What is proper way to require a method to be overridden?

[Mark Elston]

* Paddy wrote (on 1/4/2007 10:20 PM):
> belinda thom wrote:
> 
>> On Jan 4, 2007, at 9:28 PM, Carl Banks wrote:
>>
>>> jeremito wrote:
>>>> I am writing a class that is intended to be subclassed.  What is the
>>>> proper way to indicate that a sub class must override a method?
>>> You can't (easily).
>>>
>>> If your subclass doesn't override a method, then you'll get a big fat
>>> AttributeError when someone tries to call it.  But this doesn't stop
>>> someone from defining a subclass that fails to override the method.
>>> Only when it's called will the error show up.  You can, as others have
>>> noted, define a method that raises NotImplementedError.  But this
>>> still
>>> doesn't stop someone from defining a subclass that fails to override
>>> the method.  The error still only occurs when the method is called.
>>>
>>> There are some advantages to using NotImplementedError:
>>>
>>> 1. It documents the fact that a method needs to be overridden
>>> 2. It lets tools such as pylint know that this is an abstract method
>>> 3. It results in a more informative error message
>>>
>>> But, in the end, if someone wants to define a class that defiantly
>>> refuses to declare a method, you can't stop them.
>> This is the con of a dynamic language...
> In a language that has statements to force the user to over-ride a
> method when a class is sub-classed, what is to stop the lazy
> sub-classer from doing the equivalent of:
> 
>   define override_me(self, ...):
>     pass
> 
> And so get code through the compiler,, allowing them to 'meet their
> targets'?
> 

And this *could* be perfectly suitable if the class really doesn't make
use of the method.  (I have seen this in poorly designed super classes
and interface classes - sigh.)

> - Paddy.
>