[Tutor] method, type?

[Alan Gauld]

Second attempt, my PC crashed just as I was sending the first
one. You might see two similar posts, if so apologies...

On 06/01/16 14:46, Steven D'Aprano wrote:
> I don't understand what you mean by "Python doesn't support named 
> constructors". It seems to me that this is the *only* sort of 
> constructor that Python supports.

No, Python constructors have names(new/init) but they are not
used explicitly (ie. by the client code)to create objects. By
contrast languages like Delphi, Smalltalk, Objective C,
some Lisps and, I think, Eiffel (and others?) all require an
explicit call of a constructor method and there may be
multiple constructors per class each with different names
(usually describing how the construction occurs or the
nature of the object constructed).

For example in Smalltalk:

obj := MyClass new.  # new is the normal choice of name
obj2 := Date today.
obj3 := Date new.
obj4 := Time now.

All of these methods (new, today, now) are constructors(*) but
with explicit names that tells you something about the kind
of instance created.

(*)As I understand it, Smalltalk constructors are just class
methods allocated to a constructor category. But the category
is just an organisational device used by the IDE/library
it doesn't actually change the code in any way.

> As I understand it, "named constructor" comes from the C++ world, where 
> functions are matched not just by name, but by parameters as well.

That's overloading which uses the same name but with different
numbers/types of parameters and/or return type. The problem with
that is that in C++ constructors must have the name of the class.
So if you want multiple constructors that all take a single
string as their paramater then it gets tricky.

> two integers representing the position relative to the entire screen, or 
> two integers representing the position relative to the current window. 
> The C++ compiler cannot distinguish those two cases, and would give an 
> error.

Exactly so.

> The solution is to use constructors with different names, which C++ 
> calls "named constructors".

I stopped using C++ around v2 and it didn't have such a feature.
Maybe its been added since. If so that's good to know. (Time
for some googling methinks...)

> So the only way to have two different constructors is to give them 
> different names.
> 
> Hence all constructors in Python are "named constructors".

But they are not native constructors such as those used in
Smalltalk etc. They have to be factory methods that call new/init
under the covers. And that's how most OOP languages that don't
support named constructors get round it. (Although interestingly,
Objective C refers to all constructors as 'factory methods', even
the 'default' ones.)

> It would be reasonable to make this a factory function declared in the 
> global module level. But I think making it a classmethod is better. 

I won't put up much of an argument here. In C++ or Java I'd
definitely say use a class method. But it seems to me that I've
seen more factory functions in modules than class method factories.
But that's not been a detailed analysis just a gut feel for what
seems idiomatic in Python. It also seems easier for beginners
to grasp than the more esoteric notion of class methods.

> ... ensures that if you subclass the class, you automatically
> get a valid constructor as well.

Wouldn't it return an instance of the superclass rather than
the sub class? You'd need to override it wouldn't you?


-- 
Alan G
Author of the Learn to Program web site
http://www.alan-g.me.uk/
http://www.amazon.com/author/alan_gauld
Follow my photo-blog on Flickr at:
http://www.flickr.com/photos/alangauldphotos