In my experience, prototype-based OOP is less prone to the typical OOP over-analysis - if you just need one object that does something, you just assemble it, that's it, problem solved. Since the default stance isn't designing a generic class for all possible subtypes that exhibit related behavior and blah blah blah, there's less push toward overthinking things.
Prototype-based OOP is also well-suited to runtime modification, since the underlying implementation is usually simpler, but that's a distant second to the above benefit. And while prototypes are sometimes assumed to be less efficient than classes, look at the research from Self* - While some popular languages have poor implementations, it's not inherently less efficient.
Also, I suspect prototypes integrate more smoothly with non-OOP code than classes, based on experience with Lua (which isn't strictly OOP, but typically uses prototypes). To some extent, this blurs with dynamic typing, though - there's only one statically typed + prototype-based OOP language I know of, Günther Blaschek's Omega (described in _Object-Oriented Programming With Prototypes_).
* http://selflanguage.org/documentation/published/index.html Some of the people involved later worked on the "JVM". Perhaps you've heard of it?
But that's not really a property of prototype languages - it's a property of duck typing. For example, Ruby - a non prototype language - can implement exactly the same idea. Take an instance of a class, and add a method specifically to that object - no problems. Critically, the thing that makes this possible is the fact that evaluation of the existence of the method is made at call time, not at compile time.
The way I write object-centric code in Lua feels very different from in Python, even though both languages are dynamically typed (and otherwise fairly similar). It probably has to do with conventions / what's "Pythonic". I haven't used Ruby enough to comment.
Given a sufficiently flexible and late-bound class system (like Smalltalk's), the differences between prototypes and classes can blur.
But in general I would say that it is easier to build classes on top of prototype systems than vice versa, which is one argument for prototypes.
All prototype implementations aren't the same, though, any more than Smalltalk and C++ both have the same class systems! Javascript and Self are quite different in some ways, especially when it comes to delegation (ie inheritance) and iolanguage and research such as Kevo are different again.
Despite the first answer to the linked question, I don't think that either class based or prototype based languages are easier to write a VM for.
What attracts me most is that prototype systems are conceptually simpler in an Occam's Razor sort of way. Instead of needing two concepts: classes and instances, we only need one: objects.
That's true, but I find you end up having to make the difference. Yes, prototype inheritance is conceptually simpler but in practice it's more complex. In JavaScript, for example, OO code tends to be harder to follow and includes more boilerplate code. You have to make up for the simplicity of the platform in order to get your work done. In more traditional OO languages, inheritance might be more limited but it's also more straight forward and easier to implement, calling parent class methods is build in, and the "this" reference works consistently.
Explain.
Class based inheritance, with a good metaprogramming model, is strictly more flexible than prototype based inheritance because you can easily implement the latter. The shortest implementation that I know of is the following Ruby one:
Proto = Class.new(Class) # Beware: magic.
def Proto.clone
Class.new(self)
end
Then again, Javascript's implementation of anything is awful. The only reason people use it, is because there is no way around it for web client programming. I'm pretty sure we shouldn't take it as an example.
(define (make-point-2D x y)
(define (get-x) x)
(define (get-y) y)
(define (set-x! new-x) (set! x new-x))
(define (set-y! new-y) (set! y new-y))
(lambda (selector . args) ; a dispatcher
(case selector
((get-x) (apply get-x args))
((get-y) (apply get-y args))
((set-x!) (apply set-x! args))
((set-y!) (apply set-y! args))
(else (error "don't understand " selector)))))
(define (make-account balance)
(define (withdraw amount)
(if (>= balance amount)
(begin (set! balance (- balance amount))
balance)
"Insufficient funds"))
(define (deposit amount)
(set! balance (+ balance amount))
balance)
(define (dispatch m)
(cond ((eq? m 'withdraw) withdraw)
((eq? m 'deposit) deposit)
(else (error "Unknown request -- MAKE-ACCOUNT"
m))))
dispatch)