[PEAK] generic function signature changes

[Phillip J. Eby]

At 07:26 AM 6/15/2006 -0400, Kevin Dangoor wrote:
>Am I correct that it is harder to change a generic function's
>signature than it is to change an ordinary function's signature?

Actually, it's the *same* effort.  See below.


>  Let's
>say you have this function:
>
>def foo(bar)
>
>As an ordinary Python function, you can do this:
>
>def foo(bar, baz=None)
>
>without breaking any subclasses (if this were a method)

That's actually not true - you'll break subclasses that override the 
method, and this situation is precisely the same.  That is, the methods of 
a generic function are just like methods in subclasses.  In fact, if you 
are using only single-dispatch on the first argument, it is *exactly* the 
same: you have one GF method for each subclass.


>or anyone else
>using the function.

And those don't break in the generic function case, either.


>But, if you add an additional, even optional,
>parameter to a generic function, all of the implementers of that
>function need to change as well...
>
>Is that right?

Yes - just like if you add a parameter to a method in a base class or 
interface -- all the implementers or overriders of the method need to 
change as well.