A Python interface library that disallows function body content on interfaces and supports adaption.
Jump to the Reference.
- Features:
- Prevents code in method bodies of an interface class
- Ensures that method overrides have compatible signatures
- Allows concrete implementations the flexibility to implement abstract properties as instance attributes.
- Supports interface adaption.
- Treats abc interfaces that do not include any implementation as a pure interface type. This means that
class C(PureInterface, ABCInterface)
will be a pure interface if the abc interface meets the no function body content criteria. - Supports optional structural type checking for
Interface.provided_by(a)
andInterface.adapt(a)
- Warns if
provided_by
did a structural type check when inheritance would work. - Supports python 2.7 and 3.5+
The phrase pure interface applies only to the first design goal - a class that defines only an interface with no implementation is a pure interface. In every other respect the zen of 'practicality beats purity' applies.
pure_interface depends on the six and typing modules (typing is included in python 3.5 and later).
You can install released versions of pure_interface using pip:
pip install pure_interface
or you can grab the source code from GitHub.
For simplicity in these examples we assume that the entire pure_interface namespace has been imported :
from pure_interface import *
To define an interface, simply inherit from the class PureInterface
and leave all method bodies empty:
class IAnimal(PureInterface):
@property
def height(self):
pass
def speak(self, volume):
pass
As PureInterface
is a subtype of abc.ABC
the abstractmethod
and abstractproperty
decorators work as expected. For convenience the abc
module abstract decorators are included in the pure_interface
namespace, and on Python 2.7 abstractclassmethod
and abstractstaticmethod
are also available.
However these decorators are optional as ALL methods and properties on a pure interface are abstract. In the example above, both height
and speak
are considered abstract and must be overridden by subclasses. Because of this, interface classes cannot be instantiated :
IAnimal()
TypeError: Interfaces cannot be instantiated.
Including abstract decorators in your code can be useful for reminding yourself (and telling your IDE) that you need to override those methods. Another common way of informing an IDE that a method needs to be overridden is for the method to raise NotImplementedError
. For this reason methods that just raise NotImplementedError
are also considered empty.
Including code in a method will result in an InterfaceError
being raised when the module is imported. For example:
class BadInterface(PureInterface):
def method(self):
print('hello')
InterfaceError: Function "method" is not empty
Did you forget to inherit from object to make the class concrete?
Inspired by PEP-544 pure_interface
also allows using class attributes to specify required interface attributes.
The use of class attribute or @property
to define a class attribute are interchangable. This interface is equivalent to the one above:
class IAnimal(PureInterface):
height = None
def speak(self, volume):
pass
The value assigned to class attributes must be None
and the attribute is removed from the class dictionary.:
>>> IAnimal.height
AttributeError: 'IAnimal' object has no attribute 'height'
This is because IAnimal
is an interface definition and not an implementation. Of course, concrete implementations may use class attributes as normal.
In Python 3.6 and later type annotations can also be used to define interface properties:
class IAnimal(PureInterface):
height: float
def speak(self, volume):
pass
The dir()
function will include all interface attributes so that mock.Mock(spec=IAnimal)
will work as expected:
>>> dir(IAnimal)
['__abstractmethods__', '__doc__', ..., 'height', 'speak']
Simply inheriting from a pure interface and writing a concrete class will result in an InterfaceError
exception as pure_interface
will assume you are creating a sub-interface. To tell pure_interface
that a type should be concrete simply inherit from object
as well (or anything else that isn't a PureInterface
). For example:
class Animal(object, IAnimal):
def __init__(self, height):
self._height = height
@property
def height(self):
return self._height
def speak(self, volume):
print('hello')
Exception: Mixing a PureInterface
class with an abc.ABC
interface class that only defines abstract methods and properties that satisfy the empty method criteria will result in a type that is considered a pure interface.:
class ABCInterface(abc.ABC):
@abstractmethod
def foo(self):
pass
class MyPureInterface(ABCInterface):
def bar(self):
pass
Concrete implementations may implement interface properties as normal attributes, provided that they are all set in the constructor:
class Animal2(object, IAnimal):
def __init__(self, height):
self.height = height
def speak(self, volume):
print('hello')
This can simplify implementations greatly when there are lots of properties on an interface. You can also implement interface class attributes as properties if desired.
The astute reader will notice that the Animal2
bases list makes an inconsistent method resolution order. This is handled by the PureInterfaceType
meta-class by removing object
from the front of the bases list. However static checkers such as mypy will complain. To get around this, pure_interface
includes an empty Concrete
class which you can use to keep mypy happy:
class Concrete(object):
pass
class Animal2(Concrete, IAnimal):
def __init__(self, height):
self.height = height
def speak(self, volume):
print('hello')
Method overrides are checked for compatibility with the interface. This means that argument names must match exactly and that no new non-optional arguments are present in the override. This enforces that calling the method with interface parameters will aways work. For example, given the interface method:
def speak(self, volume):
Then these overrides will all fail the checks and raise an InterfaceError
:
def speak(self): # too few parameters
def speak(self, loudness): # name does not match
def speak(self, volume, language): # extra required argument
However new optional parameters are permitted, as are *args
and **kwargs
:
def speak(self, volume, language='doggy speak')
def speak(self, *args)
As with abc.ABC
, the abstract method checking for a class is done when an object is instantiated. However it is useful to know about missing methods sooner than that. For this reason pure_interface
will issue a warning during module import when methods are missing from a concrete subclass. For example:
class SilentAnimal(object, IAnimal):
def __init__(self, height):
self.height = height
will issue this warning:
readme.py:28: UserWarning: Incomplete Implementation: SilentAnimal does not implement speak
class SilentAnimal(object, IAnimal):
Trying to create a SilentAnimal
will fail in the standard abc way:
SilentAnimal()
TypeError: Can't instantiate abstract class SilentAnimal with abstract methods speak
If you have a mixin class that implements part of an interface you can suppress the warnings by adding an class attribute called pi_partial_implementation
. The value of the attribute is ignored, and the attribute itself is removed from the class. For example:
class HeightMixin(object, IAnimal):
pi_partial_implementation = True
def __init__(self, height):
self.height = height
will not issue any warnings.
The warning messages are also appended to the module variable missing_method_warnings
, irrespective of any warning filters (but only if is_development=True
). This provides an alternative to raising warnings as errors. When all your imports are complete you can check if this list is empty.:
if pure_iterface.missing_method_warnings:
for warning in pure_iterface.missing_method_warnings:
print(warning)
exit(1)
Note that missing properties are NOT checked for as they may be provided by instance attributes.
Adapters for an interface are registered with the adapts
decorator or with the register_adapter
function. Take for example an interface ISpeaker
and a class Talker
and an adapter class TalkerToSpeaker
:
class ISpeaker(PureInterface):
def speak(self, volume):
pass
class Talker(object):
def talk(self):
return 'talk'
@adapts(Talker)
class TalkerToSpeaker(object, ISpeaker):
def __init__(self, talker):
self._talker = talker
def speak(self, volume):
return self._talker.talk()
The adapts
decorator call above is equivalent to:
register_adapter(TalkerToSpeaker, Talker, ISpeaker)
The ISpeaker
parameter passed to register_adapter
is the first interface in the MRO of the class being decorated (TalkerToSpeaker
). If there are no interface types in the MRO of the decorated class an InterfaceError
exception is raised.
Adapter factory functions can be decorated too, in which case the interface being adapted to needs to be specified:
@adapts(Talker, ISpeaker)
def talker_to_speaker(talker):
return TalkerToSpeaker(talker)
The decorated adapter (whether class for function) must be callable with a single parameter - the object to adapt.
The PureInterface.adapt
method will adapt an object to the given interface such that Interface.provided_by
is True
or raise ValueError
if no adapter could be found. For example:
speaker = ISpeaker.adapt(talker)
isinstance(speaker, ISpeaker) --> True
If you want to get None
rather than an exception then use:
speaker = ISpeaker.adapt_or_none(talker)
You can filter a list of objects returning those objects that provide an interface using filter_adapt(objects)
:
list(ISpeaker.filter_adapt([None, Talker(), a_speaker, 'text']) --> [TalkerToSpeaker, a_speaker]
By default the adaption functions will return an object which provides only the functions and properties specified by the interface. For example given the following implementation of the ISpeaker
interface above:
class TopicSpeaker(ISpeaker):
def __init__(self, topic):
self.topic = topic
def speak(self, volume):
return 'lets talk about {} very {}'.format(self.topic, volume)
topic_speaker = TopicSpeaker('python')
Then:
speaker = ISpeaker.adapt(topic_speaker)
speaker is topic_speaker --> False
speaker.topic --> AttributeError("ISpeaker interface has no attribute topic")
This is controlled by the optional interface_only
parameter to adapt
which defaults to True
. Pass interface_only=False
if you want the actual adapted object rather than a wrapper:
speaker = ISpeaker.adapt(topic_speaker, interface_only=False)
speaker is topic_speaker --> True
speaker.topic --> 'Python'
Accessing the topic
attribute on an ISpeaker
may work for all current implementations of ISpeaker
, but this code will likely break at some inconvenient time in the future.
Adapters from sub-interfaces may be used to perform adaption if necessary. For example:
class IA(PureInterface):
foo = None
class IB(IA):
bar = None
@adapts(int):
class IntToB(object, IB):
def __init__(self, x):
self.foo = self.bar = x
Then IA.adapt(4)
will use the IntToB
adapter to adapt 4
to IA
(unless there is already an adapter from int
to IA
)
Structural type checking checks if an object has the attributes and methods defined by the interface.
As interfaces are inherited, you can usually use isinstance(obj, MyInterface)
to check if an interface is provided. An alternative to isinstance()
is the PureInterface.provided_by(obj)
classmethod which will fall back to structural type checking if the instance is not an actual subclass. This can be controlled by the allow_implicit
parameter which defaults to True
. The structural type-checking does not check function signatures.:
class Parrot(object):
def __init__(self):
self._height = 43
@property
def height(self):
return self._height
def speak(self, volume):
print('hello')
p = Parrot()
isinstance(p, IAnimal) --> False
IAnimal.provided_by(p) --> True
IAnimal.provided_by(p, allow_implicit=False) --> False
The structural type checking makes working with data transfer objects (DTO's) much easier.:
class IMyDataType(PureInterface):
@property
def thing(self):
pass
class DTO(object):
pass
d = DTO()
d.thing = 'hello'
IMyDataType.provided_by(d) --> True
e = DTO()
e.something_else = True
IMyDataType.provided_by(e) --> False
Adaption also supports structural typing by passing allow_implicit=True
(but this is not the default):
speaker = ISpeaker.adapt(Parrot(), allow_implicit=True)
ISpeaker.provided_by(speaker) --> True
When using provided_by()
or adapt()
with allow_implicit=True
, a warning may be issued informing you that the structurally typed object should inherit the interface. The warning is only issued if the interface is implemented by the class (and not by instance attributes as in the DTO case above) and the warning is only issued once for each class, interface pair. For example:
s = ISpeaker.adapt(Parrot())
UserWarning: Class Parrot implements ISpeaker.
Consider inheriting ISpeaker or using ISpeaker.register(Parrot)
The pure_interface
module provides 4 functions for returning information about interface types.
- type_is_pure_interface(cls)
Return True if cls is a pure interface, False otherwise or if cls is not a class.
- get_type_interfaces(cls)
Returns all interfaces in the cls mro including cls itself if it is an interface
- get_interface_method_names(interface)
Returns a frozen set of names of methods defined by the interface. If
type_is_pure_interface(interface)
returnsFalse
then an empty set is returned.- get_interface_property_names(interface)
Returns a frozen set of names of properties defined by the interface. If
type_is_pure_interface(interface)
returnsFalse
then an empty set is returned.
Much of the empty function and other checking is awesome whilst writing your code but ultimately slows down production code. For this reason the pure_interface
module has an is_development
switch.:
is_development = not hasattr(sys, 'frozen')
is_development
defaults to True
if running from source and default to False
if bundled into an executable by py2exe, cx_Freeze or similar tools.
If you manually change this flag it must be set before modules using the PureInterface
type are imported or else the change will not have any effect.
If is_development
if False
then:
- Signatures of overriding methods are not checked
- No warnings are issued by the adaption functions
- No incomplete implementation warnings are issued
- The default value of
interface_only
is set toFalse
, so that interface wrappers are not created.
You can use pure_interface
with PyContracts
Simply import the pure_contracts
module and use the ContractInterface
class defined there as you would the PureInterface
class described above. For example:
from pure_contracts import ContractInterface
from contracts import contract
class ISpeaker(ContractInterface):
@contract(volume=int, returns=unicode)
def speak(self, volume):
pass
- PureInterfaceType
Metaclass for checking interface and implementation classes. Adding PureInterfaceType as a meta-class to a class will not make that class an interface, you need to inherit from
PureInterface
class to define an interface.Classes created with a metaclass of
PureInterfaceType
will have the following property:_pi Information about the class that is used by this meta-class
- PureInterface
Base class for defining interfaces. The following methods are provided:
- adapt (obj, allow_implicit=False, interface_only=None)
Adapts
obj
to this interface. Ifallow_implicit
isTrue
permit structural adaptions. Ifinterface_only
isNone
the it is set to the value ofis_development
. Ifinterface_only
resolves toTrue
a wrapper object that provides the properties and methods defined by the interface and nothing else is returned. RaisesValueError
if no adaption is possible or a registered adapter returns an object not providing this interface.- adapt_or_none (obj, allow_implicit=False, interface_only=None)
As per adapt() except returns
None
instead of raising aValueError
- can_adapt (obj, allow_implicit=False)
Returns
True
ifadapt(obj, allow_implicit)
will succeed. Short-cut foradapt_or_none(obj) is not None
- filter_adapt (objects, allow_implicit=False, interface_only=None)
Generates adaptions of each item in objects that provide this interface. allow_implicit and interface_only are as for adapt. Objects that cannot be adapted to this interface are silently skipped.
- interface_only (implementation)
Returns a wrapper around implementation that provides the properties and methods defined by the interface and nothing else.
- provided_by (obj, allow_implicit=True)
Returns
True
if obj provides this interface. Ifallow_implicit
isTrue
the also returnTrue
for objects that provide the interface structure but do not inherit from it. RaisesValueError
is the class is a concrete type.
- Concrete
Empty class to create a consistent MRO in implementation classes.
- adapts (from_type, to_interface=None)
Class or function decorator for declaring an adapter from from_type to to_interface. The class or function being decorated must take a single argument (an instance of from_type) and provide (or return and object providing) to_interface. The adapter may return an object that provides the interface structurally only, however
adapt
must be called withallow_implicit=True
for this to work. If decorating a class, to_interface may beNone
to use the first interface in the class's MRO.- register_adapter (adapter, from_type, to_interface)
Registers an adapter to convert instances of from_type to objects that provide to_interface for the to_interface.adapt() method. adapter must be a callable that takes a single argument (an instance of from_type) and returns and object providing to_interface.
- type_is_pure_interface (cls)
Return
True
if cls is a pure interface andFalse
otherwise- get_type_interfaces (cls)
Returns all interfaces in the cls mro including cls itself if it is an interface
- get_interface_method_names (cls)
Returns a
frozenset
of names of methods defined by the interface. If cls is not a interface type then an empty set is returned.- get_interface_property_names (cls)
Returns a
frozenset
of names of properties defined by the interface If cls is not a interface type then an empty set is returned.- get_interface_attribute_names (cls)
Returns a
frozenset
of names of class attributes and annotations defined by the interface If cls is not a interface type then an empty set is returned.- get_interface_properties_and_attribute_names (cls)
Returns a
frozenset
of names of properties, attributes and annotations defined by the interface If cls is not a interface type then an empty set is returned.
- is_development
Set to
True
to enable all checks and warnings. If set toFalse
then:- Signatures of overriding methods are not checked
- No warnings are issued by the adaption functions
- No incomplete implementation warnings are issued
- The default value of
interface_only
is set toFalse
, so that interface wrappers are not created.
- missing_method_warnings
The list of warning messages for concrete classes with missing interface (abstract) method overrides. Note that missing properties are NOT checked for as they may be provided by instance attributes.