Hacker News new | comments | show | ask | jobs | submit login

The problem with the author's case for Go's is-a relationships is that it breaks down the moment you want to pass the object to a function expecting the original object.

For example, http://play.golang.org/p/EmodogIiQU

    type A struct { }
    type B struct { A }  //B is-a A

    func save(A) { //do something }
 
    b := &B{}

    save(b);  //OOOPS! b IS NOT A
If Go had is-a relationships, the code above would be valid. Instead, Go only implements has-a relationships, and simply provides shortcuts to calling B.A.foo() as B.foo().

One could create B.save(), which would call save(b.A), but the very reason you're now proxying the call to save() is because there is no is-a relationship in Go.

We all know about interfaces, but the problem is that is-a relationships do exist, and you can't always use interfaces, because often you want to share the data encapsulated by the objects, not only the behavior. One ends up creating methods to fetch each piece of data, but in code that is supposed to be performant, calling methods instead of accessing fields is suboptimal.




In my experience its always really telling that there are no incredibly simple is-a relationships to use in inheritance justifications (as opposed to tons of goto real world examples for just about every other feature under the sun). Its always either incredibly abstract (B is-a A), or incredibly contrived (Triangle is-a Shape). I've spent a lot of time in inheritance heavy code, and I've yet to find something that wouldn't be just as, if not more, elegant without inheritance. I've spent the most time in Cocoa (which I believe to be very well designed BTW), but the inheritance there is clearly not needed IMO. I usually find one of the following to be true:

1) Very shallow inheritance trees that could have very easily (and more logically) been replaced with interfaces. For example, NSResponder being everything's superclass even though just about all of its methods are empty implementations ("subclassers responsibility"), aka, it clearly should have been an interface.

2) Confused/strangely complex is-a relationships (mutable array is-a immutable array, what? so if I specifically specify NSArray, I may still get a mutable array and the type system will be happy???).

3) Strange rules around what methods are overridable, and more importantly, how specifically they can be overridden. Why can't I in a UIView subclass override -subviews to ensure that it always has the same subviews? Well, implementation detail, that's why. Normally this would be fine, but since anyone is allowed to muck around in a subclass, suddenly I need to know the way it was implemented. This of course conflicts other parts of the framework where you are definitely expected to override the method and not use a setter.

I have yet to be presented with one of these "killer" is-a relationships that must exist. If the sole excuse is "performance", then sure I'll conceded. I guess I don't work in environments where member access is the performance bottleneck so I guess I can't relate.


Views in UI toolkits are the best example I know for an is-a relationship. A button is a view, a slider is a view. A container may have heterogenous children, but all of them are views. An interface doesn't work, because you would have to implement basic properties like containment, geometry, etc. separately for every widget.

To my knowledge, nobody has found any better design than this, though some have found worse ones (e.g. HTML and its bizarre input element).

To your specific points:

1. Yes, NSResponder should definitely have been an interface. See for example the weird way that documents and delegates are spliced into the responder chain.

2. The fact that NSMutable* is-a NS* is a lovely design. Mutability can be understood as adding setters to a class that doesn’t have them. The alternative seems to be weird duplicative splits like ArrayList/Array, or String/StringBuilder/CharSequence.

3. Agreed; it’s a failing of ObjC that it’s usually unspecified whether a method is designed to be called, to be overridden, or both. I wish this were enforced at the language level.


> In my experience its always really telling that there are no incredibly simple is-a relationships to use in inheritance justifications (as opposed to tons of goto real world examples for just about every other feature under the sun). Its always either incredibly abstract (B is-a A), or incredibly contrived (Triangle is-a Shape). I've spent a lot of time in inheritance heavy code, and I've yet to find something that wouldn't be just as, if not more, elegant without inheritance.

The flow tree (render object tree) in Servo (or any other browser engine) must use inheritance: we have a heterogeneous tree of objects that all share a common set of fields (position, intrinsic widths, collapsible margins, some various bits that store state during reflow), but they all use virtual methods because they must lay out their contents differently.

We can't use composition because we wouldn't get virtual methods. We can't use an interface because then we would be forced into virtual dispatch for all of those fields that are shared between flows.

Rust doesn't have OO yet either, so we're forced to hack around it in weird ways (usually via a small amount of unsafe code to simulate inheritance).

> I have yet to be presented with one of these "killer" is-a relationships that must exist. If the sole excuse is "performance", then sure I'll conceded. I guess I don't work in environments where member access is the performance bottleneck so I guess I can't relate.

A browser engine is exactly that sort of environment. Forcing all member access to go through virtual dispatch would murder the performance of any browser.

Note that this was exactly the sort of thing that OO was designed for in Simula: heterogeneous trees of objects that all share some common fields but have different virtual methods. This generalizes to GUI libraries, game worlds etc—in short, simulations :)


> The flow tree (render object tree) in Servo (or any other browser engine) must use inheritance: we have a heterogeneous tree of objects that all share a common set of fields (position, intrinsic widths, collapsible margins, some various bits that store state during reflow), but they all use virtual methods because they must lay out their contents differently.

Haven't used Servo, but one of the big eye opening composition experiences for me was Unity's Scene Graph. Whereas Cocoa uses an inheritance model for its view-tree, Unity has a tree of transforms that you do not subclass or change in any way, and then you add behaviors to those transforms. If you want it to render, you can attach a renderer, if you want to hit test, you attach a collider. If you want any arbitrary other thing to happen, you create that behavior. Its really nice, the idea of "tree" is completely separate from all other concepts. Rendering a 3D game, on mobile, at 60fps (on GC-ed Mono no less), makes me feel pretty good about its performance characteristics. Most our perf issues were with limiting draw calls and optimizing shaders, not method calling.

Similarly, I worked a lot on a browser engine in the past and virtual method dispatch was again not this clear cut performance killer.


> Most our perf issues were with limiting draw calls and optimizing shaders, not method calling.

Sounds like the work done by tree traversals weren't high overhead in general for your workload. But it does matter for some workloads.

> Similarly, I worked a lot on a browser engine in the past and virtual method dispatch was again not this clear cut performance killer.

We're seeing large gains from, as far as we can tell, having fewer virtual method calls than other engines. Eliminating virtual dispatch opens up a huge range of call-site optimizations since the methods can often be statically inlined (as well as reducing the load on the branch target buffer).

> Similarly, I worked a lot on a browser engine in the past and virtual method dispatch was again not this clear cut performance killer.

That doesn't match my experience. Devirtualization opens up lots of inlining opportunities, and inlining is one of the most critical optimizations that compilers can do (mostly because of the other optimizations that it opens up; e.g. const propagation, GVN, etc. etc.)

See this study: http://hubicka.blogspot.com/2014/04/devirtualization-in-c-pa...

Devirtualization optimizations improve Dromaeo by 7-8%. That's a significant win, especially since devirtualization is only a best-effort optimization and Dromaeo has a lot of JS in it.


    We can't use composition because we wouldn't get virtual methods.

  We can't use an interface because then we would be forced into virtual dispatch for all of those fields that are shared between flows.

  we have a heterogeneous tree of objects that all share a common set of fields (position, intrinsic widths, collapsible margins, some various bits that store state during reflow),
You're description seems to me to scream functions: That work on plain structs.

I'm not an expert in performance though so I don't if that is just as bad as the other alternatives you mentioned. But you didn't mention them so I thought I'd ask if you considered that as a valid approach and if so why you rejected them?


I don't know what you mean by functions on that work on plain structs; can you elaborate?


Inheritance is usually a mess because people use it to model objects when they should instead be modeling types: a triangle is not a shape, but a Shape is a sum type of Circle, Square, Triangle, etc.


Shape is definitely much more accurately modeled as an interface than a base class, since it has no functionality on its own.


I see 'is a' and think algebraic data types.


IOW, the Liskov substitution principle is unsatisfied. The language has chosen by design not to resolve "methods"† up the "hierarchy"†, therefore the only way to reinstate it is to implement func save(B) { }, which has either the benefit of making you think whether you need to persist more fields or make explicit that you don't. And as you said, implementing proxy accessors to satisfy an interface just to simulate inheritance is obviously not the right choice.

† neither of which exist, since there's no objects and no inheritance. In simplifying things, this brings a constraint. Given how method resolution is a pain point WRT implementation complexity and performance (see e.g Ruby), this is a reasonable tradeoff. I am glad we have such an interesting choice of languages.


Just call save(b.A) or save(&b.A).

No need for methods, is performant. I can't think of a reason--besides extra typing--why this wouldn't be sufficient.

http://play.golang.org/p/7vE_wN6EBv


It's sufficient, in that it does what you want - it calls save() on b's A.

But it's not sufficient if what you want is to treat a B as an A. There is no is-a, so you have to keep treating a B as something different from an A.


This becomes difficult to discuss in the abstract. In my experience, save() is a clear example of exactly why you want composition and not inheritance. When I am saving the B record, I want it to let the A record's save do its thing without interfering. Let's for the sake of argument use Person and Employee instead of A and B so we can use a concrete example. We'll say Person has a name field and an address field. Employee owns a Person, and has an EmployeeID as well as Salary. You want:

    Employee : Person
    {
        string employeeID;
        double salary;
    }
I want:

    Employee
    {
        Person person;
        string employeeID;
        double salary;
    }
If we're saving into JSON, your save contents are mixing with the Person's save contents ("{name, address, salary, employeeID}"). What happens when a "basic income" law is passed, and all of a sudden the Person's implementor decides to add a salary field to the Person object? My method continues working without hiccup, because I was explicit about how I saved:

    save()
    {
     write("{ employeeid:whatever, salary:whatever, person: ");
     save(person);
     write("}");
    }
My save file displays the same encapsulation as my code, and thus behaves correctly when a base class changes by default. On the other hand, in subclass-land you'd now have competing salary fields, so you'd have to explicitly prepare for that, instead of getting it semantically for free. For example, you could start defensively programming by name-spacing the save properties: "{person-name, person-address, person-salary, employee-salary, employee-employeeID}".

In my experience this reliance on what "things are" is a bad way to think about programming. It doesn't help anyone to argue the philosophy of whether Employee is a Person or not, because I can easily sidestep the argument by saying: "OK fine, Employees are Persons... but I'm not longer writing the Employee class, I'm writing the EmployeeRecord class. And as such the person instance itself is part of the employee set if it has an associated EmployeeRecord. I have now satisfied the is-a relationship without an is-a language feature". Kind of like how certain Rectangle ARE squares regardless of whether they happen to be members of the Square class. Its a membership requirement, not an instantiation requirement. Its just words.


Why not?

    type Saveable interface {
        ...
    }

    func save(a Saveable) { .... }

    save(a);  // ok, assuming A is Saveable
    save(b);  // necessarily ok, given than A is Saveable




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact

Search: