dirty_templates.py

class DirtyTemplateMeta(type):
    '''
    Metaclass for :class:`DirtyTemplate`
    '''
    def __new__(cls, name, bases, dct, *, template_bases=()):
        new_cls = super().__new__(cls, name, bases, dct)
        new_cls.template_dict = dct
        new_cls.template_bases = template_bases
        return new_cls


class DirtyTemplate(metaclass=DirtyTemplateMeta):
    '''
    Define a DirtyTemplate.

    This is a horrible hack job that allows (in theory anyway) to define
    a 'Template' class which can be used a bit like templates in C++, in
    particular to define a class once that can be rendered into multiple
    defined classes each with different base classes.

    This allows for some pretty wild stuff.

    So we can define a template that bases ``dict``, implements a method, and
    overrides another.

    Warning
    -------
    As I say, this is a terrible hack. I was experimenting with ways to be able
    to avoid repeating definitions but, due to some third-party dark magics,
    was unable to rely on mixin's or composition to get the job done.

    I'm not saying it's a good idea, or that it'll work, just a bit of
    experimentation.

    Examples
    --------

    >>> class SomeTemplate(DirtyTemplate, template_bases=(dict,)):
    ...     def some_method(self):
    ...         return "hello world"
    ...     def __getitem__(self, key):
    ...         return "constant"
    ...

    This shouldn't be instansiated directly. Instead it can be used to
    render different versions of the same class with different base classes.

    >>> @SomeTemplate.template
    ... class ImplA:
    ...     def some_method(self):
    ...         return "not hello"
    ...

    However, each rendering always has the template as the 'top class' rather
    than what you might expect.

    >>> a = ImplA()
    >>> a.some_method()
    'hello world'
    >>> a["thing"]
    'constant'

    It even has the base classes defined on the template;

    >>> ImplA.__bases__
    (<class 'dirty_templates.ImplA'>, <class 'dict'>)

    The behaviour from these base classes works as expected.
    >>> a["thing"] = 4
    >>> a.values()
    dict_values([4])


    In the next example we render a version of the template that inherits from
    a subclass of one of the bases of the template;

    >>> import collections
    >>> @SomeTemplate.template
    ... class ImplB(collections.OrderedDict):
    ...     pass
    ...


    Here again the template behaviour is used;

    >>> b = ImplB()
    >>> b.some_method()
    'hello world'
    >>> b["thing"]
    'constant'

    Note though, because the wrapped ``ImplB`` class already bases ``dict``
    (via OrderedDict) it's not added to the list of ``__bases__`` like it was
    in ``ImplA``

    >>> ImplB.__bases__
    (<class 'dirty_templates.ImplB'>,)

    but again the behaviour of ``OrderedDict`` is accessible;

    >>> b["thing"] = 4
    >>> b.move_to_end("thing")

    Finally we can obviously define new methods in the rendered classes;

    >>> @SomeTemplate.template
    ... class ImplC(collections.OrderedDict):
    ...     def some_new_method(self):
    ...         return "new hello"
    ...
    ...
    >>> c = ImplC()
    >>> c.some_method()
    'hello world'
    >>> c.some_new_method()
    'new hello'
    >>> c["thing"]
    'constant'

    '''
    # This is set by metaclass
    template_dict = None
    template_bases = ()

    @classmethod
    def template(cls, new_cls):
        '''
        Decorator to generate new templated classes.
        '''
        name = new_cls.__name__
        new_cls.__name__ = f'{name}_inner'

        bases = [new_cls]
        bases.extend(
            base for base in cls.template_bases
            if not issubclass(new_cls, base)
        )

        return type(name, tuple(bases), cls.template_dict)