Fantom

 

6. TypeSystem

Overview

The Fantom type system serves two primary purposes:

  • Encapsulation: a mechanism to group fields and methods together
  • Contracts: a mechanism to model semantics

Encapsulation

Types encapsulate a set of uniquely named slots. There are two types of slots: fields store state and methods model behavior. Grouping a set of slots together enables us to create packaged units of software which match our domain model.

This encapsulations serves several purposes:

  • Contracts: it enables the explicit contracts discussed next
  • Structure: it enables the three part namespace of pod::type.slot
  • Inheritance: grouping slots together enable reuse through inheritance

Contracts

Types are also used to specify an explicit and implicit contract. The explicit contract specifies what the class can do by the set of fields and methods it exposes. For example given a File, we know that it will support an explicit set of methods like exists, isDir, and size. This set of methods defines the contract for what we can and cannot do with a File. The compiler can use this information to perform type checking and report errors if you are attempting to use unknown slots. Sometimes you'll find compile time type checking gets in the way - in that case you simply switch from the . operator to the -> operator to delay type checking until runtime (or do something clever in your trap method).

The implicit contract specifies semantics that a human can understand - if I tell you a variable is a File, then you probably have a good understanding of what that variable is modeling. Programming is largely about mapping a problem domain into code - type systems help us annotate our code with domain specific terminology.

Types

There are two kinds of types in Fantom:

  • Classes
  • Mixins

Classes

Classes are the primary mechanism for specifying types. All objects are instances of exactly one class which can be queried at runtime via the Type.of method. Classes support single inheritance just like Java or C#.

Mixins

A mixin is a kind of type which is not designed to be used stand alone. Instead a mixin packages a group of slots together to be inherited into a class (or another mixin).

Mixins are similar to interfaces in Java or C#, but much more flexible. A Java or C# interface is purely a type definition of abstract methods, it can't actually include any behavior itself. Fantom mixins can declare concrete methods which provide a lot more power.

You can't create instances of a mixin - they are an abstract type designed to provide reuse when inherited into classes. Mixins also can't store state - although they can contain abstract fields to define a type contract which requires a field signature.

Pure Object Oriented

Fantom is fundamentally an object-oriented language. It is "pure" in the sense that everything is an object including core types such as Int. For example, Java primitive types such as boolean and int do not subclass from java.lang.Object which creates a fractured type system. Fantom defines a unified class hierarchy with sys::Obj as the root. Since everything is an object, you can call methods on everything:

Type.of(false) =>  sys::Bool
1972.isEven    =>  true
'x'.toChar     =>  "x"

Arrays are another type system anomaly not supported by Fantom. For example, in Java arrays are reference types which can be used as a java.lang.Object type, but they aren't proper classes with nice OO methods. In most circumstances, the List class is used instead of arrays. Plus you will use Buf instead of byte[] and StrBuf instead of char[].

Nullable Types

Types may be nullable or non-nullable. A non-nullable type is guaranteed to never store the null value. Nullable types are indicated with a trailing "?". This means non-nullable is the default unless otherwise specified:

Str   // never stores null
Str?  // might store null

The compiler prevents obvious mistakes when using a nullable expression when a non-nullable type is expected:

  • null literal
  • safe invoke method call or field access
  • as operator

Additional checks are implicitly done at runtime when coercing a nullable type to a non-nullable type. This allows your code to fail fast at the point where null bug was introduced versus propagating into unrelated code.

Value-Types

The special types Bool, Int, and Float are value-types. These types are optimized by the runtime to be passed by value instead of as a reference to an object. This allows Fantom to achieve the same performance as using primitives in Java and value-types in C#.

Value-types can be nullable also. For instance a variable declared to be Int? can store null.

Value-types differ from reference types in that fields default to false/zero instead of null. However the nullable versions of value-types do default to null.

A mapping of Fantom types to their runtime representations:

Fantom   Default   Java                .NET
------   -------   ----                ------
Bool     false     boolean             bool
Bool?    null      java.lang.Boolean   bool?
Int      0         long                long
Int?     null      java.lang.Long      long?
Float    0.0f      double              double
Float?   null      java.lang.Double    double?

By convention Fantom APIs use null to indicate an non-normal condition. For example, often in a Java API which returns an int such as String.indexOf() or InputStream.read() a special value of -1 will be used to indicate a non-normal result. This can be especially problematic when -1 is a valid result. In Fantom APIs we return Int? and use null instead of a special value like -1.

Statically Typed

Fantom is statically typed - all method and fields signatures require type declarations. This is a religious issue for many developers, but we believe type declarations just add too much value for code analysis and readability to throw them out for a bit of code compression.

However there are definitely times when a static type system gets in the way of an elegant solution. So Fantom provides some dynamic typing features too:

Implicit Casts

Anyplace where a compile time type check would typically require a cast in Java or C#, the compiler will implicitly insert a cast for you. The cast ensures that the JVM or CLR generates a runtime exception if the type check fails. If the compiler knows that the types are incompatible, then it will generate a compile time error.

Formally the rules are expressed as anytime where Type A is used and Type B is expected:

  1. If A.fits(B) the call is statically known to be correct
  2. Otherwise if B.fits(A) then we insert an implicit cast
  3. Otherwise it is a compile time error

For example:

Int func(Int x) { ... }

Int i := 5
Num n := 5
Str s := "foo"

// statically correct as is: Int.fits(Int)
func(i)  =>  func(i)

// implicit cast inserted: Int.fits(Num)
func(n)  =>  func((Int)n)

// compile time error: !Int.fits(Str)
func(s)  =>  error

This feature allows you to use Obj as a wildcard type which is assignable to anything. This is often used with in conjunction with dynamic invokes which return Obj?:

Str name := x->person->name
if (test->isTrue) {...}
File(x->uri)

Coercion from a non-nullable type to a nullable type is safe. A coercion from a nullable type to a non-nullable is implicitly allowed, but is checked at runtime:

Str? x := null
Str  y := x      // implicit cast as y := (Str)x

The above code will compile with the impilicit cast. However at runtime it will fail with a NullErr.

Type Signatures

We call the syntax used to express a type declaration a type signature. Type signatures are used extensively in your source code, in the fcode formats, and in the reflection APIs. The formal signature for a type is its qualified name or qname. Although in source code, we typically use the simple name in combination with the using statement. There is also a special syntax for expressing signatures of generic types.

Collections

There are two primary classes for managing collections: List and Map. Both of these types have a special literal syntax and a special type signature syntax.

List

Lists are a sequential collection of objects with fast integer indexing. A Fantom list is very similar to an ArrayList in Java or C# with similar performance tradeoffs: fast indexing and appending, but slower inserts and removes from the middle. Lists have a literal syntax and a special type signature syntax.

Map

Maps are a hashmap of key-value pair, very similar to an HashMap or Hashtable in Java or C#. Maps have a literal syntax and a special type signature syntax.

Generics

Although there isn't a general purpose generics mechanism yet, Fantom does use generics in a limited fashion. Specifically three classes use generics:

These are the only three generic types in Fantom. Each generic type uses a set of generic parameters in its method signatures. Generic parameters are always one of the following single ASCII letters: A-H, L, M, R, and V. The meaning of each generic parameter is discussed below.

To use a generic we have to specify a type for each of the generic parameters - we call this process parameterization. Fantom doesn't use a general purpose parameterization syntax like List<Str> as used by Java and C#. Instead each of the three generic types has its own custom parameterization syntax discussed below.

List Type Signatures

The List class uses two generic parameters:

  • V: type of item stored by the list
  • L: type of the parameterized list

The parameterization syntax of List is designed to mimic the array syntax of Java and C#:

// format
V[]

// examples
Str[]     // list of Strs
Int?[]    // list of Int?
Int[][]   // list of Int[] (list of a list of Ints)

The L generic parameter is used to indicate the parameterized type itself. For example the following is the signature of the List.add method:

L add(V item)

Given type Str[], then V maps to Str and L maps to Str[]. So the add method for Str[] is parameterized as:

Str[] add(Str item)

Map Type Signatures

The Map class uses three generic parameters:

  • K: type of key stored by the map
  • V: type of value stored by the map
  • M: type of the parameterized map

The parameterization syntax of Map is designed to mimic the map literal syntax:

// format
[K:V]          // formal signature
K:V            // brackets are optional in most cases

// examples
[Str:User]     // map of Users keyed by Str
Str:User       // same as above without optional brackets
Uri:File?      // map of File? keyed by Uri
[Uri:File]?    // map of Uri:File where the entire map variable might be null
Str:File[]     // map of File[] keyed by Str
[Str:File][]   // list of Str:File (brackets not optional)

The formal syntax for Map parameterization is [K:V]. Typically the brackets are optional, and by convention left off. But in some complicated type declarations you will need to use the brackets such as the [Str:File][] example above. Brackets are always used in APIs which return formal signatures.

Func Type Signature

The Func class uses nine generic parameters:

  • A to H: the function parameter types
  • R: the function return types

The parameterization syntax of Func is designed to match the syntax used by closures:

// format
|A a, B b ... H h -> R|

// examples
|Int a, Int b->Str|  // function which takes two Int args and returns a Str
|Int, Int->Str|      // same as above omitting parameter names
|->Bool|             // function which takes zero args and returns Bool
|Str s->Void|        // function which takes one Str arg and returns void
|Str s|              // same as above, omitting optional void return
|->Void|             // function which takes no arguments and returns void
|->|                  // shortcut for above

Function signatures are used extensively in functional programming and closures. It can be a bit tricky to grasp at first, but what we are parameterizing is the Func class itself - the arguments passed to the function and the return type.

To understand this a bit better, let's consider a Java example. We often want to declare the type of a "callback method" - in Java we typically do this by creating an interface. We then use that interface type whenever we need to specify a method that requires that callback:

interface Comparator
{
  int compare(Object a, Object b);
}

void sort(Comparator comparator)

In Fantom we skip the interface part and just declare the callback type using an in-place function signature:

Void sort(|Obj a, Obj b->Int| comparator)

This signature says that sort takes one argument called comparator which references a Func that takes two Objs and returns an Int.

But typically we are sorting a List which itself has been parameterized. List comes with a built-in sort method which has the actual signature:

L sort(|V a, V b->Int| c := null)

This method combines List's generic V parameter with a function signature. So given a list of type Str[], then the parameterized version of sort would be:

Str[] sort(|Str a, Str b->Int| c := null)

Function signatures are covered in yet more detail in the Functions chapter.

Subtype Substitution

The Liskov Substitution Principle defines that a subtype may be substituted for a super type and the behavior remains unchanged. As a general rule this principle applies to the Fantom type system and standard library.

Substitution rules are used to determine if type A is assignable to type B. During compilation we use these rules for checking variable assignment, field assignment, and method call parameters. These rules are also used by the is/isnot/as operators and the Type.fits method.

Fantom's subtyping substitution rules are defined as follows where A fits B if:

  1. A is B
  2. A is a class which extends class B
  3. A is a class which implements mixin B
  4. A is a mixin which implements mixin B
  5. A and B are sys::List type where all apply:
    1. A:V fits B:V
  6. A and B are sys::Map types where all apply:
    1. A:K fits B:K
    2. A:V fits B:V
  7. A and B are sys::Func types where all apply:
    1. B:R is Void or A:R fits B:R
    2. A airty <= B arity (number of parameters)
    3. B param i fits A param i for each parameter position i

Note function types have special "reverse" substitution rules, see Functions chapter for more extensive discussion.

Nullability is handled specially in that all cases above, non-nullable type A fits nullable B. However nullability is not considered by the is family of operators, nor by Type.fits.

Note that collection types List and Map allow contra-variance similar to how Java arrays work. For example Int[] is considered substitutable for Num[]. For the List getter methods this is true. However, for add/set methods this is not true, since one might add Float to a Num[], but this would be illegal for an Int[]. Despite this hole, we still make this trade-off for pragmatic reasons since most public APIs use readonly or immutable lists so extensively.

However note that in the JVM implementation, Lists are backed by the a Java array of the appropriate type, which in general will perform runtime checks during set/add operations. However, there is no absolute guarantee that contra-variant set/adds will be checked.