Metaclass programming in Python, Part 3
Metaprogramming without metaclasses
Last year, I attended the EuroPython 2006 conference. The conference was good, the organization perfect, the talks of very high level, the people extremely nice. Nonetheless, I noticed something of a disturbing trend in the Python community that prompted this article. Almost simultaneously, my co-author David Mertz was reflecting on a similar issue with some submitted patches to Gnosis Utilities. The trend at issue is the trend towards cleverness. Unfortunately, whereas cleverness in the Python community was once largely confined to Zope and Twisted, it now is appearing everywhere.
We have nothing against cleverness in experimental projects and learning exercises. Our gripe is with cleverness in production frameworks that we are forced to cope with as users. In this article, we hope to make a small contribution away from cleverness, at least in an area where we have some expertise, that being metaclass abuses.
For this article, we take a ruthless stance: we consider metaclass abuse any usage of a metaclass where you could have solved the same problem equally well without a custom metaclass.
One of the most common metaprogramming scenarios is the creation of classes with attributes and methods that are dynamically generated. Contrary to popular belief, this is a job where most of the time you do not need and you do not want a custom metaclass.
This article is intended for two sets of readers: average programmers who would benefit from knowing a few meta-programming tricks but are scared off by brain-melting concepts; and clever programmers who are too clever and should know better. The problem for the latter is that it is easy to be clever, whereas it takes a lot of time to become unclever. For instance, it took us a few months to understand how to use metaclasses, but a few years to understand how not to use them.
About class initialization
During class creation, attributes and methods of classes are set once and for all. Or rather, in Python, methods and attributes can be changed at nearly any point, but only if naughty programmers sacrifice transparency.
In various common situations, you may want to create classes in more dynamic ways than simply running static code for their creation. For instance, you may want to set some default class attributes according to parameters read from a configuration file; or you may want to set class properties according to the fields in a database table. The easiest way to dynamically customize class behavior uses an imperative style: first create the class, then add methods and attributes.
For example, an excellent programmer of our acquaintance, Anand Pillai, has
proposed a path to Gnosis Utilities' subpackage
gnosis.xml.objectify that does exactly this. A base class
gnosis.xml.objectify._XO_ that is specialized (at
runtime) to hold "xml node objects" is "decorated" with a number of
enhanced behaviors, like so:
Listing 1. Dynamic enhancement of a base class
setattr(_XO_, 'orig_tagname', orig_tagname) setattr(_XO_, 'findelem', findelem) setattr(_XO_, 'XPath', XPath) setattr(_XO_, 'change_pcdata', change_pcdata) setattr(_XO_,'addChild',addChild)
You might think, reasonably enough, that the same enhancement can be accomplished simply by subclassing the XO base class. True, in one sense, but Anand has provided about two dozen possible enhancements, and particular users might want some of them, but not others. There are too many permutations to easily create subclasses for every enhancement scenario. Still, the above code is not exactly pretty. You could accomplish the above sort of job with a custom metaclass, attached to XO, but with behavior determined dynamically. But that brings us back to the excessive cleverness (and opacity) that we hope to avoid.
A clean, and non-ugly, solution to the above need might be to add class decorators to Python. If we had those, we might write code similar to this:
Listing 2. Adding class decorators to Python
features = [('XPath',XPath), ('addChild',addChild), ('is_root',is_root)] @enhance(features) class _XO_plus(gnosis.xml.objectify._XO_): pass gnosis.xml.objectify._XO_ = _XO_plus
That syntax, however, is a thing of the future, if it becomes available at all.
When metaclasses become complicated
It might seem like all the fuss in this paper so far is about nothing. Why
not, for example, just define the metaclass of XO as
and be done with it.
Enhance.__init__() can happily add
whatever capabilities are needed for the particular use in question. This
might look like so:
Listing 3. Defining XO as Enhance
class _XO_plus(gnosis.xml.objectify._XO_): __metaclass__ = Enhance features = [('XPath',XPath), ('addChild',addChild)] gnosis.xml.objectify._XO_ = _XO_plus
Unfortunately, things are not so simple once you start to worry about inheritance. Once you have defined a custom metaclass for your base class, all the derived classes will inherit the metaclass, so the initialization code will be run on all derived classes, magically and implicitly. This may be fine in specific circumstances (for instance, suppose you have to register in your framework all the classes you define: using a metaclass ensures that you cannot forget to register a derived class), however, in many cases you may not like this behavior because:
- You believe that explicit is better than implicit.
- The derived classes have the same dynamic class attributes of the base class. Setting them again for each derived class is a waste, since they would be available anyway by inheritance. This may be an especially significant issue if the initialization code is slow or computationally expensive. You might add a check in the metaclass code to see if the attributes were already set in a parent class, but this adds plumbing and it does not give real control on a per-class basis.
- A custom metaclass will make your classes somewhat magic and nonstandard: you may not want to increase your chances to incur in metaclass conflicts, issues with "__slots__", fights with (Zope) extension classes, and other guru-level intricacies. Metaclasses are more fragile than many people realize. We have rarely used them for production code, even after four years of usage in experimental code.
- You feel that a custom metaclasses is overkill for the simple job of class initialization, and you would rather use a simpler solution.
In other words, you should use a custom metaclass only when your real intention is to have code running on derived classes without users of those classes noticing it. If this is not your case, skip the metaclass and make your life (and that of your users) happier.
The classinitializer decorator
What we present in the rest of this paper might be accused of cleverness. But the cleverness need not burden users, just us authors. Readers can do something much akin to the hypothetical (non-ugly) class decorator we propose, but without encountering the inheritance and metaclass conflict issues the metaclass approach raises. The "deep magic" decorator we give in full later generally just enhances the straightforward (but slightly ugly) imperative approach, and is "morally equivalent" to this:
Listing 4. Imperative approach
def Enhance(cls, **kw): for k, v in kw.iteritems(): setattr(cls, k, v) class ClassToBeInitialized(object): pass Enhance(ClassToBeInitialized, a=1, b=2)
The above imperative enhancer is not so bad. But it has a few drawbacks: it make you repeat the class name; readability is suboptimal since class definition and class initialization are separated -- for long class definitions you can miss the last line; and it feels wrong to first define something and then immediately mutate it.
classinitializer decorator provides a declarative
solution. The decorator converts
Enhance(cls,**kw) into a
method that can be used in a class definition:
Listing 5. The magic decorator in basic operation
>>> @classinitializer # add magic to Enhance ... def Enhance(cls, **kw): ... for k, v in kw.iteritems(): ... setattr(cls, k, v) >>> class ClassToBeInitialized(object): ... Enhance(a=1, b=2) >>> ClassToBeInitialized.a 1 >>> ClassToBeInitialized.b 2
If you have used Zope interfaces, you may have seen examples of class
zope.interface.implements). In fact,
classinitializer is implemented by using a trick copied from
zope.interface.advice, which credits Phillip J. Eby. The
trick uses the "__metaclass__" hook, but it does not use a custom
ClassToBeInitialized keeps its original metaclass,
i.e. the plain built-in metaclass
type of new style classes:
>>> type(ClassToBeInitialized) <type 'type'>
In principle, the trick also works for old style classes, and it would be easy to write an implementation keeping old style classes old style. However, since according to Guido "old style classes are morally deprecated", the current implementation converts old style classes into new style classes:
Listing 6. Promotion to newstyle
>>> class WasOldStyle: ... Enhance(a=1, b=2) >>> WasOldStyle.a, WasOldStyle.b (1, 2) >>> type(WasOldStyle) <type 'type'>
One of the motivations for the
classinitializer decorator is
to hide the plumbing and to make mere mortals able to implement their own
class initializers in an easy way, without knowing the details of how
class creation works and the secrets of the _metaclass_ hook. The other
motivation is that even for Python wizards it is very inconvenient to
rewrite the code managing the _metaclass_ hook every time one writes a new
As a final note, let us point out that the decorated version of
Enhance is smart enough to continue to work as a
non-decorated version outside a class scope, provided that you pass to it
an explicit class argument:
>>> Enhance(WasOldStyle, a=2) >>> WasOldStyle.a 2
The (overly) deep magic
Here is the code for
classinitializer. You do not need to
understand it to use the decorator:
Listing 7. The classinitializer decorator
import sys def classinitializer(proc): # basic idea stolen from zope.interface.advice, P.J. Eby def newproc(*args, **kw): frame = sys._getframe(1) if '__module__' in frame.f_locals and not \ '__module__' in frame.f_code.co_varnames: # we are in a class if '__metaclass__' in frame.f_locals: raise SyntaxError("Don't use two class initializers or\n" "a class initializer together with a __metaclass__ hook") def makecls(name, bases, dic): try: cls = type(name, bases, dic) except TypeError, e: if "can't have only classic bases" in str(e): cls = type(name, bases + (object,), dic) else: # other strange errs, e.g. __slots__ conflicts raise proc(cls, *args, **kw) return cls frame.f_locals["__metaclass__"] = makecls else: proc(*args, **kw) newproc.__name__ = proc.__name__ newproc.__module__ = proc.__module__ newproc.__doc__ = proc.__doc__ newproc.__dict__ = proc.__dict__ return newproc
From the implementation it is clear how class initializers work: when you
call a class initializer inside a class, your are actually defining a
_metaclass_ hook that will be called by the class' metaclass (typically
type). The metaclass will create the class (as a new style
one) and will pass it to the class initializer procedure.
Tricky points and caveats
Since class initializers (re)define the _metaclass_ hook, they don't play well with classes that define a _metaclass_ hook explicitly (as opposed to implicitly inheriting one). If a _metaclass_ hook is defined after the class initializer, it silently overrides it.
Listing 8. table project index.html home
>>> class C: ... Enhance(a=1) ... def __metaclass__(name, bases, dic): ... cls = type(name, bases, dic) ... print 'Enhance is silently ignored' ... return cls ... Enhance is silently ignored >>> C.a Traceback (most recent call last): ... AttributeError: type object 'C' has no attribute 'a'
While unfortunate, there is no general solution to this issue; we simply document it. On the other hand, if you call a class initializer after the _metaclass_ hook, you will get an exception:
Listing 9. Local metaclass raises an error
>>> class C: ... def __metaclass__(name, bases, dic): ... cls = type(name, bases, dic) ... print 'calling explicit __metaclass__' ... return cls ... Enhance(a=1) ... Traceback (most recent call last): ... SyntaxError: Don't use two class initializers or a class initializer together with a __metaclass__ hook
Raising an error is preferable to silently overriding your explicit _metaclass_ hook. As a consequence, you will get an error if you try to use two class initializers at the same time, or if you call the same one twice:
Listing 10. Doubled enhancement creates a problem
>>> class C: ... Enhance(a=1) ... Enhance(b=2) Traceback (most recent call last): ... SyntaxError: Don't use two class initializers or a class initializer together with a__metaclass__ hook
On the plus side, all issues for inherited _metaclass_ hooks and for custom metaclasses are handled:
Listing 11. Happy to enhance inherited metaclass
>>> class B: # a base class with a custom metaclass ... class __metaclass__(type): ... pass >>> class C(B): # class with both custom metaclass AND class initializer ... Enhance(a=1) >>> C.a 1 >>> type(C) <class '_main.__metaclass__'>
The class initializer does not disturb the metaclass of
which is the one inherited by base
B, and the inherited
metaclass does not disturb the class initializer, which does its job just
fine. You would have run into trouble, instead, if you tried to call
Enhance directly in the base class.
Putting it together
With all this machinery defined, customizing class initialization becomes rather straightforward, and elegant looking. It might be something as simple as:
Listing 12. Simplest form enhancement
class _XO_plus(gnosis.xml.objectify._XO_): Enhance(XPath=XPath, addChild=addChild, is_root=is_root) gnosis.xml.objectify._XO_ = _XO_plus
This example still uses the "injection" which is somewhat superfluous to
the general case; i.e. we put the enhanced class back into a specific name
in the module namespace. It is necessary for the particular module, but
will not be needed most of the time. In any case, the argument to to
Enhance() need not be fixed in code as above, you can equally
Enhance(**feature_set) for something completely dynamic.
The other point to keep in mind is that your
function can do rather more than the simple version suggested above. The
decorator is more than happy to tweak more sophisticated enhancement
functions. For example, here is one that adds "records" to a class:
Listing 13. Variations on class enhancement
@classinitializer def def_properties(cls, schema): """ Add properties to cls, according to the schema, which is a list of pairs (fieldname, typecast). A typecast is a callable converting the field value into a Python type. The initializer saves the attribute names in a list cls.fields and the typecasts in a list cls.types. Instances of cls are expected to have private attributes with names determined by the field names. """ cls.fields =  cls.types =  for name, typecast in schema: if hasattr(cls, name): # avoid accidental overriding raise AttributeError('You are overriding %s!' % name) def getter(self, name=name): return getattr(self, '_' + name) def setter(self, value, name=name, typecast=typecast): setattr(self, '_' + name, typecast(value)) setattr(cls, name, property(getter, setter)) cls.fields.append(name) cls.types.append(typecast)
The differing concerns of (a) what is enhanced; (b) how the magic works; and (c) what the basic class itself does are kept orthogonal:
Listing 14. Customizing a record class
>>> class Article(object): ... # fields and types are dynamically set by the initializer ... def_properties([('title', str), ('author', str), ('date', date)]) ... def __init__(self, values): # add error checking if you like ... for field, cast, value in zip(self.fields, self.types, values): ... setattr(self, '_' + field, cast(value)) >>> a=Article(['How to use class initializers', 'M. Simionato', '2006-07-10']) >>> a.title 'How to use class initializers' >>> a.author 'M. Simionato' >>> a.date datetime.date(2006, 7, 10)
- "Charming Python: Decorators make magic easy" (developerWorks, December 2006) takes a look at the newest Python facility for meta-programming.
- Read Michele's "A simple and useful doctester for your documentation".
- See all Linux tips and Linux tutorials on developerWorks.