PyEcore is a Model Driven Engineering (MDE) framework written for Python. Precisely, it is an implementation of EMF/Ecore for Python, and it tries to give an API which is compatible with the original EMF Java implementation.

PyEcore allows you to handle models and metamodels (structured data model), and gives the key you need for building MDE-based tools and other applications based on a structured data model. It supports out-of-the-box:

• Data inheritance,
• Two-ways relationship management (opposite references),
• XMI (de)serialization,
• JSON (de)serialization,
• Notification system,
• Reflexive API...

Let's see how we can create a very simple "dynamic" metamodel (as opposed to static ones, see the documentation for more details):

>>> from pyecore.ecore import EClass, EAttribute, EString, EObject
>>> Graph = EClass('Graph')  # We create a 'Graph' concept
>>> Node = EClass('Node')  # We create a 'Node' concept
>>>
>>> # We add a "name" attribute to the Graph concept
>>> Graph.eStructuralFeatures.append(EAttribute('name', EString,
                                                default_value='new_name'))
>>> # And one on the 'Node' concept
>>> Node.eStructuralFeatures.append(EAttribute('name', EString))
>>>
>>> # We now introduce a containment relation between Graph and Node
>>> contains_nodes = EReference('nodes', Node, upper=-1, containment=True)
>>> Graph.eStructuralFeatures.append(contains_nodes)
>>> # We add an opposite relation between Graph and Node
>>> Node.eStructuralFeatures.append(EReference('owned_by', Graph, eOpposite=contains_nodes))

With this code, we have defined two concepts: Graph and Node. Both have a name, and there exists a containment relationship between them. This relation is bi-directional, which means that each time a Node object is added to the nodes relationship of a Graph, the owned_by relation of the Node is also updated (it also works the other way around).

Let's create some instances of our freshly created metamodel:

>>> # We create a Graph
>>> g1 = Graph(name='Graph 1')
>>> g1
<pyecore.ecore.Graph at 0x7f0055554dd8>
>>>
>>> # And two node instances
>>> n1 = Node(name='Node 1')
>>> n2 = Node(name='Node 2')
>>> n1, n2
(<pyecore.ecore.Node at 0x7f0055550588>,
 <pyecore.ecore.Node at 0x7f00555502b0>)
>>>
>>> # We add them to the Graph
>>> g1.nodes.extend([n1, n2])
>>> g1.nodes
EOrderedSet([<pyecore.ecore.Node object at 0x7f0055550588>,
             <pyecore.ecore.Node object at 0x7f00555502b0>])
>>>
>>> # bi-directional references are updated
>>> n1.owned_by
<pyecore.ecore.Graph at 0x7f0055554dd8>

This example gives a quick overview of some of the features you get for free when using PyEcore.

The project slowly grows and it still requires more love.

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

You can read the documentation at this address:

https://pyecore.readthedocs.io/en/latest/

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.

These dependencies are directly installed if you choose to use pip.

The tests use py.test and 'coverage'. Everything is driven by Tox, so in order to run the tests simply run:

$ tox

• There is some meta-property that could be missing inside PyEcore. If you see one missing, please open a new ticket!
• Proxies are not "removed" once resolved as in the the Java version, instead they act as transparent proxies and redirect all calls to the 'proxied' object.
• PyEcore is able to automatically load some model/metamodel dependencies on its own.

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, only single root models),
• Eclipse XMI export (partially, only single root models),
• simple notification/Event system,
• EOperations support,
• code generator for the static part,
• EMF proxies (first version),
• object deletion (first version),
• EMF commands (first version),
• EMF basic command stack,
• EMF very basic Editing Domain,
• JSON import (simple JSON format),
• JSON export (simple JSON format),
• introduce behavior @runtime,
• resources auto-load for some cross-references,
• derived collections,
• multiple roots resources,
• xsi:schemaLocation support for XMI resources,
• URI mapper like,
• EGeneric support (first simple version),
• URI converter like

The things that are in the roadmap:

• new implementation of EOrderedSet, EList, ESet and EBag,
• new implementation of EStringToStringMapEntry and EFeatureMapEntry,
• improve documentation,
• copy/paste (?).

There aren't too many projects proposing to handle models and metamodels 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, only supports Class/Primitive Types (no Enumeration), XMI import/export and does not provide a reflexion layer. The project didn't move since 2005.

EMF4CPP proposes a C++ implementation of EMF. This implementation also introduces Python scripts to call the generated C++ code from a Python environment. It seems that the EMF4CPP does not provide a reflexive layer either.

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 and seems to provide a reflexive layer. The project does not appear to have moved since a while.

Thanks for making PyEcore better!

• Mike Pagel (@moltob), who is also the author of pyecoregen and pymultigen (on which pyecoregen is based)
• Terry Kingston (@TerryKingston)
• Afonso Pinto (@afonsobspinto)
• Andy (@CFAndy)
• annighoefer (@annighoefer)
• Rodriguez Facundo (@rodriguez-facundo)
• Filippo Ledda (@filippometacell)
• Ewoud Werkman (@ewoudwerkman)
• Eugene (@4ekin)
• Alexandre Acebedo (@aacebedo)
• Jinhu (@jinhu)
• Pablo Campillo (@pablo-campillo)
• Jose Antonio Moral (@jamoralp)
• Marco Heinemann (@ubmarco)
• Christian Drabek (@cdrabek)
• Dan Yeaw (@danyeaw)
• Arkadiusz Michał Ryś (@arrys)

• This article on the blog of Professor Jordi Cabot gives more information and implementation details about PyEcore.