Robin Becker wrote: > On 26/02/2013 18:38, Peter Otten wrote: >> Robin Becker wrote: > ...........3: >> >> $ python -m timeit -s 'from new import instancemethod >>> from math import sqrt >>> class A(int): pass >>> A.m = instancemethod(sqrt, None, A) >>> a = A(42) >>> ' 'a.m()' >> 1000000 loops, best of 3: 0.5 usec per loop >> $ python -m timeit -s 'from math import sqrt >>> class A(int): >>> def m(self): >>> return sqrt(self) >>> a = A(42) >>> ' 'a.m()' >> 1000000 loops, best of 3: 0.473 usec per loop >> >> > this analysis might be relevant if I wanted to use sqrt. However, in my > case the method takes > > > > py C > utf8 bytes 50 20 usec > unicode 39 15 > > here py refers to a native python method and C to the extension method > after adding to the class. Both are called via an instance of the class.
I think you misunderstood. You compare the time it takes to run the function coded in C and its Python equivalent -- that difference is indeed significant. But what I was trying to measure was the difference between two ways to wrap the C function: Given a function cfunc implemented in C and the two ways of turning it into a method (1) class A(object): pass def method(self, ...): return cfunc(self, ...) A.method = method (2) class A(object): pass A.method = new.instancemethod(cfunc, None, A) I interpreted my timeit results as an indication that both ways have roughly the same overhead (method (1) being 0.027 usec faster). I don't have your code available, that's why I picked math.sqrt as an example for cfunc. I expect that you will get a similar result with your actual cfunc and therefore can (and should IMO) use method (1) in both Python 2 and 3 -- but of course not being able to measure it myself it may turn out I'm wrong. -- http://mail.python.org/mailman/listinfo/python-list