bound vs unbound functions

[Kent Johnson]

Michael Tobis wrote:
> I'm trying to do metaprogramming. I'm sure I've got this all wrong
> wrong wrong, but somehow my approach hasn't yet hit a brick wall.
> 
> Anyway, I'd like to dynamically add a method to an instance at
> instantiation time. Something like
> 
> ######
> In [71]: class quux(object):
> ....:     def __init__(self,stuff):
> ....:         template = "def foo(self,b): print b + %s" % stuff
> ....:         exec(template)
> ....:         self.bazz = foo
> ....:
> 
> 
> In [72]: q = quux(5)
> 
> In [73]: q.bazz(4)
> ---------------------------------------------------------------------------
> TypeError                                 Traceback (most recent call
> last)
> 
> /Users/tobis/PyNSol/<console>
> 
> TypeError: foo() takes exactly 2 arguments (1 given)
> 
> In [74]: q.bazz("not much",4)
> 9
> ########

The thread Steve quoted suggests using new.instancemethod():

import new

class quux(object):
     def __init__(self,stuff):
         template = "def foo(self,b): print b + %s" % stuff
         exec(template)
         self.bazz = new.instancemethod(foo, self, quux)


There is no need for exec; you can define foo() directly as a nested function:

class quux(object):
     def __init__(self,stuff):
         def foo(self,b):
             print b + stuff
         self.bazz = new.instancemethod(foo, self, quux)


Of course for this simple example you can just remember stuff as an attribute:

class quux(object):
     def __init__(self,stuff):
         self.stuff = stuff

     def bazz(self, b):
         print b + self.stuff

Kent

> 
> So the straightforward question is why, even though bazz is a method of
> class quux, it doesn't have that extra call parameter 'self'. Is this a
> problem? If I actually need a reference to self is it OK to do:
> 
> In [76]: q.bazz(q,4)
> 
> ?
> 
> The more vague question is why do people despise 'exec', and how should
> I do this sort of thing instead?