Duck typing, must it be dynamic?

Question

Wikipedia used to say* about duck-typing:

In computer programming with object-oriented programming languages, duck typing is a style of dynamic

Answer 1

I see "duck typing" more as a programming style, whereas "structural typing" is a type system feature.

Structural typing refers to the ability of the type system to express types that include all values that have certain structural properties.

Duck typing refers to writing code that just uses the features of values that it is passed that are actually needed for the job at hand, without imposing any other constraints.

So I could use structural types to code in a duck typing style, by formally declaring my "duck types" as structural types. But I could also use structural types without "doing duck typing". For example, if I write interfaces to a bunch of related functions/methods/procedures/predicates/classes/whatever by declaring and naming a common structural type and then using that everywhere, it's very likely that some of the code units don't need all of the features of the structural type, and so I have unnecessarily constrained some of them to reject values on which they could theoretically work correctly.

So while I can see how there is common ground, I don't think duck typing subsumes structural typing. The way I think about them, duck typing isn't even a thing that might have been able to subsume structural typing, because they're not the same kind of thing. Thinking of duck typing in dynamic languages as just "implicit, unchecked structural types" is missing something, IMHO. Duck typing is a coding style you choose to use or not, not just a technical feature of a programming language.

For example, it's possible to use isinstance checks in Python to fake OO-style "class-or-subclass" type constraints. It's also possible to check for particular attributes and methods, to fake structural type constraints (you could even put the checks in an external function, thus effectively getting a named structural type!). I would claim that neither of these options is exemplifying duck typing (unless the structural types are quite fine grained and kept in close sync with the code checking for them).

Answer 2

It's my understanding that structural typing is used by type inferencers and the like to determine type information (think Haskell or OCaml), while duck typing doesn't care about "types" per se, just that the thing can handle a specific method invocation/property access, etc. (think respond_to? in Ruby or capability checking in Javascript).

Answer 3

Structural Type System

A structural type system compares one entire type to another entire type to determine whether they are compatible. For two types A and B to be compatible, A and B must have the same structure – that is, every method on A and on B must have the same signature.

Duck Typing

Duck typing considers two types to be equivalent for the task at hand if they can both handle that task. For two types A and B to be equivalent to a piece of code that wants to write to a file, A and B both must implement a write method.

Summary

Structural type systems compare every method signature (entire structure). Duck typing compares the methods that are relevant to a specific task (structure relevant to a task).

Answer 4

There are always going to be examples from some programming languages that violate some definitions of various terms. For example, ActionScript supports doing duck-typing style programming on instances that are not technically dynamic.

var x:Object = new SomeClass();
if ("begin" in x) {
    x.begin();
}

In this case we tested if the object instance in "x" has a method "begin" before calling it instead of using an interface. This works in ActionScript and is pretty much duck-typing, even though the class SomeClass() may not itself be dynamic.

Answer 5

Duck typing means If it just fits, it's OK

This applies to both dynamically typed

def foo obj
    obj.quak()
end

or statically typed, compiled languages

template <typename T>
void foo(T& obj) {
    obj.quak();
}

The point is that in both examples, there has not been any information on the type given. Just when used (either at runtime or compile-time!), the types are checked and if all requirements are fulfilled, the code works. Values don't have an explicit type at their point of declaration.

Structural typing relies on explicitly typing your values, just as usual - The difference is just that the concrete type is not identified by inheritance but by it's structure.

A structurally typed code (Scala-style) for the above example would be

def foo(obj : { def quak() : Unit }) {
    obj.quak()
}

Don't confuse this with the fact that some structurally typed languages like OCaml combine this with type inference in order to prevent us from defining the types explicitly.

Answer 6

I'm not sure if it really answers your question, but...

Templated C++ code looks very much like duck-typing, yet is static, compile-time, structural.

template<typename T>
struct Test
{
    void op(T& t)
    {
        t.set(t.get() + t.alpha() - t.omega(t, t.inverse()));
    }
};