Dynamically adding key-arguments to method

Question

I would like to set the default key-arguments of an instance method dynamically. For example, with

class Module(object):
    def __init__(self, **kargs):
        set-default-key-args-of-method(self.run, kargs)  # change run arguments
    def run(self, **kargs):
        print kargs

We would have:

m = Module(ans=42)

m.run.im_func.func_code.co_argcount  # => 2
m.run.im_func.func_code.co_varnames  # => ('self','ans','kargs')
m.run.im_func.func_defaults          # => (42,)
m.run()                              # print {'ans':42}

I tried something with types.CodeType (which I don't really understand) for a function (not a method) and got it to work (well not-to-fail), but the added key-arguments did not show in the kargs dictionary of the function (it only print {})

The change has to be done for the current instance only. Actually, I am using a class right now (I'm OO in my mind) so I would like to do it with a class method, but a function is maybe better. Something like:

def wrapped_run(**kargs):
    def run(**key_args):
        print key_args

    return wrap-the-run-function(run, kargs) 

run = wrapped_run(ans=42)

run.func_code.co_argcount  # => 1
run.func_code.co_varnames  # => ('ans','key_args')  ## keep the 'key_args' or not
run.func_defaults          # => (42,)
run()                      # print {'ans':42}

Any advise or idea is welcome.

A little on the context:

The Module class is some kind a function wrapper, which can be use to include the lower-end function in a dataflow system automatically but add intermediate procedures. I would like the module run function (actually, it will probably be it's __call___ function) to have the correct API in order for the dataflow system to nicely generate the correct module's input transparently.

I'm using python 2.7

Answer 1

For the sake of closure, I give the only solution that was found: use exec (proposed by mgilson)

import os, new

class DynamicKargs(object):
    """
    Class that makes a run method with same arguments
    as those given to the constructor
    """
    def __init__(self, **kargs):
        karg_repr = ','.join([str(key)+'='+repr(value) \
                              for key,value in kargs.iteritems()])
        exec 'def run(self,' + karg_repr + ',**kargs):\n    return self._run(' + karg_repr + ',**kargs)'

        self.run = new.instancemethod(run, self)

    def _run(self, **kargs):
        print kargs

# this can also be done with a function
def _run(**kargs):
    print kargs

def dynamic_kargs(**kargs):
    karg_repr = ','.join([str(key)+'='+repr(value) for key,value in kargs.iteritems()])
    exec 'def run(' + karg_repr + ',**kargs):\n    return _run(' + karg_repr + ',**kargs)'
    return run


# example of use
# --------------
def example():
    dyn_kargs = DynamicKargs(question='ultimate', answer=42)
    print 'Class example \n-------------'
    print 'var number:', dyn_kargs.run.im_func.func_code.co_argcount
    print 'var names: ', dyn_kargs.run.im_func.func_code.co_varnames
    print 'defaults:  ', dyn_kargs.run.im_func.func_defaults
    print 'run print: ', 
    dyn_kargs.run()
    print ''

    dyn_kargs = dynamic_kargs(question='foo', answer='bar')
    print 'Function example \n----------------'
    print 'var number:', dyn_kargs.func_code.co_argcount
    print 'var names: ', dyn_kargs.func_code.co_varnames
    print 'defaults:  ', dyn_kargs.func_defaults
    print 'run print: ', 
    dyn_kargs()

The example function prints:

Class example 
-------------
var number: 3
var names:  ('self', 'answer', 'question', 'kargs')
defaults:   (42, 'ultimate')
run print:  {'answer': 42, 'question': 'ultimate'}

Function example 
----------------
var number: 2
var names:  ('answer', 'question', 'kargs')
defaults:   ('bar', 'foo')
run print:  {'answer': 'bar', 'question': 'foo'}

However:

• there might be problem if arguments value are not well represented by their repr
• I think it is too complicated (thus not pythonic), and personally, I did not use it

Answer 2

You might be looking for something like this:

class Module(object):
    def __init__(self, **kargs):
        old_run = self.run.im_func
        def run(self,**kwargs):
            kargs_local = kargs.copy()
            kargs.update(kwargs)
            return old_run(self,**kargs)
        self.run = run.__get__(self,Module)

    def run(self,**kargs):
        print kargs

m1 = Module(foo=3,bar='baz')
m1.run()
print type(m1.run)

m2 = Module(foo=4,qux='bazooka')
m2.run()
print type(m2.run)

I've just created a wrapper instancemethod around the previous function. (partially inspired by this post).

Alternatively:

from functools import partial
from types import MethodType

class Module(object):
    def __init__(self, **kargs):
        self.run = MethodType(partial(self.run.im_func,**kargs),self,Module)

    def run(self,**kargs):
        print kargs

but this still doesn't give the API you're looking for...

Answer 3

kwargs is a dictionary, and all we should need to do is save it for later. Then the user can override it with their values.

class Module(object):
    def __init__(self, **kwargs):
        self.defaults = kwargs
    def run(**kwargs):
        values = dict(self.defaults.items() + kwargs.items())
        print values

EDIT

Are you perhaps looking for lambda function generation?

def wrapfunc(**kwargs):
    def run(kwargs):
        print kwargs
    return lambda x: run(dict(kwargs.items() + x.items()))

run = wrapfunc(ans=42)
run({})
run({'ans':12})

Answer 4

This doesn't sound like a good idea. Rather than mucking about with function signatures, it would be better to define a set of defaults as an instance variable, and use it in the function:

class Module(object):
    def __init__(self, **kwargs):
        self.run_defaults = kwargs
    def run(self, **kwargs):
        actual_values = self.run_defaults.copy()
        actual_values.update(kwargs)
        print actual_values