pyecore 0.1.3

Dynamic Metamodels

Dynamic metamodels reflects the ability to create metamodels “on-the-fly”. You can create metaclass hierarchie, add EAttribute and EReference.

In order to create a new metaclass, you need to create an EClass instance:

>>> import pyecore.ecore as Ecore
>>> MyMetaclass = Ecore.EClass('MyMetaclass')

You can then create instances of your metaclass:

>>> instance1 = MyMetaclass()
>>> instance2 = MyMetaclass()
>>> assert instance1 is not instance2

>From the created instances, we can go back to the metaclasses:

>>> instance1.eClass
<EClass name="MyMetaclass">

Then, we can add metaproperties to the freshly created metaclass:

>>> instance1.eClass.eAttributes
[]
>>> MyMetaclass.eStructuralFeatures.append(Ecore.EAttribute('name', Ecore.EString))
>>> instance1.eClass.eStructuralFeatures
[<pyecore.ecore.EAttribute object at 0x7f7da72ba940>]
>>> str(instance1.name)
'None'
>>> instance1.name = 'mystuff'
>>> instance1.name
'mystuff'

We can also create a new metaclass B and a new metareferences towards B:

>>> B = Ecore.EClass('B')
>>> MyMetaclass.eStructuralFeatures.append(Ecore.EReference('toB', B, containment=True))
>>> b1 = B()
>>> instance1.toB = b1
>>> instance1.toB
<pyecore.ecore.B object at 0x7f7da70531d0>
>>> b1.eContainer() is instance1   # because 'toB' is a containment reference
True

Opposite and ‘collection’ meta-references are also managed:

>>> C = Ecore.EClass('C')
>>> C.eStructuralFeatures.append(Ecore.EReference('toMy', MyMetaclass))
>>> MyMetaclass.eStructuralFeatures.append(Ecore.EReference('toCs', C, upper=-1, eOpposite=C.eStructuralFeatures[0]))
>>> instance1.toCs
[]
>>> c1 = C()
>>> c1.toMy = instance1
>>> instance1.toCs  # 'toCs' should contain 'c1' because 'toMy' is opposite relation of 'toCs'
[<pyecore.ecore.C object at 0x7f7da7053390>]

Static Metamodels

The static definition of a metamodel using PyEcore mostly relies on the classical classes definitions in Python. The following example is more related to a ‘by hand’ static metamodel definition. This way of producing metamodels is kinda deprecated as a MTL generator (in /generator) automatically produces a static metamodel from the .ecore definition.

$ cat example.py
"""
static metamodel example
"""
from pyecore.ecore import EObject, EAttribute, EReference, EString, MetaEClass

nsURI = 'http://example/1.0'


class B(EObject, metaclass=MetaEClass):
    def __init__(self):
        pass


class C(EObject, metaclass=MetaEClass):
    def __init__(self):
        pass


class MyMetaclass(EObject, metaclass=MetaEClass):
    name = EAttribute(eType=EString)
    toB = EReference(eType=B, containment=True)
    toCs = EReference(eType=C, upper=-1)

    def __init__(self):
        pass

# We need to update C in order to add the opposite meta-reference
# At the moment, the information need to be added in two places
C.toMy = EReference('toMy', MyMetaclass, eOpposite=MyMetaclass.toCs)
C.eClass.eStructuralFeatures.append(C.toMy)

$ python
...
>>> import example
>>> instance1 = example.MyMetaclass()
>>> c1 = C()
>>> c1.toMy = instance1
>>> assert c1 is instance1.toCs[0] and c1.toMy is instance1

The automatic code generator defines a Python package hierarchie instead of only a Python module. This allows more freedom for dedicated operations and references between packages.

Static/Dynamic EOperation

PyEcore also support EOperation definition for static and dynamic metamodel. For static metamodel, the solution is simple, a simple method with the code is added inside the defined class. The corresponding EOperation is created on the fly. Theire is still some “requirements” for this. In order to be understood as an EOperation candidate, the defined method must have at least one parameter and the first parameter must always be named self.

For dynamic metamodels, the simple fact of adding an EOperation instance in the EClass instance, adds an “empty” implementation:

>>> import pyecore.ecore as Ecore
>>> A = Ecore.EClass('A')
>>> operation = Ecore.EOperation('myoperation')
>>> param1 = Ecore.EParameter('param1', eType=Ecore.EString, required=True)
>>> operation.eParameters.append(param1)
>>> A.eOperations.append(operation)
>>> a = A()
>>> help(a.myoperation)
Help on method myoperation:

myoperation(param1) method of pyecore.ecore.A instance
>>> a.myoperation('test')
...
NotImplementedError: Method myoperation(param1) is not yet implemented

For each EParameter, the required parameter express the fact that the parameter is required or not in the produced operation:

>>> operation2 = Ecore.EOperation('myoperation2')
>>> p1 = Ecore.EParameter('p1', eType=Ecore.EString)
>>> operation2.eParameters.append(p1)
>>> A.eOperations.append(operation2)
>>> a = A()
>>> a.operation2(p1='test')  # Will raise a NotImplementedError exception

You can then create an implementation for the eoperation and link it to the EClass:

>>> def myoperation(self, param1):
...:    print(self, param1)
...:
>>> A.python_class.myoperation = myoperation

To be able to propose a dynamic empty implementation of the operation, PyEcore relies on Python code generation at runtime.

Notifications

PyEcore gives you the ability to listen to modifications performed on an element. The EObserver class provides a basic observer which can receive notifications from the EObject it is register in:

>>> import library as lib  # we use the wikipedia library example
>>> from pyecore.notification import EObserver, Kind
>>> smith = lib.Writer()
>>> b1 = lib.Book()
>>> observer = EObserver(smith, notifyChanged=lambda x: print(x))
>>> b1.authors.append(smith)  # observer receive the notification from smith because 'authors' is eOpposite or 'books'

The EObserver notification method can be set using a lambda as in the previous example, using a regular function or by class inheritance:

>>> def print_notif(notification):
...:    print(notification)
...:
>>> observer = EObserver()
>>> observer.observe(b1)
>>> observer.notifyChanged = print_notif
>>> b1.authors.append(smith)  # observer receive the notification from b1

Using inheritance:

>>> class PrintNotification(EObserver):
...:    def __init__(self, notifier=None):
...:        super().__init__(notifier=notifier)
...:
...:    def notifyChanged(self, notification):
...:        print(notification)
...:
...:
>>> observer = PrintNotification(b1)
>>> b1.authors.append(smith)  # observer receive the notification from b1

The Notification object contains information about the performed modification:

• new -> the new added value (can be a collection) or None is remove or unset

• old -> the replaced value (always None for collections)

• feature -> the EStructuralFeature modified

• notifer -> the object that have been modified

• kind -> the kind of modification performed

The different kind of notifications that can be currently received are:

• ADD -> when an object is added to a collection

• ADD_MANY -> when many objects are added to a collection

• REMOVE -> when an object is removed from a collection

• SET -> when a value is set in an attribute/reference

• UNSET -> when a value is removed from an attribute/reference

Deep Journey Inside PyEcore

This section will provide some explanation of how PyEcore works.

>>> A = EClass('A')
>>> eobject = A()  # We create an A instance
>>> eobject.eClass
<EClass name="A">
>>> eobject2 = eobject.eClass()  # We create another A instance
>>> assert isinstance(eobject2, eobject.__class__)
>>> from pyecore.ecore import EcoreUtils
>>> assert EcoreUtils.isinstance(eobject2, A)

>>> MyClass = EClass('MyClass')  # We create an EClass instance
>>> type(MyClass)
pyecore.ecore.EClass
>>> MyClass.python_class
pyecore.ecore.MyClass
>>> myclass_instance = MyClass()  # MyClass is callable, creates an instance of the 'python_class' class
>>> myclass_instance
<pyecore.ecore.MyClass at 0x7f64b697df98>
>>> type(myclass_instance)
pyecore.ecore.MyClass
# We can access the EClass instance from the created instance and go back
>>> myclass_instance.eClass
<EClass name="MyClass">
>>> assert myclass_instance.eClass.python_class is MyClass.python_class
>>> assert myclass_instance.eClass.python_class.eClass is MyClass
>>> assert myclass_instance.__class__ is MyClass.python_class
>>> assert myclass_instance.__class__.eClass is MyClass
>>> assert myclass_instance.__class__.eClass is myclass_instance.eClass

>>> B = EClass('B')  # in complement, we create a new B metaclass
>>> list(B.eAllSuperTypes())
[]
>>> B.eSuperTypes.append(A)  # B inherits from A
>>> list(B.eAllSuperTypes())
{<EClass name="A">}
>>> B.python_class.mro()
[pyecore.ecore.B,
 pyecore.ecore.A,
 pyecore.ecore.EObject,
 pyecore.ecore.ENotifier,
 object]
>>> b_instance = B()
>>> assert isinstance(b_instance, A.python_class)
>>> assert EcoreUtils.isinstance(b_instance, A)

Importing an Existing XMI Metamodel/Model

XMI support is still a work in progress, but the XMI import is on good tracks. Currently, only basic XMI metamodel (.ecore) and model instances can be loaded:

>>> from pyecore.resources import ResourceSet, URI
>>> rset = ResourceSet()
>>> resource = rset.get_resource(URI('path/to/mm.ecore'))
>>> mm_root = resource.contents[0]
>>> rset.metamodel_registry[mm_root.nsURI] = mm_root
>>> # At this point, the .ecore is loaded in the 'rset' as a metamodel
>>> resource = rset.get_resource(URI('path/to/instance.xmi'))
>>> model_root = resource.contents[0]
>>> # At this point, the model instance is loaded!

The ResourceSet/Resource/URI will evolve in the future. At the moment, only basic operations are enabled: create_resource/get_resource/load/save....

int_conversion = lambda x: int(x)  # translating str to int durint load()
String = Ecore.EDataType('String', str)
Double = Ecore.EDataType('Double', int, 0, from_string=int_conversion)
Int = Ecore.EDataType('Int', int, from_string=int_conversion)
IntObject = Ecore.EDataType('IntObject', int, None,
                            from_string=int_conversion)
Boolean = Ecore.EDataType('Boolean', bool, False,
                          from_string=lambda x: x in ['True', 'true'])
Long = Ecore.EDataType('Long', int, 0, from_string=int_conversion)
EJavaObject = Ecore.EDataType('EJavaObject', object)
xmltype = Ecore.EPackage()
xmltype.eClassifiers.extend([String,
                             Double,
                             Int,
                             EJavaObject,
                             Long,
                             Boolean,
                             IntObject])
xmltype.nsURI = 'http://www.eclipse.org/emf/2003/XMLType'
xmltype.nsPrefix = 'xmltype'
xmltype.name = 'xmltype'
rset.metamodel_registry[xmltype.nsURI] = xmltype

# Then the resource can be loaded (here from an http address)
resource = rset.get_resource(HttpURI('http://myadress.ecore'))
root = resource.contents[0]

rset.get_resource(URI('uri/towards/my/first/resource'))
resource = rset.get_resource(URI('uri/towards/my/secon/resource'))

# Other model is 'external_model'
resource = rset.create_resource(URI('the/wanted/uri'))
resource.append(external_model)

Exporting an Existing XMI Resource

As for the XMI import, the XMI export (serialization) is still somehow very basic. Here is an example of how you could save your objects in a file:

>>> # we suppose we have an already existing model in 'root'
>>> from pyecore.resources.xmi import XMIResource
>>> from pyecore.resources import URI
>>> resource = XMIResource(URI('my/path.xmi'))
>>> resource.append(root)  # We add the root to the resource
>>> resource.save()  # will save the result in 'my/path.xmi'
>>> resource.save(output=URI('test/path.xmi'))  # save the result in 'test/path.xmi'

You can also use a ResourceSet to deal with this:

>>> # we suppose we have an already existing model in 'root'
>>> from pyecore.resources import ResourceSet, URI
>>> rset = ResourceSet()
>>> resource = rset.create_resource(URI('my/path.xmi'))
>>> resource.append(root)
>>> resource.save()

Installation

PyEcore is available on pypi, you can simply install it using pip:

$ pip install pyecore

The installation can also be performed manually (better in a virtualenv):

$ python setup.py install

Dependencies

The dependencies required by pyecore are:

• ordered-set which is used for the ordered and unique collections expressed in the metamodel,

• lxml which is used for the XMI parsing.

Run the Tests

Tests uses py.test and ‘coverage’. Everything is driven by Tox, so in order to run the tests simply run:

$ tox

Liberty Regarding the Java EMF Implementation

• There is some meta-property that are not still coded inside PyEcore. More will come with time,

• Resource can only contain a single root at the moment,

• External resources (like http://www.eclipse.org/emf/2003/XMLType) must be create by hand an loaded in the global_registry or as a resource of a ResourceSet,

• Proxies are not “removed” once resolved as in the the Java version, instead they acts as transparent proxies and redirect each calls to the ‘proxied’ object.

State

In the current state, the project implements:

• the dynamic/static metamodel definitions,

• reflexive API,

• inheritance,

• enumerations,

• abstract metaclasses,

• runtime typechecking,

• attribute/reference creations,

• collections (attribute/references with upper bound set to -1),

• reference eopposite,

• containment reference,

• introspection,

• select/reject on collections,

• Eclipse XMI import (partially),

• Eclipse XMI export (partially),

• simple notification/Event system,

• EOperations support,

• code generator for the static part,

• EMF proxies.

The things that are in the roadmap:

• object deletion,

• documentation,

• command system (?).

Existing Projects

There is not so much projects proposing to handle model and metamodel in Python. The only projects I found are:

• PyEMOF (http://www.lifl.fr/~marvie/software/pyemof.html)

• EMF4CPP (https://github.com/catedrasaes-umu/emf4cpp)

• PyEMOFUC (http://www.istr.unican.es/pyemofuc/index_En.html)

PyEMOF proposes an implementation of the OMG’s EMOF in Python. The project targets Python2 and supports XMI import/export. The project didn’t move since 2005, but seems quite complete.

EMF4CPP proposes a C++ implementation of EMF. This implementation also introduces Python scripts to call the generated C++ code from a Python environment.

PyEMOFUC proposes, like PyEMOF, a pure Python implementation of the OMG’s EMOF. If we stick to a kind of EMF terminology, PyEMOFUC only supports dynamic metamodels. The project does not seems to move a lot.

Additional Resources

• The article at this address: http://modeling-languages.com/pyecore-python-eclipse-modeling-framework gives more information and implementations details about PyEcore.