Secretly passing parameter to function

[Chris Kaynor]

On Wed, Dec 5, 2012 at 10:50 AM, Olivier Scalbert <
olivier.scalbert at algosyn.com> wrote:

> Hi all !
>
> I have a problem that is not easy to explained, so I have tried to reduce
> it a lot.
>
> We are using a framework, that we can not modify.
>
> in framework.py:
> def do(something):
>    '''
>    Here we are in a framework that can not be modified ...
>    It does a lot of things
>
>    and finally:
>    '''
>    something()
>
> in test.py:
> from framework import *
>
> def step1():
>    print "Do step1"
>
> def step2():
>    print "Do step2"
>
>
> # We ask the framework to do some work.
> do(step1)
> do(step2)
> do(step3)
>
>
> We are writing step1, step2, ... and asking the framework to process them.
> Everything is ok, until we want to add a parameter to some steps.
> We want to be able to do that:
>
> in test.py:
> from framework import *
>
> def step1(param):
>    print "Do step1 with param"
>
> def step2():
>    print "Do step2"
>
>
> # We ask the framework to do some work.
>
> do(step1, param = None)
> do(step1, param = [0, 1, 5]) # again
> do(step2)
>
> Probably the easiest solution would be to use functools.partial to create
a proxy function, as follows:

import functools
do(functools.partial(step1, param=None))
do(functools.partial(step1, param=[0,1,5]))
do(step2)

Effectively what functools.partial does is returns a new function with the
arguments defined in the constructor. It looks something like:

def partial(func, *args, **kwargs):
    def newFunc(*cargs, **ckwargs):
        return func(*args+cargs, **kwargs+ckwargs)
    return newFunc

Note, that implementation of partial may not exactly match the semantics of
the real thing, and is untested and may just flat out not work...I'd
recommend using the real one instead.

In this case, it will produce a function that really takes no additional
arguments, as all arguments are defined as part of the creation, meaning
the new function will match the requirements by the framework of taking 0
arguments, despite the actual function taking and receiving one keyword
argument (param).

There is one caveat of using functools.partial: for positional arguments,
you cal only fill the arguments in left-to-right order; you cannot specify
the second argument of a list, other than specifying it by name. For
example, for the function "def test(a, b, c)", you can specify arguments
"a", "a and b", or "a and b and c" by position, but you cannot specify only
the arguments "b", "c", "a and c", or "b and c" by position. You can,
however, specify them by name, as keyword-arguments.



An alternative approach to the problem is to either use lamdas or write a
specialized wrapper like my example partial above to specify the arguments.
These eliminate the restriction on argument order for partial.

An example for lamdas would be:
do(lambda: step1(param=None))
do(lambda: step1(param=[0,1,5]))
do(step2)

This works much the same way as partial, in that it creates a new, unnamed
function which specifies the arguments.



> Of course it does not work ...
> TypeError: do() takes exactly 1 argument (2 given)
>
> And we can not modify the framework (in which "do" is defined.
>
> One solution would be to use a global variable that can be set before each
> step. But it is not very elegant ...
>
> One other approach would be to add dynamically an attribute the the step1
> function, and retrieve it inside the function, but it is perhaps overkill.
>
> Do you have some ideas ?
>
> Thanks,
>
>
> Olivier
>
> --
> http://mail.python.org/**mailman/listinfo/python-list<http://mail.python.org/mailman/listinfo/python-list>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/python-list/attachments/20121205/8df2183c/attachment.html>