positional argument; this value is not checked to see if it implements the interface:

>>> ICartesianPoint(0, 'bob')
'bob'

If an object already implements the interface, then it will be returned:

>>> @zope.interface.implementer(ICartesianPoint)
... class CartesianPoint(object):
...     """The default cartesian point is the origin."""
...     def __init__(self, x=0, y=0):
...         self.x = x
...         self.y = y
...     def __repr__(self):
...         return "CartesianPoint(%s, %s)" % (self.x, self.y)

>>> obj = CartesianPoint()
>>> ICartesianPoint(obj) is obj
True

__conform__

PEP 246 outlines a requirement:

When the object knows about the [interface], and either considers itself compliant, or knows how to wrap itself suitably.

This is handled with __conform__. If an object implements __conform__, then it will be used to give the object the chance to decide if it knows about the interface. This is true even if the class declares that it implements the interface.

>>> @zope.interface.implementer(ICartesianPoint)
... class C(object):
...     def __conform__(self, proto):
...          return "This could be anything."

>>> ICartesianPoint(C())
'This could be anything.'

If __conform__ returns None (because the object is unaware of the interface), then the rest of the adaptation process will continue. Here, we demonstrate that if the object already provides the interface, it is returned.

>>> @zope.interface.implementer(ICartesianPoint)
... class C(object):
...     def __conform__(self, proto):
...          return None

>>> c = C()
>>> ICartesianPoint(c) is c
True

Adapter hooks (see __adapt__ and adapter hooks) will also be used, if present (after a __conform__ method, if any, has been tried):

>>> from zope.interface.interface import adapter_hooks
>>> def adapt_tuple_to_point(iface, obj):
...     if isinstance(obj, tuple) and len(obj) == 2:
...         return CartesianPoint(*obj)

>>> adapter_hooks.append(adapt_tuple_to_point)
>>> ICartesianPoint((1, 1))
CartesianPoint(1, 1)

>>> adapter_hooks.remove(adapt_tuple_to_point)
>>> ICartesianPoint((1, 1))
Traceback (most recent call last):
...
TypeError: ('Could not adapt', (1, 1), <InterfaceClass builtins.ICartesianPoint>)

__adapt__ and adapter hooks

Interfaces implement the PEP 246 __adapt__ method to satisfy the requirement:

When the [interface] knows about the object, and either the object already complies or the [interface] knows how to suitably wrap the object.

This method is normally not called directly. It is called by the PEP 246 adapt framework and by the interface __call__ operator once __conform__ (if any) has failed.

The __adapt__ method is responsible for adapting an object to the receiver.

The default version returns None (because by default no interface “knows how to suitably wrap the object”):

>>> ICartesianPoint.__adapt__(0)

unless the object given provides the interface (“the object already complies”):

>>> @zope.interface.implementer(ICartesianPoint)
... class C(object):
...     pass

>>> obj = C()
>>> ICartesianPoint.__adapt__(obj) is obj
True

Customizing __adapt__ in an interface

It is possible to replace or customize the __adapt___ functionality for particular interfaces, if that interface “knows how to suitably wrap [an] object”. This method should return the adapted object if it knows how, or call the super class to continue with the default adaptation process.

>>> import math
>>> class IPolarPoint(zope.interface.Interface):
...     r = zope.interface.Attribute("Distance from center.")
...     theta = zope.interface.Attribute("Angle from horizontal.")
...     @zope.interface.interfacemethod
...     def __adapt__(self, obj):
...          if ICartesianPoint.providedBy(obj):
...              # Convert to polar coordinates.
...              r = math.sqrt(obj.x ** 2 + obj.y ** 2)
...              theta = math.acos(obj.x / r)
...              theta = math.degrees(theta)
...              return PolarPoint(r, theta)
...          return super(type(IPolarPoint), self).__adapt__(obj)

>>> @zope.interface.implementer(IPolarPoint)
... class PolarPoint(object):
...     def __init__(self, r=0, theta=0):
...        self.r = r; self.theta = theta
...     def __repr__(self):
...        return "PolarPoint(%s, %s)" % (self.r, self.theta)
>>> IPolarPoint(CartesianPoint(0, 1))
PolarPoint(1.0, 90.0)
>>> IPolarPoint(PolarPoint())
PolarPoint(0, 0)

See also

zope.interface.interfacemethod(), which explains how to override functions in interface definitions and why, prior to Python 3.6, the zero-argument version of super cannot be used.

Using adapter hooks for loose coupling

Commonly, the author of the interface doesn’t know how to wrap all possible objects, and neither does the author of an object know how to __conform__ to all possible interfaces. To support decoupling interfaces and objects, interfaces support the concept of “adapter hooks.” Adapter hooks are a global sequence of callables hook(interface, object) that are called, in order, from the default __adapt__ method until one returns a non-None result.

Note

In many applications, a Adapter Registry is installed as the first or only adapter hook.

We’ll install a hook that adapts from a 2D (x, y) Cartesian point on a plane to a three-dimensional point (x, y, z) by assuming the z coordinate is 0. First, we’ll define this new interface and an implementation:

>>> class ICartesianPoint3D(ICartesianPoint):
...      z = zope.interface.Attribute("Depth.")
>>> @zope.interface.implementer(ICartesianPoint3D)
... class CartesianPoint3D(CartesianPoint):
...     def __init__(self, x=0, y=0, z=0):
...        CartesianPoint.__init__(self, x, y)
...        self.z = 0
...     def __repr__(self):
...        return "CartesianPoint3D(%s, %s, %s)" % (self.x, self.y, self.z)

We install a hook by simply adding it to the adapter_hooks list:

>>> from zope.interface.interface import adapter_hooks
>>> def returns_none(iface, obj):
...     print("(First adapter hook returning None.)")
>>> def adapt_2d_to_3d(iface, obj):
...     if iface == ICartesianPoint3D and ICartesianPoint.providedBy(obj):
...         return CartesianPoint3D(obj.x, obj.y, 0)
>>> adapter_hooks.append(returns_none)
>>> adapter_hooks.append(adapt_2d_to_3d)
>>> ICartesianPoint3D.__adapt__(CartesianPoint())
(First adapter hook returning None.)
CartesianPoint3D(0, 0, 0)
>>> ICartesianPoint3D(CartesianPoint())
(First adapter hook returning None.)
CartesianPoint3D(0, 0, 0)

Hooks can be uninstalled by removing them from the list:

>>> adapter_hooks.remove(returns_none)
>>> adapter_hooks.remove(adapt_2d_to_3d)
>>> ICartesianPoint3D.__adapt__(CartesianPoint())

Persistence, Sorting, Equality and Hashing

Tip

For the practical implications of what’s discussed below, and some potential problems, see Equality, Hashing, and Comparisons.

Just like Python classes, interfaces are designed to inexpensively support persistence using Python’s standard pickle module. This means that one process can send a reference to an interface to another process in the form of a byte string, and that other process can load that byte string and get the object that is that interface. The processes may be separated in time (one after the other), in space (running on different machines) or even be parts of the same process communicating with itself.

We can demonstrate this. Observe how small the byte string needed to capture the reference is. Also note that since this is the same process, the identical object is found and returned:

>>> import sys
>>> import pickle
>>> class Foo(object):
...    pass
>>> sys.modules[__name__].Foo = Foo # XXX, see below
>>> pickled_byte_string = pickle.dumps(Foo, 0)
>>> len(pickled_byte_string)
21
>>> imported = pickle.loads(pickled_byte_string)
>>> imported == Foo
True
>>> imported is Foo
True
>>> class IFoo(zope.interface.Interface):
...     pass
>>> sys.modules[__name__].IFoo = IFoo # XXX, see below
>>> pickled_byte_string = pickle.dumps(IFoo, 0)
>>> len(pickled_byte_string)
22
>>> imported = pickle.loads(pickled_byte_string)
>>> imported is IFoo
True
>>> imported == IFoo
True

References to Global Objects

The eagle-eyed reader will have noticed the two funny lines like sys.modules[__name__].Foo = Foo. What’s that for? To understand, we must know a bit about how Python “pickles” (pickle.dump or pickle.dumps) classes or interfaces.

When Python pickles a class or an interface, it does so as a “global object” [6]. Global objects are expected to already exist (contrast this with pickling a string or an object instance, which creates a new object in the receiving process) with all their necessary state information (for classes and interfaces, the state information would be things like the list of methods and defined attributes) in the receiving process, so the pickled byte string needs only contain enough data to look up that existing object; this data is a reference. Not only does this minimize the amount of data required to persist such an object, it also facilitates changing the definition of the object over time: if a class or interface gains or loses methods or attributes, loading a previously pickled reference will use the current definition of the object.

The reference to a global object that’s stored in the byte string consists only of the object’s __name__ and __module__. Before a global object obj is pickled, Python makes sure that the object being pickled is the same one that can be found at getattr(sys.modules[obj.__module__], obj.__name__); if there is no such object, or it refers to a different object, pickling fails. The two funny lines make sure that holds, no matter how this example is run (using some doctest runners, it doesn’t hold by default, unlike it normally would).

We can show some examples of what happens when that condition doesn’t hold. First, what if we change the global object and try to pickle the old one?

>>> sys.modules[__name__].Foo = 42
>>> pickle.dumps(Foo)
Traceback (most recent call last):
...
_pickle.PicklingError: Can't pickle <class 'Foo'>: it's not the same object as builtins.Foo

A consequence of this is that only one object of the given name can be defined and pickled at a time. If we were to try to define a new Foo class (remembering that normally the sys.modules[__name__].Foo = line is automatic), we still cannot pickle the old one:

>>> orig_Foo = Foo
>>> class Foo(object):
...    pass
>>> sys.modules[__name__].Foo = Foo # XXX, usually automatic
>>> pickle.dumps(orig_Foo)
Traceback (most recent call last):
...
_pickle.PicklingError: Can't pickle <class 'Foo'>: it's not the same object as builtins.Foo

Or what if there simply is no global object?

>>> del sys.modules[__name__].Foo
>>> pickle.dumps(Foo)
Traceback (most recent call last):
...
_pickle.PicklingError: Can't pickle <class 'Foo'>: attribute lookup Foo on builtins failed

Interfaces and classes behave the same in all those ways.

What’s This Have To Do With Sorting, Equality and Hashing?

Another important design consideration for interfaces is that they should be sortable. This permits them to be used, for example, as keys in a (persistent) BTree. As such, they define a total ordering, meaning that any given interface can definitively said to be greater than, less than, or equal to, any other interface. This relationship must be stable and hold the same across any two processes.

An object becomes sortable by overriding the equality method __eq__ and at least one of the comparison methods (such as __lt__).

Classes, on the other hand, are not sortable [4]. Classes can only be tested for equality, and they implement this using object identity: class_a == class_b is equivalent to class_a is class_b.

In addition to being sortable, it’s important for interfaces to be hashable so they can be used as keys in dictionaries or members of sets. This is done by implementing the __hash__ method [5].

Classes are hashable, and they also implement this based on object identity, with the equivalent of id(class_a).

To be both hashable and sortable, the hash method and the equality and comparison methods must be consistent with each other. That is, they must all be based on the same principle.

Classes use the principle of identity to implement equality and hashing, but they don’t implement sorting because identity isn’t a stable sorting method (it is different in every process).

Interfaces need to be sortable. In order for all three of hashing, equality and sorting to be consistent, interfaces implement them using the same principle as persistence. Interfaces are treated like “global objects” and sort and hash using the same information a reference to them would: their __name__ and __module__.

In this way, hashing, equality and sorting are consistent with each other, and consistent with pickling:

>>> class IFoo(zope.interface.Interface):
...     pass
>>> sys.modules[__name__].IFoo = IFoo # XXX, usually automatic
>>> f1 = IFoo
>>> pickled_f1 = pickle.dumps(f1)
>>> class IFoo(zope.interface.Interface):
...     pass
>>> sys.modules[__name__].IFoo = IFoo # XXX, usually automatic
>>> IFoo == f1
True
>>> unpickled_f1 = pickle.loads(pickled_f1)
>>> unpickled_f1 == IFoo == f1
True

This isn’t quite the case for classes; note how f1 wasn’t equal to Foo before pickling, but the unpickled value is:

>>> class Foo(object):
...     pass
>>> sys.modules[__name__].Foo = Foo # XXX, usually automatic
>>> f1 = Foo
>>> pickled_f1 = pickle.dumps(Foo)
>>> class Foo(object):
...     pass
>>> sys.modules[__name__].Foo = Foo # XXX, usually automatic
>>> f1 == Foo
False
>>> unpickled_f1 = pickle.loads(pickled_f1)
>>> unpickled_f1 == Foo # Surprise!
True
>>> unpickled_f1 == f1
False

For more information, and some rare potential pitfalls, see Equality, Hashing, and Comparisons.

Footnotes

[1]

The main reason we subclass Interface is to cause the Python class statement to create an interface, rather than a class. It’s possible to create interfaces by calling a special interface class directly. Doing this, it’s possible (and, on rare occasions, useful) to create interfaces that don’t descend from Interface. Using this technique is beyond the scope of this document.

[2]	Classes are factories. They can be called to create their instances. We expect that we will eventually extend the concept of implementation to other kinds of factories, so that we can declare the interfaces provided by the objects created.

[3]

The goal is substitutability. An object that provides an extending interface should be substitutable for an object that provides the extended interface. In our example, an object that provides IBaz should be usable wherever an object that provides IBlat is expected. The interface implementation doesn’t enforce this, but maybe it should do some checks.

[4]	In Python 2, classes could be sorted, but the sort was not stable (it also used the identity principle) and not useful for persistence; this was considered a bug that was fixed in Python 3.

[5]	In order to be hashable, you must implement both __eq__ and __hash__. If you only implement __eq__, Python makes sure the type cannot be used in a dictionary, set, or with hash(). In Python 2, this wasn’t the case, and forgetting to override __hash__ was a constant source of bugs.

[6]	From the name of the pickle bytecode operator; it varies depending on the protocol but always includes “GLOBAL”.