Link
ping's cookbook
Search

Type system

What are types?

A type is a node used to describe the interface an instance can have, here are a few examples:

// Foo is now an interface type.
class Foo {}

// FooFn is now an alias of the `Foo Function()` type.
typedef FooFn = Foo Function();

// You can now create interface types of Bar with any subtype of Foo as the type argument.
class Bar<T extends Foo> {
  // T is a subtype of Foo in this context.
}

At the highest level, there are only a handful kinds of types:

  1. dynamic
  2. void
  3. interface types
  4. function types
  5. parameter types

The most common is interface types, which describe a class and resolved type arguments.

dart:core contains a bunch of classes with special type properties, I’ll cover those below.

Instances

Throughout an object’s lifetime, it has a single type, this type is determined when constructed and can never be changed:

int x = 2;
num y = x;
print(x is int); // true
print(y is int); // true
int z = y as int; // works

The type used to declare a variable is only the interface, it can store any assignable instance that implements said interface.

Methods

When you call a method on an instance, the type the instance was created with always determines the implementation of that method, for example:

class Foo {
  void hi() => print("i am foo");
}

class Bar implements Foo {
  void hi() => print("i am bar");
}

void callHi(Foo foo) => foo.hi();

void main() {
  callHi(Bar()); // prints "i am bar"
}

Here, Bar’s implementation of hi will always override calls from its instances, regardless of what context its in.

All types visible to dart code are a subtype of Object and inherit the default implementation of its interface.

Dart is strongly typed, that means the compiler can make strong guarantees about the type a value will have at runtime.

Strong typing does not mean methods are guaranteed to exist though, if a method is missing when called, dart calls the noSuchMethod method which will throw a NoSuchMethodError by default.

(42 as dynamic).foo(); // throws NoSuchMethodError

All field access on instances is done through calls to setter and getter methods.

When you declare a field inside of a class, it implicitly declares setter and getter methods that read and write to an internal variable. This is different from C# for example, where setters / getters and fields are incompatible declarations.

class Foo {
  int a; // This declares both set:a and get:a
}

class Bar extends Foo {
  int get a => super.a * 2; // This overrides get:a without touching set:a
}

main() {
  var foo = Bar();
  foo.a = 2;
  print(foo.a); // prints 4
}

Assignability

A variable can contain values that are not actual subtypes of its declared type, specifically null:

int x;
print(x is int); // false

This prints false because the is operator is a subtype check, not an assignability check.

The as operator on the other hand does do an assignability check:

int x;
print(x as int); // null, works

This is because a value x is assignable to T if either:

  1. x’s runtime type is a subtype of T.
  2. x is null and T is nullable.

Null vs void vs dynamic vs Object

The Null class is special, it throws a formatted NoSuchMethodError when methods other than get:hashCode, get:runtimeType, and operator== are called.

The dynamic and void types are both effectively an alias for Object, but change how visible methods are:

  • With Object, you can only access methods from the Object interface (just like a regular class), i.e. hashCode.
  • With void, you can store and cast, but not access any methods.
  • With dynamic, you can access any methods and call it with any arguments, those return values are also treated as dynamic.

Closures

Extraction is the process of taking an instance method and turning it into a closure, this is commonly called a tear-off.

Methods can be extracted by calling the getter with their name:

typedef ToStringFn = String Function();
ToStringFn getToString(Object x) => x.toString;

In this example we extract the toString method from an arbitrary object x, giving us a closure that can be called as if it was a regular instance call on x.

The above code is effectively the same thing as:

typedef ToStringFn = String Function();
ToStringFn getToString(Object x) => () => x.toString();

Except the former is a bit more efficient.

Functions are very special, they can actually refer to two different things:

  1. Function types declared with arguments and return type, i.e. void Function() foo;.
  2. The Function class as an interface type, which is a super type of any function.

Function types are similar to generic interface types, but can describe parameter names and types.

All function types are subtypes of Function, regardless of their return type and arguments:

print(print is Function); // true

Callable classes

Classes can be callable… kinda.

class Foo {
  void call() => print('hi');
}

void main() {
  Foo()(); // prints "hi"
}

This is actually a little deceiving, Foo instances themselves are not actually callable, what’s happening is that the call method is being implicitly extracted.

Implicit tear-offs have some caveats, for example:

void callFoo(void Function() x) {
  print(x is Foo); // false
  print(x is Function); // true
  x();
}

void main() {
  var x = Foo();
  print(x is Foo); // true
  print(x is Function); // false
  callFoo(x);
}

The x here seems to be in a superposition between being Foo and being Function, this is because x is being implicitly converted into a closure before being passed to callFoo.