commit d24d1ca2d439a185416374eacca460e7b8213f21 Author: Isis Lovecruft isis@torproject.org Date: Wed Oct 31 16:12:26 2012 +0000
* Initial outline of metaclasses for dynamic class generation. * Data Descriptor class. * Utilities for tricks and hacks to play to pass data between instances, and call super on a parent class that you don't yet know the name of. * I know this stuff looks hacky, and thus it's not being used anywhere yet, but I think it could provide later optimizations, both for runtime for the code and also for decision trees to handle progressions of tests without knowing what the future tests will be. Maybe. --- ooni/utils/hacks.py | 32 +++++++ ooni/utils/meta.py | 241 +++++++++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 273 insertions(+), 0 deletions(-)
diff --git a/ooni/utils/hacks.py b/ooni/utils/hacks.py index cf5f080..e778540 100644 --- a/ooni/utils/hacks.py +++ b/ooni/utils/hacks.py @@ -57,3 +57,35 @@ def patched_reduce_ex(self, proto): else: return copy_reg._reconstructor, args
+class MetaSuper(type): + """ + Metaclass for creating subclasses which have builtin name munging, so that + they are able to call self.__super.method() from an instance function + without knowing the instance class' base class name. + + For example: + + from hacks import MetaSuper + class A: + __metaclass__ = MetaSuper + def method(self): + return "A" + class B(A): + def method(self): + return "B" + self.__super.method() + class C(A): + def method(self): + return "C" + self.__super.method() + class D(C, B): + def method(self): + return "D" + self.__super.method() + + assert D().method() == "DCBA" + + Subclasses should not override "__init__", nor should subclasses have + the same name as any of their bases, or else much pain and suffering + will occur. + """ + def __init__(cls, name, bases, dict): + super(autosuper, cls).__init__(name, bases, dict) + setattr(cls, "_%s__super" % name, super(cls)) diff --git a/ooni/utils/meta.py b/ooni/utils/meta.py new file mode 100644 index 0000000..054e580 --- /dev/null +++ b/ooni/utils/meta.py @@ -0,0 +1,241 @@ +#-*- coding: utf-8 -*- +# +# meta.py +# ------- +# Meta classes for coercing subclasses which don't exist yet. +# +# :authors: Isis Lovecruft +# :version: 0.1.0-alpha +# :copyright: 2012 Isis Lovecruft +# :license: see attached LICENSE file +# + +class MetaDescriptor(type): + """ + bug: "Attribute error: class <> has no attribute __bases__" + + There are only objects. However, there *are* two kinds of objects: + type-objects and non-type-objects. + + There are only two objects which do not have an attribute named + "__bases__": + + 1) Instances of the builtin object ``object`` itself (i.e. the + superclass of any top-level class in python), whose __class__ is + itself, and whose type is ``type(type)``: + + >>> o = object() + >>> type(o) + <type 'type'> + >>> o.__class__ + <type 'object'> + + The o.__class__ part seems to imply that the __bases__ of + ``object`` should be itself. + + 2) Old Style Classes. Which are despicable demons, deserving to be + sent straight back to hell. No offense to the good Satanas. The + only thing these have by default is __doc__ and __module__. Also, + and this is importants too: Old Style Classes do not have an + attribute named __class__, because they do not derive from + anything. + + Incidentally, the type of ``type(type)`` is itself, and the "__bases__" of + ``type(type)`` is... + + >>> t = type(type) + >>> t.__name__ + 'type' + >>> type(t) + <type 'type'> + >>> t.__class__ + <type 'type'> + >>> t.__bases__ + (<type 'object'>,) + + ``type(object)``. WTF. This is to say that the "__bases__" of ``object`` + is the ``type`` of itself. This strange loop is where all black magic + enters into Python. + + If we do "class Metaclass(type): pass", we can then call + ``super(Metaclass, mcls)``, and the resulting ``super`` object is actually + just ``type``: + + o Its type is ``type(type)``. + o Its __bases__ ``type(object)``. + + For example, ``super(Metaclass, mcls).__new__(mcls, *a, *kw)`` is the same + as saying ``type(mcls, "Metaclass", (type, ), {} )``, except that super + does some namespace munging with calling "self.__super" on its own type, + which is probably equivalent to the strange loops with type(type) and + type(object), but my brain is already flipping out (and I keep a string + cosmology textbook on my nightstand!). + + However, we should not ever be able to call + + >>> super(type, type(mcls)).__new__(type(mcls), 'type', (type(type),) {} ) + TypeError: object.__new__(type) is not safe, use type.__new__() + + Q: Why all this fuss? + + A: We need to force future class-level attributes of subclasses of + TestCase to be accessible (also at the class-level, without + instatiations) by TestCase. I.e.: + 1) class SubTestCase has class attribute optParameters, but no + class for doing anything with them, and they shouldn't have to. + They should just be able to define the options. + 2) Therefore, TestCase needs to have data descriptors, which get + inherited. + 3) We need to be able to do this without dangerous namespace + munging, because we cannot control the namespace of future + tests. Therefore, we cannot use hacks like "self.__super". + + We need a Metaclass, which creates a Metafactory for classes which are + dynamic test descriptors. If this is confusing, leave it alone, there's + witches is these woods. + + http://stackoverflow.com/a/10707719 + http://docs.python.org/2/howto/descriptor.html + http://www.no-ack.org/2011/03/strange-behavior-with-properties-on.html + http://www.cafepy.com/article/python_types_and_objects/ + http://stackoverflow.com/questions/100003/what-is-a-metaclass-in-python + + """ + ## DO NOT GIVE A METACLASS ATTRIBUTES HERE. + #descriptors = { } + + from byteplay import Code, opmap + from functools import partial + + def __new__(im_so_meta_even_this_acronym, name, base, dict): + + def hofstaeder_decorator(meta_decorator): + def meta_decorator_factory(*args, **kwargs): + def meta_decorator_wrapper(func): + return meta_decorator(func, *args, **kwargs) + return meta_decorator_wrapper + return meta_decorator_factory + + def _transmute(opcode, arg): + if ((opcode == opmap['LOAD_GLOBAL']) and + (arg == 'self')): + return opmap['LOAD_FAST'], arg + return opcode, arg + + def selfless(child): + code = Code.from_code(child.func_code) + code.args = tuple(['self'] + list(code.args)) + code.code = [_transmute(op, arg) for op, arg in code.code] + function.func_code = code.to_code() + return function + + acronym = ( ( k,v ) for k,v in dict.items( ) + if not k.startswith('__') ) + metanym, polymorph = ( (k for ( k,v ) in acronym), + (v for ( k,v ) in acronym) ) + morphonemes = (("get%s"%n, "set%s"%n, "del%s"%n) for n in metanym ) + + oracles = [] + for getter, setter, deleter in morphonemes: + childnym = getter[3:] + + @hofstaeder_decorator + def meta_decorator(func, *args, **kwargs): + def decorator_wrapper(first, last): + return func(first, last) + return decorator_wrapper + + @meta_decorator(getter, setter, deleter) + def decorated_property(first, last): + childnym = getter[3:] + class DataDescriptor(object): + @selfless + def __init__(childnym=None, polymorph): + setattr(self, childnym, polymorph) + + @property + @selfless + def getter(): + return self.childnym + return DataDescriptor(first, last) + + oracles.append(decorated_property(childnym, polymorph)) + + return super( + MetaDescriptor, im_so_meta_even_this_acronym).__new__( + im_so_meta_even_this_acronym, + metanym, + polymorph, + dict(oracles) ) +''' + @property + def x(self): ## or getx + return self._x + @x.setter + def x(self, value): ## or setx + self._x = value + @x.deleter + def x(self): ## or delx + del self._x + ## or 'x = property(getx, setx, delx, "documentation")' + ## ^ ^ ^ ^ + ## just need @property's name, initial value can be None + +Metaclass + Creates Metaclasses for each data descriptor in each SubTestCase + so, per SubTestCase, we get (usually two) descriptors: + optParameters and input + +''' + +def applyClassAttribute(obj, cls, get='optParameters'): + """ + I get attributes from an outside instances' dictionary and attempt to + apply them to a class. I require that the attributes I am trying to set be + data descriptors which is just Python name munging trick that is mild and + harmless enough to have it's own builtin decorator helper: + + class Foo(object): + def __init__(self, *a, **kw): + if 'thing' in kw: + self._thing = thing + @property + def thing(self): + return self._thing + @property.setter + def thing(self, value): + self._thing = value + @property.delter + def thing(self): + return del(self) + """ + from ooni.utils.assertions import isClass, isNotClass + + try: + assert isNotClass(obj), "must be an instance" + assert isClass(cls), "not a class" + ## obj is probably an instance + C = obj.__class__ ## of a subclass of nettest.TestCase + + assert issubclass(C, cls), "not a subclass of %s" % cls + assert C.__dict__.__contains__('optParameters'), \ + "%s in %s.__dict__ not found" % (get, C) + except AssertionError, ae: + log.debug(ae) + else: + attributes = classify_class_attrs(C) + ## uncomment this to have class attributes spewn everywhere: + #log.debug("Found class attributes:\n%s" % pprint(attributes)) + for attr in attributes: + if attr.name == str(get): + setattr(obj, str(get), attr.object) + if not hasattr(obj, str(get)): + log.debug("Unable to find class attribute %s" % get) + else: + log.debug("Applying %s.%s = %s to descriptor..." + % (C.name, attr.name, attr.object)) + ## This was an unfinished attempt at fixing a class' __bases__, I do + ## not know if it was heading in the right direction. It can be + ## removed it is still crufting up the space. --isis + if '__bases__' or '_parents' in C.__dict__: + pass