A metaclass is used to wrap all (or just some) methods for logging purposes. The underlying mechanism can be used as well to check pre/post conditions, attribute access,... The basic point is, that the actual class must not be changed in any way to achive the desired effect.
| Python |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 | import re
log = open('log','w')
indent = 0
indStr = ' '
def logmethod(methodname):
def _method(self,*argl,**argd):
global indent
#parse the arguments and create a string representation
args = []
for item in argl:
args.append('%s' % str(item))
for key,item in argd.items():
args.append('%s=%s' % (key,str(item)))
argstr = ','.join(args)
print >> log,"%s%s.%s(%s) " % (indStr*indent,str(self),methodname,argstr)
indent += 1
# do the actual method call
returnval = getattr(self,'_H_%s' % methodname)(*argl,**argd)
indent -= 1
print >> log,'%s:'% (indStr*indent), str(returnval)
return returnval
return _method
class LogTheMethods(type):
def __new__(cls,classname,bases,classdict):
logmatch = re.compile(classdict.get('logMatch','.*'))
for attr,item in classdict.items():
if callable(item) and logmatch.match(attr):
classdict['_H_%s'%attr] = item # rebind the method
classdict[attr] = logmethod(attr) # replace method by wrapper
return type.__new__(cls,classname,bases,classdict)
class Test(object):
__metaclass__ = LogTheMethods
logMatch = '.*'
def __init__(self):
self.a = 10
def meth1(self):pass
def add(self,a,b):return a+b
def fac(self,val): # faculty calculation
if val == 1:
return 1
else:
return val * self.fac(val-1)
if __name__ == '__main__':
l = Test()
l.meth1()
print l.add(1,2)
print l.fac(10)
|
Discussion
This recipe shows, how a metaclass can help various programmers needs. In this example, all method calls (and there return value) are written into a log file. Running the code will result in the following log file:
<__main__.Test object at 0x8158c24>.__init__() : None <__main__.Test object at 0x8158c24>.meth1() : None <__main__.Test object at 0x8158c24>.add(1,2) : 3 <__main__.Test object at 0x8158c24>.fac(10) ||<__main__.Test object at 0x8158c24>.fac(9) ||||<__main__.Test object at 0x8158c24>.fac(8) ||||||<__main__.Test object at 0x8158c24>.fac(7) ||||||||<__main__.Test object at 0x8158c24>.fac(6) ||||||||||<__main__.Test object at 0x8158c24>.fac(5) ||||||||||||<__main__.Test object at 0x8158c24>.fac(4) ||||||||||||||<__main__.Test object at 0x8158c24>.fac(3) ||||||||||||||||<__main__.Test object at 0x8158c24>.fac(2) ||||||||||||||||||<__main__.Test object at 0x8158c24>.fac(1) ||||||||||||||||||: 1 ||||||||||||||||: 2 ||||||||||||||: 6 ||||||||||||: 24 ||||||||||: 120 ||||||||: 720 ||||||: 5040 ||||: 40320 ||: 362880
: 3628800
The only interesting bit is the rebinding of the methods and their replacement with a wrapper method. To make this example a little less dull, only methods are wrapped whose method name match some regular expression. This might be actually close to Aspect Oriented Programming (but I don't know enough about this to say for sure)


Comments
Non-metaclass solution. You can perform the same task on the fly, on any object, without metaclasses. This means you can reroute object methods from outside the object itself. Just reproduce the body of LogTheMethods in a "normal" function. The only bit you need to change is, instead of assigning classdict[attr] = logmethod(attr), you use new.instancemethod(function, instance, class)
In this example, I rewrote the block so that one needs only pass in a reference to any method, and the other properties are introspected. I use this technique to "lock" object methods at runtime: accessing the original method requires a corresponding key. None of the locked objects need a metaclass defined, or __metaclass__ declared.
Sign in to comment