docs/archive/LangRef.html

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
   "http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
  <title>Swift Language Reference Manual</title>
  <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  <meta name="author" content="Chris Lattner">
  <meta name="description"
      content="Swift Language Reference Manual.">
  <link rel="stylesheet" href="_static/swift.css" type="text/css">

  <script type="text/javascript" src="toc.js"></script>

</head>

<body>

<h1>Swift Language Reference</h1>

<p>
<!-- The Table of Contents is automatically inserted in this <div>.
 Do not delete this <div>. -->
<div id="nav"></div>
</p>

  <!-- ********************************************************************* -->
  <h2 id="decl">Declarations</h2>
  <!-- ********************************************************************* -->

  <pre class="grammar">
    decl ::= <a href="#decl-class">decl-class</a>
    decl ::= <a href="#decl-constructor">decl-constructor</a>
    decl ::= <a href="#decl-deinit">decl-deinit</a>
    decl ::= <a href="#decl-extension">decl-extension</a>
    decl ::= <a href="#decl-func">decl-func</a>
    decl ::= <a href="#decl-import">decl-import</a>
    decl ::= <a href="#decl-enum">decl-enum</a>
    decl ::= <a href="#decl-enum-element">decl-enum-element</a>
    decl ::= <a href="#decl-protocol">decl-protocol</a>
    decl ::= <a href="#decl-struct">decl-struct</a>
    decl ::= <a href="#decl-typealias">decl-typealias</a>
    decl ::= <a href="#decl-let">decl-let</a>
    decl ::= <a href="#decl-var">decl-var</a>
    decl ::= <a href="#decl-subscript">decl-subscript</a>
 </pre>


  <!-- ===================================================================== -->
  <h3 id="decl-top-level">Module-Scope Declarations</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    top-level ::= <a href="#brace-item-list">brace-item</a>*
  </pre>

  <p>The top level of a swift source file is grammatically identical to the
  contents of a func decl.  Some declarations, however, are restricted to
  module scope.
  </p>

  <!-- _____________________________________________________________________ -->
  <h4 id="brace-item-list">Brace Enclosed Items</h4>

  <pre class="grammar">
    brace-item-list ::= '{' brace-item* '}'

    brace-item      ::= <a href="#decl">decl</a>
    brace-item      ::= <a href="#expr">expr</a>
    brace-item      ::= <a href="#stmt">stmt</a>
  </pre>

  <p>The brace item list provides a sequencing operation which evaluates the
  members of its body in order.  Function bodies and the bodies of control
  flow statements use braces.  Also, the <a
    href="#decl-top-level">source file</a> itself is effectively a
  brace item list, but without the braces.
  </p>

  <!-- ===================================================================== -->
  <h3 id="decl-import">import Declarations</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    decl-import ::=  <a href="#attribute-list">attribute-list</a> 'import' import-kind? import-path

    import-kind ::= 'typealias'
    import-kind ::= 'struct'
    import-kind ::= 'class'
    import-kind ::= 'enum'
    import-kind ::= 'protocol'
    import-kind ::= 'var'
    import-kind ::= 'func'

    import-path ::= <a href="#any-identifier">any-identifier</a> ('.' <a href="#any-identifier">any-identifier</a>)*
  </pre>

  <p>'import' declarations allow named values and types to be accessed with
  local names, even when they are defined in other modules and namespaces.  See
  the section on <a href="#namebind">name binding</a> for more
  information on how these work.  import declarations are only allowed at
  module scope.</p>

  <p>'import' directives only impact a single source file: imports in one
  swift file do not affect name lookup in another file. import directives can
  only occur at the top level of a file, not within a function or namespace.</p>

  <p>An import without an explicit import-kind names a module; all of the
  module's members are imported into the current scope. The module's name is
  also imported into the current scope in order to allow qualified access to
  the module's members, which can be useful for disambiguation.</p>

  <p>If an import-kind is provided, the last element of the import path is
  taken to be the name of a decl <em>within</em> the module named by the rest of
  the path. Only that name is introduced into the current scope; the name of the
  module itself is <em>not</em> accessible, nor any other decls within the
  module.</p>

  <p>Different import-kinds perform different filters on the decls within a
  module:</p>

  <ul>
    <li><code>typealias</code> can be used to import any concrete type (struct,
    class, enum, or another typealias). It cannot be used to import protocols,
    which are often used for more than just their existential type.</li>

    <li><code>struct</code>, <code>class</code>, <code>enum</code> can be used
    to import any type whose <i>canonical type</i> is a struct, class,
    or enum, respectively. (This allows "Int" to be imported as a struct, for
    example, even though its definition in the standard library may be a
    typealias for another struct type.)</li>

    <li><code>protocol</code> is used to import a protocol</li>

    <li><code>var</code> is used to import a module-scoped variable</li>

    <li><code>func</code> will import all overloads of a function</li>
  </ul>

  <pre class="example">
    <i>// Import all of the top level symbols and types in a module.</i>
    import Swift

    <i>// Import a single type.</i>
    import typealias Swift.BufferedStream

    <i>// Import all addition overloads.</i>
    import func Swift.+
  </pre>

  <!-- ===================================================================== -->
  <h3 id="decl-extension">extension Declarations</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    decl-extension ::= 'extension' <a href="#type-identifier">type-identifier</a> <a href="#inheritance">inheritance</a>? '{' <a href="#decl">decl</a>* '}'
  </pre>

  <p>'extension' declarations allow adding member declarations to existing
     types, even in other source files and modules.  There are different
     semantic rules for each type that is extended.
  </p>

  <!-- _____________________________________________________________________ -->
  <h4 id="decl-extension-enum-struct"><a href="#decl-enum">enum</a>, <a
    href="#decl-struct">struct</a>, and <a href="#decl-class">class</a>
    declaration extensions</h4>

  <p>FIXME: Write this section.</p>

  <!-- ===================================================================== -->
  <h3 id="decl-let">let Declarations</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    decl-let    ::= <a href="#attribute-list">attribute-list</a> 'let' <a href="#pattern">pattern</a> initializer?  (',' pattern initializer?)*
    initializer ::= '=' <a href="#expr">expr</a>
  </pre>


  <p>'let' declarations define an immutable binding between an initializer value
  and a name.</p>

<p>Here are some examples of 'let' declarations:</p>

<pre class="example">
  <i>// Simple examples.</i>
  let a = 4
  let c : Int = 42

  <i>// This decodes the tuple return value into independently named parts</i>
  <i>// and both 'val' and 'err' are in scope after this line.</i>
  let (val, err) = foo()

  // let declarations require an initializer (though the type is optional).
  let b : Int        <i>// error: let requires an initializer</i>

  // Let bindings of classes make the binding immutable, not the object.
  class Rocket {
    func blastOff() { ... }
  }
  let rocket = Rocket()
  rocket.blastOff()      // okay
  rocket = Rocket()   <i>// error, cannot change a let binding</i>
</pre>

  <!-- "var Declarations" was converted to ReST -->

  <!-- ===================================================================== -->
  <h3 id="decl-func">func Declarations</h3>
  <!-- ===================================================================== -->

  <!-- "func Declarations" was converted to ReST -->

  <!-- _____________________________________________________________________ -->
  <h4 id="func-signature">Function signatures</h4>

  <pre class="grammar">
    func-signature ::= func-arguments func-signature-result?
    func-arguments ::= <a href="#pattern-tuple">pattern-tuple</a>+
    func-arguments ::= selector-tuple
    <a id="selector-tuple">selector-tuple</a> ::= '(' <a href="#pattern-tuple">pattern-tuple-element</a> ')' (<a href="#identifier">identifier-or-any</a> '(' pattern-tuple-element ')')+
    func-signature-result ::= '-&gt;' <a href="#type">type</a>
  </pre>

  <p>A function signature specifies one or more sets of parameter
    patterns, plus an optional result type.</p>

  <p>When a result type is not written, it is implicitly the empty tuple type,
    <tt>()</tt>.</p>

  <p>In the body of the function described by a particular signature,
    all the variables bound by all of the parameter patterns are in
    scope, and the function must return a value of the result type.</p>

  <p>An outermost pattern in a function signature must be <a
  href="#fully_typed_types">fully-typed</a> and irrefutable. If a result type is
  given, it must also be fully-typed.</p>

  <p>The type of a function with signature <tt>(P<sub>0</sub>)(P<sub>1</sub>)...(P<sub><i>n</i></sub>) -&gt; R</tt>
    is <tt>T<sub>0</sub> -&gt; T<sub>1</sub> -&gt; ... -&gt; T<sub><i>n</i></sub> -&gt; R</tt>,
    where <tt>T<sub><i>i</i></sub></tt> is the bottom-up type of the pattern
    <tt>P<sub><i>i</i></sub></tt>.  This is called "currying".  The
    behavior of all the intermediate functions (those which do not
    return <tt>R</tt>) is to capture their arguments, plus any
    arguments from prior patterns, and returns a function which takes
    the next set of arguments.  When the "uncurried" function is
    called (the one taking <tt>T<sub><i>n</i></sub></tt> and returning
    <tt>R</tt>), all of the arguments are then available and the
    function body is finally evaluated as normal.</p>

  <p>A function declared with a selector-style signature
    <tt>func(a<sub>0</sub>:T<sub>0</sub>) name<sub>1</sub>(a<sub>1</sub>:T<sub>1</sub>) ... name<sub><i>n</i></sub>(a<sub><i>n</i></sub>:T<sub><i>n</i></sub>) -&gt; R</tt>
    has the type <tt>(_:T<sub>0</sub>, name<sub>1</sub>:T<sub>1</sub>, ... name<sub><i>n</i></sub>:T<sub><i>n</i></sub>) -&gt; R</tt>,
    that is, the names of the fields in the argument tuple are the
    <tt>name<sub><i>n</i></sub></tt> identifiers preceding each argument
    pattern. However, in the body of a function
    described by a signature, those arguments will be bound using the
    corresponding
    <tt>a<sub><i>n</i></sub></tt> patterns inside
    the arguments. This allows for Cocoa-style keyword function
    names such as <tt>doThing(x, withThing:y)</tt> to be defined without
    requiring that an awkward keyword name be the same as the
    variable name.

  <p>Here are some examples of func definitions:</p>

  <pre class="example">
    <i>// Implicitly returns (), aka <a href="#stdlib-Void">Void</a></i>
    func a() {}

    <i>// Same as 'a'</i>
    func a1() -&gt; Void {}

    <i>// Function pointers to a function expression.</i>
    var a2 = func ()-&gt;() {}
    var a3 = func () {}
    var a4 = func {}

    <i>// Really simple function</i>
    func c(_ arg : Int) -&gt; Int { return arg+4 }

    <i>// Simple operators.</i>
    func [infix_left=190] +  (lhs : Int, rhs : Int) -&gt; Int
    func [infix_left=160] == (lhs : Int, rhs : Int) -&gt; Bool

    <i>// Curried function with multiple return values:</i>
    func d(_ a : Int) (b : Int) -&gt; (res1 : Int, res2 : Int) {
      return (a,b)
    }

    <i>// A more realistic example on a trivial type.</i>
    struct bankaccount {
      amount : Int

     static func bankaccount() -> bankaccount {
        // Custom 'constructor' logic goes here.
      }
      func deposit(_ arg : Int) {
        amount = amount + arg
      }

      static func someMetatypeMethod() {}
    }

    <i>// Dot syntax on metatype.</i>
    bankaccount.someMetatypeMethod()

    <i>// A function with selector-style signature.</i>

    enum PersonOfInterest {
      case ColonelMustard
      case MissScarlet
    }
    enum Room {
      case Conservatory
      case Ballroom
    }
    enum Weapon {
      case Candlestick
      case LeadPipe
    }

    func accuseSuspect(_ suspect:PersonOfInterest)
        inRoom(room:Room)
        withWeapon(weapon:Weapon) {
      print("It was \(suspect) in the \(room) with the \(weapon)")
    }

    <i>// Calling a selector-style function.</i>
    accuseSuspect(.ColonelMustard, inRoom:.Ballroom, withWeapon:.LeadPipe)
  </pre>

  <!-- ===================================================================== -->
  <h3 id="decl-typealias">typealias Declarations</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    We use the keyword "typealias" instead of "typedef" because it really is an
    alias for an existing type, not a "definition" of a new type.
  </div>


  <pre class="grammar">
    decl-typealias ::= typealias-head '=' <a href="#type">type</a>
    <a name="typealias-head"></a>typealias-head ::= 'typealias' <a href="#identifier">identifier</a> <a href="#inheritance">inheritance</a>?
  </pre>

  <p>'typealias' makes a named alias of a type, like a typedef in C.  From that
  point on, the alias may be used in all situations the specified name is. If an <a href="#inheritance">inheritance</a> clause is provided, it specifies protocols to which the aliased type shall conform.</p>

  <p>Here are some examples of type aliases:</p>

  <pre class="example">
    <i>// location is an alias for a tuple of ints.</i>
    typealias location = (x : Int, y : Int)

    <i>// pair_fn is a function that takes two ints and returns a tuple.</i>
    typealias pair_fn = (Int) -&gt; (Int) -&gt; (first : Int, second : Int)
  </pre>

  <!-- ===================================================================== -->
  <h3 id="decl-enum">enum Declarations</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    decl-enum ::= <a href="#attribute-list">attribute-list</a> 'enum' <a href="#identifier">identifier</a> <a href="#generic-params">generic-params</a>? <a href="#inheritance">inheritance</a>? enum-body
    enum-body ::= '{' decl* '}'

    decl-enum-element ::= <a href="#attribute-list">attribute-list</a> 'case' enum-case (',' enum-case)*
    enum-case ::= <a href="#identifier">identifier</a> <a href="#type-tuple">type-tuple</a>? ('->' <a href="#type">type</a>)?
  </pre>

  <p>An <tt>enum</tt> declaration creates a <a href="#type-enum">enum type</a>.
  Here are some examples of enum declarations:</p>

  <pre class="example">
    <i>// Declares three enums.</i>
    enum DataSearchFlags {
      case None
      case Backward
      case Anchored
    }

    <i>// Shorthand for the above.</i>
    enum DataSearchFlags {
      case None, Backward, Anchored
    }

    <i>// Declare discriminated union with enum decl.</i>
    enum SomeInts {
      case None
      case One(Int)
      case Two(Int, Int)
    }

    func f1(_ searchpolicy : DataSearchFlags)  <i>// DataSearchFlags is a valid type name</i>
    func test1() {
      f1(DataSearchFlags.None)  <i>// Use of constructor with qualified identifier</i>
      f1(.None)                 <i>// Use of constructor with context sensitive type inference</i>

      <i>// "None" has no type argument, so the constructor's type is "DataSearchFlags".</i>
      var a : DataSearchFlags = .None
    }

    enum SomeMoreInts {
      case None           <i>// Doesn't conflict with previous "None".</i>
      case One(Int)
      case Two(Int, Int)
    }

    func f2(_ a : SomeMoreInts)

    func test2() {
      <i>// Constructors for enum element can be used in the obvious way.</i>
      f2(.None)
      f2(.One(4))
      f2(.Two(1, 2))

      <i>// Constructor for None has type "SomeMoreInts".</i>
      var a : SomeMoreInts = SomeMoreInts.None

      <i>// Constructor for One has type "(Int) -&gt; SomeMoreInts".</i>
      var b : (Int) -&gt; SomeMoreInts = SomeMoreInts.One

      <i>// Constructor for Two has type "(Int,Int) -&gt; SomeMoreInts".</i>
      var c : (Int,Int) -&gt; SomeMoreInts = SomeMoreInts.Two
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="decl-struct">struct Declarations</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    decl-struct ::= <a href="#attribute-list">attribute-list</a> 'struct' <a href="#identifier">identifier</a> <a href="#generic-params">generic-params</a>? <a href="#inheritance">inheritance</a>? '{' decl-struct-body '}'
    decl-struct-body ::= <a href="#decl">decl</a>*
  </pre>

  <p>A struct declares a simple value type that can contain data members and
     have methods.</p>

  <p>The body of a 'struct' is a list of decls. Stored (non-computed) 'var'
     decls declare members with storage in the struct. Other declarations act
     like they would in an <a href="#decl-extension">extension</a> of the
     struct type.</p>

  <p>Here are a few simple examples:</p>

  <pre class="example">
    struct S1 {
      var a : Int, b : Int
    }

    struct S2 {
      var a : Int
      func f() -> Int { return b }
      var b : Int
    }
  </pre>


  <p>Here are some more realistic examples of structs:</p>

  <pre class="example">
    struct Point { x : Int, y : Int }
    struct Size { width : Int, height : Int }
    struct Rect {
      origin : Point,
      size : Size

      typealias CoordinateType = Int

      func area() -> Int { return size.width*size.height }
    }

    func test4() {
      var a : Point
      var b = Point.Point(1, 2)    // Silly but fine.
      var c = Point(y = 1, x = 2)  // Using metatype.

      var x1 = Rect(a, Size(42, 123))
      var x2 = Rect(size = Size(width = 42, height=123), origin = a)

      var x1_area = x1.width*x1.height
      var x1_area2 = x1.area()
    }
  </pre>

  <div class="commentary">
    Structs do not support inheritance due to undesirable ripple effects across
    the design of the language. For example, method dispatch would arguably need
    to become virtual, not static. The storage of the type would arguably need
    to become indirected so that an array of T could be implemented sanely
    (because we don't know if T is actually a T, or a subclass of T). We'd need
    to store the "isa"/vtable in the struct so that virtual method dispatch
    could be implemented, and this has additional storage costs. None of these
    tradeoffs make sense for the intended use cases we have in mind for structs
    (Ints, Floats, Points, Rects, UUIDs, IP addresses, C struct interop, etc,
    etc). Said differently: we're trying to force a hard wall
    between types that need indirect access by their nature and those types
    that need direct access by their nature. The former are called classes. The
    latter are called structs.
  </div>

  <!-- ===================================================================== -->
  <h3 id="decl-class">class Declarations</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    decl-class ::= <a href="#attribute-list">attribute-list</a> 'class' <a href="#identifier">identifier</a> <a href="#generic-params">generic-params</a>? <a href="#inheritance">inheritance</a>? '{' decl-class-body '}'
    decl-class-body ::= <a href="#decl">decl</a>*
  </pre>

  <p>A class declares a reference type referring to an object which can contain
     data members and have methods.  Classes support single inheritance;
     a parent class should be listed as the first type in the
     inheritance list.</p>

  <p>The body of a 'class' is a list of decls. Stored (non-computed) 'var' decls
     declare members with storage in the class. Non-type 'var' and 'func'
     decls declare instance members;type 'var' and 'func' decls declare
     members of the class itself.  Both class and instance members can
     be overridden by a subclass.</p>
  <p>Type declarations inside a class act essentially the same way as type
     declarations outside a class.</p>

  <p>FIXME: For the moment, see classes.rst for more details on the
     class system.</p>
  <p>FIXME: Add a reference to the section on generics.</p>

  <p>The only way to create a new instance of a class is with a
    call to one of the class's constructors.</p>

  <p>Here is a simple example:</p>

  <pre class="example">
    class C1 {
      var a : Int
      var b : Int
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="decl-protocol">Protocol Declarations</h3>
  <!-- ===================================================================== -->

 <pre class="grammar">
    decl-protocol ::= <a href="#attribute-list">attribute-list</a> 'protocol' <a href="#identifier">identifier</a> <a href="#inheritance">inheritance</a>? '{' protocol-member* '}'
  </pre>

  <p>A protocol declaration describes an abstract interface implemented by
     another type.  It consists of a set of declarations, which may be instance
     methods or properties. A type <i>conforms</i> to a protocol if it
     provides declarations that correspond to each of the declarations in
     a protocol.</p>

  <p>Here are some examples of protocols:</p>

  <pre class="example">
    protocol Document {
      var title : String
    }
 </pre>

  <!-- _____________________________________________________________________ -->
  <h4 id="protocol-member-func">'func' protocol elements</h4>

  <pre class="grammar">
    protocol-member ::= <a href="#decl-func">decl-func</a>
</pre>

  <p>'func' members of a protocol define a value of function type that may be
  accessed with dot syntax on a value of the protocol's type.  The function
  gets an implicit "self" argument of the protocol type or (for a type
  function) of the metatype of the protocol.</p>

 <!-- _____________________________________________________________________ -->
  <h4 id="protocol-member-var">'var' protocol elements</h4>

  <pre class="grammar">
    protocol-member ::= <a href="#decl-var">decl-var</a>
  </pre>

  <p>'var' members of a protocol define "property" values that may be accessed
  with dot syntax on a value of the protocol's type. The actual
  variables may have no storage, and will always be accessed by a getter
  and setter. Thus, the variables shall have neither an initializer
  nor a getter/setter clause.</p>

  <!-- _____________________________________________________________________ -->
  <h4 id="protocol-member-subscript">'subscript' protocol elements</h4>

  <pre class="grammar">
    protocol-member ::= <a href="#subscript-head">subscript-head</a>
  </pre>

  <p>'subscript' members of a protocol define subscripting operations
  that may be accessed with the subscript operator ('[]') applied to a
  value of the protocol's type. </p>

  <div class="commentary">
    TODO: There is currently no way to express a requirement for a
    read-only or write-only subscript operation or variable. We may
    end up doing this with some kind of 'const' or 'immutable'
    attribute.
  </div>

   <!-- _____________________________________________________________________ -->
  <h4 id="protocol-member-typealias">'typealias' protocol elements (associated types)</h4>

  <pre class="grammar">
    protocol-member ::= <a href="#typealias-head">typealias-head</a> ('=' <a href="#type">type</a>)?
  </pre>

  <p>'typealias' members of a protocol define associated types, which
  are types used within the description of a protocol (typically in
  the inputs and outputs of 'func' members) that vary from one
  conforming type to another. When an associated type has an <a
  href="#inheritance">inheritance</a> clause, any type meant to
  satisfy the associated type requirement must conform to each of the
  protocols specified within that inheritance clause. If a type is
  provided after the '=', it is a default definition for the
  associated type that will be used as the type witness if the type
  witness cannot be determined in any other way.</p>

  <pre class="example">
    protocol SequenceType {
      typename Iterator : IteratorProtocol
      func makeIterator() -> Iterator
    }
  </pre>

  <!-- "subscript Declarations" was converted to ReST -->

  <!-- ===================================================================== -->
  <h3 id="decl-constructor">constructor Declarations</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    decl-constructor ::= <a href="#attribute-list">attribute-list</a> 'init' <a href="#generic-params">generic-params</a>? constructor-signature <a href="#brace-item-list">brace-item-list</a>

    constructor-signature ::= <a href="#pattern-tuple">pattern-tuple</a> constructor-result?
    constructor-signature ::= <a href="#identifier">identifier-or-any</a> <a href="#selector-tuple">selector-tuple</a> constructor-result?

    constructor-result ::= '->' 'Self'
  </pre>

  <p>'constructor' declares a constructor for a class, struct, or enum.  Such
     a declaration is used whenever an object is constructed.  Specifically,
     for classes, it is used when a new expression is written, and for structs
     and enums, it is used for function application when the "function"
     is a metatype.</p>

  <p>FIXME: We haven't decided the precise rules for when constructors are
     implicitly declared.  Default construction doesn't work right for structs
     or enums.  We haven't decided what the restrictions are if a member
     isn't default-constructible.</p>

  <p>A simple example:</p>

  <pre class="example">
    struct X {
      var member : Int
      init(x : Int) {
        member = x
      }
    }
    var a = X(10)
  </pre>

  <p>If a class is derived from a superclass, it must explicitly invoke a
  superclass constructor using the <tt>super.init</tt> syntax.
  <tt>super.init</tt> may only be used in a subclass constructor;
  it is not valid in a struct, enum, or root class constructor. Additionally,
  <tt>super.init</tt> may only be referenced exactly once per derived
  constructor. An example:</p>

  <pre class="example">
    class View {
      var bounds : Rect
      init(bounds:Rect) {
        self.bounds = bounds
      }
    }

    class Button : View {
      var onClick : () -&gt; ()
      init(bounds:Rect, onClick:() -&gt; ()) {
        super.init(bounds)
        self.onClick = onClick
      }
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="decl-deinit">deinitializer Declarations</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    decl-deinit ::= <a href="#attribute-list">attribute-list</a> 'deinit' <a href="#brace-item-list">brace-item-list</a>
  </pre>

  <p>'deinit' declares a deinitializer for a class.  This function is called
     when there are no longer any references to a class object, just before it
     is destroyed.  Note that deinitializers can only be declared for classes,
     and cannot be declared in extensions. Subclass deinitializers implicitly
     invoke their superclass deinitializers after executing.</p>

  <p>FIXME: We haven't really decided the precise rules here, but it's probably
     a fatal error to either throw an exception or stash a reference to 'self'
     in a deinitializer.  Not sure what happens when we cause the reference count
     of another object to reach zero inside a deinitializer.  We might eventually
     allow deinitializers in extensions once we have ivars in extensions.</p>

  <p>A simple example:</p>

  <pre class="example">
    class X {
      var fd : Int
      deinit {
        close(fd)
      }
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="attribute-list">Attribute Lists</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    attribute-list        ::= /*empty*/
    attribute-list        ::= attribute-list-clause attribute-list
    attribute-list-clause ::= '@' attribute
    attribute-list-clause ::= '@' attribute ','? attribute-list-clause

    attribute      ::= attribute-infix
    attribute      ::= attribute-resilience
    attribute      ::= attribute-inout
    attribute      ::= attribute-autoclosure
    attribute      ::= attribute-noreturn
  </pre>

  <p>An attribute list is written as a sequence of attributes, each of which has
     a leading '@' sign.  Attributes can be optionally comma separated.
     Attributes may not be repeated within a list.</p>

  <!-- _____________________________________________________________________ -->
  <h4 id="attribute-infix">Infix Attributes</h4>

  <pre class="grammar">
    attribute-infix ::= 'infix_left'  '=' <a href="#integer_literal">integer_literal</a>
    attribute-infix ::= 'infix_right' '=' <a href="#integer_literal">integer_literal</a>
    attribute-infix ::= 'infix        '=' <a href="#integer_literal">integer_literal</a>
  </pre>

  <p>The infix attributes may only be applied to the declaration of a
  function of binary operator type whose name is an
  <a href="#operator"><tt>operator</tt></a>.  The name indicates the
  associativity of the operator, either left associative, right associative, or
  non-associative.</p>

  <p>FIXME: Implement these restrictions.</p>

  <!-- _____________________________________________________________________ -->
  <h4 id="attribute-resilience">Resilience Attribute</h4>

  <pre class="grammar">
    attribute-resilience ::= 'resilient'
    attribute-resilience ::= 'fragile'
    attribute-resilience ::= 'born_fragile'
  </pre>

  <p>See the resilience design.</p>

  <!-- _____________________________________________________________________ -->
  <h4 id="attribute-inout"><tt>inout</tt> Attribute</h4>

  <pre class="grammar">
    attribute-inout ::= 'inout'
  </pre>

  <p><tt>inout</tt> is only valid in a <tt>type</tt> that
    appears within either a <a href="#pattern"><tt>pattern</tt></a> of
    a <tt>function-signature</tt> or the input type of a function
    type.
  </p>

  <p><tt>inout</tt> indicates that the argument will be passed as an "in-out"
    parameter. The caller must pass an lvalue decorated with the <tt>&amp;</tt>
    prefix operator as the argument. Semantically, the value of the argument
    is passed "in" to the callee to a local value, and that local value is
    stored back "out" to the lvalue when the callee exits. This is normally
    indistinguishable from pass-by-reference semantics.</p>

  <p><tt>inout</tt> differs from traditional pass-by-reference when closures
    are involved. If a closure captures an <tt>inout</tt> parameter, the
    <em>local value</em> is captured, not the reference. The local value at
    the time of function exit is written back to the lvalue.
    If the closure outlives the lifetime of the call, the local value lives
    independent of the original lvalue; further mutations within the closure
    do not affect the lvalue that was passed as the byref argument.

    For example, the following code:

  <pre class=example>
    func foo(x: inout Int) -> () -> Int {
      func bar() -> Int {
        x += 1
        return x
      }
      // Call 'bar' once while the inout is active.
      bar()
      return bar
    }

    var x = 219
    var f = foo(&x)
    // x is updated to the value of foo's local x at function exit.
    print("global x = \(x)")
    // These calls only update the captured local 'x', which is now independent
    // of the inout parameter.
    print("local x = \(f())")
    print("local x = \(f())")
    print("local x = \(f())")

    print("global x = \(x)")
  </pre>

  will print:

  <pre class=example>
    global x = 220
    local x = 221
    local x = 222
    local x = 223
    global x = 220
  </pre>

  <p>The type being annotated must be <a href="#materializable">materializable</a>.
    The type after annotation is never materializable.

  <p>FIXME: we probably need a const-like variant, which permits
    r-values (and avoids writeback when the l-value is not physical).
    We may also need some way of representing <q>this will be
    consumed by the nth curry</q>.
  </p>

  <!-- _____________________________________________________________________ -->
  <h4 id="attribute-autoclosure">autoclosure Attribute</h4>

  <pre class="grammar">
    attribute-autoclosure ::= 'autoclosure'
  </pre>

  <p>The <tt>autoclosure</tt> attribute modifies a <a
  href="#type-function">function type</a>, changing the behavior of any
  assignment into (or initialization of) a value with the function type.
  Instead of requiring that the rvalue and lvalue have the same function type,
  an "auto closing" function type requires its initializer expression to have
  the same type as the function's result type, and it implicitly binds a
  closure over this expression.  This is typically useful for function arguments
  that want to capture computation that can be run lazily.</p>

  <p><tt>autoclosure</tt> is only valid in a <tt>type</tt> of a
  syntactic function type that is defined to take a syntactic empty tuple.
  </p>

  <pre class="example">
  <i>// An auto closure value.  This captures an implicit closure over the</i>
  <i>// specified expression, instead of the expression itself.</i>
  var a : @autoclosure () -&gt; Int = 4

  <i>// Definition of an 'assert' function.  Assertions and logging routines</i>
  <i>// often want to conditionally evaluate their argument.</i>
  func assert(_ condition : @autoclosure () -> Bool)

  <i>// Definition of the || operator - it captures its right hand side as</i>
  <i>// an autoclosure so it can short-circuit evaluate it.</i>
  func [infix_left=110] || (lhs: Bool, rhs: @autoclosure ()->Bool) -&gt; Bool

  <i>// Example uses of these functions:</i>
  assert(i &lt; j)
  if (a == 0 || b == 42) { ... }
  </pre>

  <!-- _____________________________________________________________________ -->
  <h4 id="attribute-noreturn">No Return Attribute</h4>

  <pre class="grammar">
    attribute-noreturn ::= 'noreturn'
  </pre>

  <p>Attribute <tt>noreturn</tt> is only valid in the attribute list of a
  function declaration or in the attribute list of a <tt>type</tt>
  that describes a syntactic function type.
  </p>

  <p><tt>noreturn</tt> indicates to the compiler that the function will not
  return to the caller. This attribute should be used to suppress the
  uninitialized variable, missing return warnings and errors. The compiler is
  also allowed to more aggressively optimize the code in presence of this
  attribute.
  </p>

  <p>If a function with no a <tt>noreturn</tt> attribute contains a
  <tt>return</tt> statement, an error will be raised.
  </p>


  <!-- ********************************************************************* -->
  <h2 id="type">Types</h2>
  <!-- ********************************************************************* -->

  <pre class="grammar">
    type ::= <a href="#attribute-list">attribute-list</a> <a href="#type-function">type-function</a>
    type ::= <a href="#attribute-list">attribute-list</a> <a href="#type-array">type-array</a>

    type-simple ::= <a href="#type-identifier">type-identifier</a>
    type-simple ::= <a href="#type-tuple">type-tuple</a>
    type-simple ::= <a href="#type-composition">type-composition</a>
    type-simple ::= <a href="#type-metatype">type-metatype</a>
    type-simple ::= <a href="#type-optional">type-optional</a>
  </pre>

  <p>Swift has a small collection of core datatypes that are built into the
      compiler.  Most user-facing datatypes are defined by the
      <a href="#stdlib">standard library</a> or declared as a user defined
      types.</p>

  <!-- _____________________________________________________________________ -->
  <h3>Metatypes</h3>

  <p id="metatype">Each type has a corresponding <i>metatype</i>, with the same
    name as the type, that is injected into the standard name lookup scope when
    a type is <a href="#decl">declared</a>.  This allows access to '<a
    href="#decl-func">static functions</a>' through dot syntax.  For example:</p>

  <pre class="example">
    // Declares a type 'foo' as well as its metatype.
    struct foo {
      static func bar() {}
    }

    // Declares x to be of type foo.  A reference to a name in type context
    // refers to the type itself.
    var x : foo

    // Accesses a static function on the foo metatype.  In a value context, the
    // name of its type refers to its metatype.
    foo.bar()
  </pre>

  <!-- _____________________________________________________________________ -->
  <h3 id="fully_typed_types">Fully-Typed Types</h3>

  <p>A type may be <i>fully-typed</i>.  A type is fully-typed <i>unless</i> one
    of the following conditions hold:</p>

  <ol>
    <li>It is a function type whose result or input type is not
        fully-typed.</li>
    <li>It is a tuple type with an element that is not
        fully-typed.  A tuple element is fully-typed unless it has no
        explicit type (which is permitted for defaultable elements) or its
        explicit type is not fully-typed.  In other words, a type is
        fully-typed unless it syntactically contains a tuple element with
        no explicit type annotation.</li>
  </ol>

  <p>A type being 'fully-typed' informally means that the type is specified
     directly from its type annotation without needing contextual or other
     information to resolve its type.</p>

  <!-- _____________________________________________________________________ -->
  <h3>Materializable Types</h3>

  <p id="materializable">A type may be <i>materializable</i>.  A type
    is materializable unless it is 1) annotated with
    a <a href="#attribute-inout"><tt>inout</tt></a> attribute or 2) a
    tuple with a non-materializable element type.  In general, variables
    must have materializable type.</p>


  <!-- ===================================================================== -->
  <h3 id="type-identifier">Named Types</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    type-identifier ::= type-identifier-component ('.' type-identifier-component)*
    type-identifier-component ::= <a href="#identifier">identifier</a> <a href="#generic-args">generic-args</a>?
  </pre>

  <p>Named types may be used simply by using their name.  Named types are
     introduced by <a href="#decl-typealias">typealias</a> declarations or
     through type declarations that expand to one.</p>

  <pre class="example">
    typealias location = (x : Int, y : Int)
    var x : location      <i>// use of a named type.</i>
  </pre>

  <p>Type names may use dot syntax to refer to names types declared in other
  modules or types nested within other types.</p>

  <pre class="example">
    <i>// Direct reference to a member of another module.</i>
    var x : Swift.Int
  </pre>

  <p>Each component of a named type may be followed by a list of generic
  parameters for that component enclosed in angle brackets <tt>&lt;&gt;</tt>.

  <pre class="example">
    <i>// A generic class definition.</i>
    class Dict&lt;K, V&gt; { }

    <i>// A variable of a generic instance type.</i>
    var map : Dict&lt;String, Int&gt;
  </pre>

  <!-- ===================================================================== -->
  <h3 id="type-tuple">Tuple Types</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    Tuples are everywhere in Swift: even the argument list of a function is a
    tuple of those arguments.
  </div>

  <pre class="grammar">
    type-tuple ::= '(' type-tuple-body? ')'
    type-tuple-body ::= type-tuple-element (',' type-tuple-element)* '...'?
    type-tuple-element ::= identifier ':' <a href="#type">type</a>
    type-tuple-element ::= <a href="#type">type</a>
  </pre>

  <p>Syntactically, tuple types are simply a (possibly empty) list of
  elements enclosed in parentheses.  A tuple type with a single, anonymous
  element is exactly that type: the parentheses are treated as
  grouping parentheses.</p>

  <p>Tuples are the low-level form of data aggregation in Swift, and are used as
  the building block of <a href="#type-function">function</a> argument lists,
  multiple return values, <a href="#decl-enum">enum</a> bodies, etc. Because
  tuples are widely accessible and available everywhere in the language,
  aggregate data access and transformation is uniform and powerful.</p>

  <p>Each element of a tuple contains an optional name followed by a type.</p>

  <p>If the tuple body ends with '...', the tuple is a varargs tuple. The type
     of the last element is changed from T to T[], and there are special rules
     for converting an expression to varargs tuple type.</p>

  <pre class="example">
  <i>// Variable definitions.</i>
  var a : ()
  var b : (Int, Int)
  var c : (x : (), y : Int)

  <i>// Tuple type inferred from an initializers:</i>
  var m = ()                     <i>// Type = ()</i>
  var n = (x: 1, y: 2)           <i>// Type = (x : Int, y : Int)</i>
  var o = (1, 2, 3)              <i>// Type = (Int, Int, Int)</i>

  <i>// Function argument and result is a tuple type.</i>
  func foo(_ x : Int, y : Int) -&gt; (val : Int, err : Int)

  <i>// enum and struct declarations with tuple values.</i>
  struct S {
    var (a : Int, b : Int)
  }
  enum Vertex {
    case Point2(x : Int, y : Int)
    case Point3(x : Int, y : Int, z : Int)
    case Point4(w : Int, x : Int, y : Int, z : Int)
  }
  </pre>


  <!-- ===================================================================== -->
  <h3 id="type-function">Function Types</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    type-function ::= <a href="#type">type-tuple</a> '-&gt;' <a href="#type">type</a>
  </pre>

  <p>Function types have a single input and single result type, separated by
     an arrow.  Because each of the types is allowed to be a tuple, we trivially
     support multiple arguments and multiple results.  "Function" types are
     more properly known as a "closure" type, because they can embody any
     context captured when the function value was formed.</p>

  <p>The result type of a function type must
     be <a href="#materializable">materializable</a>.  The argument type of a
     function is always required to be parenthesized (a tuple).  The behavior
     of function types may be modified with the <a
      href="#attribute-autoclosure"><tt>autoclosure</tt> attribute</a>.</p>

  <p>Because of the grammar structure, a nested function type like
     "(a) -&gt; (b) -&gt; c" is parsed as "(a) -&gt; ((b) -&gt; c)".  This means
     that if
     you declare this that you can pass it one argument to get a function that
     "takes b and returns c" or you can pass two arguments to "get a c".  This
     is known as <a href="http://en.wikipedia.org/wiki/Currying">currying</a>.
     For example:
  </p>

  <pre class="example">
    <i>// A simple function that takes a tuple and returns Int:</i>
    var a : (a : Int, b : Int) -&gt; Int

    <i>// A simple function that returns multiple values:</i>
    var a : (a : Int, b : Int) -&gt; (val: Int, err: Int)

    <i>// Declare a function that returns a function:</i>
    var x : (Int) -&gt; (Int) -&gt; Int

    <i>// y has type (Int) -&gt; Int</i>
    var y = x(1)

    <i>// z1 and z2 both has type Int, and both have the same value (assuming
    // the function had no side effects).</i>
    var z1 = x(1)(2)
    var z2 = y(2)

    <i>// An auto closure value.  This captures an implicit closure over the</i>
    <i>// specified expression, instead of the expression itself.</i>
    var a : @autoclosure () -> Int = 4
  </pre>

  <!-- ===================================================================== -->
  <h3 id="type-enum">Enum Types</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    'enum' types are known as <a
      href="http://en.wikipedia.org/wiki/Algebraic_data_type">algebraic data
      types</a> (ADTs) by the broader programming language community.
    We name them 'enum' after C enums, because ADTs
    fulfill many of the same roles as enums in the C tradition.
  </div>

  <p>an enum type is a simple discriminated union: the runtime representation
  of a value of enum type only has one of the specified elements at a time.</p>

  <p>All of the element types of an enum type must
  be <a href="#materializable">materializable</a>.</p>

  <p>an enum type is defined by a <a href="#decl-enum">enum decl</a>.

  <p>Values of enum type may not be default initialized unless the user
  provides a no-argument constructor.</p>

  <p>The enum metatype has a member corresponding to each declared element.
  For elements with a declared type, this member is a function which can
  construct an enum containing that element.  For elements without a
  declared type, the member is simply an enum value for that element.  A
  enum value has no accessible members except those explicitly defined
  by the user.</p>

  <p>A reference to a member of the enum metatype can be shortened using <a
  href="#expr-delayed-identifier">delayed identifier resolution</a>
  with <a href="#typecheck_context">context sensitive type inference</a>.
  </p>

  <p>The enum's value can be tested and accessed by pattern-matching the enum
    against a <a href="#pattern-enum-element">enum element pattern</a>.

  <p>TODO: Should attributes be allowed on enum elements?
  TODO: Eventually, with generics we'll have equality and inequality operators.
    Enum decls should be able to implicitly define these for their types.
  TODO: Need pattern matching and element extraction.
  </p>

  <!-- ===================================================================== -->
  <h3 id="type-array">Array Types</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    Note that array types are parsed inside-out, with the first
    bounds clause being the outermost one.  This little oddity is required
    for the bounds of nested arrays to correspond in sequence to subscript
    indexes.  That is, given an array "x : Int[5][7][11][13]" and a
    chained subscript expression of the form "x[i][j][k][l]", we really
    want "i" to be bounded by 5, "j" by 7, and so on.  This is probably
    the only case where C's rule of "declaration follows use" really makes
    sense.  There's precedent for this in many languages, including Java and C#.
  </div>

  <pre class="grammar">
    type-array ::= <a href="#type">type-simple</a>
    type-array ::= <a href="#type">type-array</a> '[' ']'
    type-array ::= <a href="#type">type-array</a> '[' <a href="#expr">expr</a> ']'
  </pre>

  <p>Array types include a base type and an optional size.  Array types indicate
  a linear sequence of elements stored consecutively memory.  Array elements may
  be efficiently indexed in constant time.  All array indexes are bounds checked
  and out of bound accesses are diagnosed with either a compile time or
  runtime failure (TODO: runtime failure mode not specified).</p>

  <p>While they look syntactically very similar, an array type with a size has
  very different semantics than an array without.  In the former case, the type
  indicates a declaration of actual storage space.  In the later case, the type
  indicates storage space allocated elsewhere of runtime-specified size.
  </p>

  <p>For an array with a size, the size must be more than zero (no
  indices would be valid).  For now, the array size must be a literal integer.
  TODO: Define a notion like C's integer-constant-expression for how constant
  folding works.</p>

  <p>The element type of an array type must
    be <a href="#materializable">materializable</a>.</p>

  <p>FIXME: Int[][] not valid because the element type isn't sized.  We need
  some constraint to reject this, or do we?</p>

  <p>Some example array types:</p>

  <pre class="example">
    <i>// A simple array declaration:</i>
    var a : Int[4]

    <i>// A reference to another array:</i>
    var b : Int[] = a

    <i>// Declare a two dimensional array:</i>
    var c : Int[4][4]

    <i>// Declare a reference to another array, two dimensional:</i>
    var d : Int[4][]

    <i>// Declare an array of function pointers:</i>
    var array_fn_ptrs : (: (Int) -&gt; Int)[42]
    var g = array_fn_ptrs[12](4)

    <i>// Without parens, this is a function that returns a fixed size array:</i>
    var fn_returning_array : (Int) -&gt; Int[42]
    var h : Int[42] = fn_returning_array(4)

    <i>// You can even have arrays of tuples and other things, these work right
    // through composition:</i>
    var array_of_tuples : (a : Int, b : Int)[42]
    var tuple_of_arrays : (a : Int[42], b : Int[42])

    array_of_tuples[12].a = array_of_tuples[13].b
    tuple_of_arrays.a[12] = array_of_tuples.b[13]
  </pre>

  <!-- _____________________________________________________________________ -->
  <h3 id="type-metatype">Metatype Types</h3>
  <pre class="grammar">
    type-metatype ::= type-simple '.' 'Type'
  </pre>

<p>Every type has an associated metatype <tt>type(of: T)</tt>.  A value of the metatype
    type is a reference to a global object which describes the type.
    Most metatype types are singleton and therefore require no
    storage, but metatypes associated with <a href="#decl-class">class
    types</a> follow the same subtyping rules as their associated
    class types and therefore are not singleton.</p>

    <!-- _____________________________________________________________________ -->
    <h3 id="type-optional">Optional Types</h3>
    <div class="commentary">
      Similar constructs exist in Haskell (<a
      href="http://hackage.haskell.org/packages/archive/base/latest/doc/html/Data-Maybe.html">Maybe</a>),
      the Boost library (<a
      href="http://www.boost.org/doc/libs/1_54_0/libs/optional/doc/html/index.html">Optional</a>),
      and C++14 (<a href="http://en.cppreference.com/w/cpp/utility/optional">optional</a>).
    </div>

    <pre class="grammar">
      type-optional ::= type-simple '?'-postfix
    </pre>

    <p>An optional type is syntactic sugar for the library type
    Optional&lt;T&gt;. This is a <a href="#decl-enum">enum</a> with two
    cases: None and Some, used to represent a value that may or may not be
    present.</p>

    <p>Swift provides a number of special, builtin behaviors involving
      this library type:
      <ul>
        <li>There is an implicit conversion from any type <code>T</code> to the
          corresponding optional type <code>T?</code>.</li>
        <li><code>weak</code> variables must have type <code>T?</code>
          and automatically become <code>None</code> when the referent begins
          deallocation.</li>
        <li>The <a href="#expr-optional">optional-chaining operator</a> works
          on values of type <code>T?</code>.</li>
        <li>Several other expressions generate values of type
          <code>T?</code>.</li>
      </ul>
      To support these intrinsic use cases, the library is required to
      provide functions with these exact signatures:
      <ul>
        <li><code>func _doesOptionalHaveValueAsBool<T>(v : T?) -> Bool</code></li>
        <li><code>func _diagnoseUnexpectedNilOptional()</code></li>
        <li><code>func _getOptionalValue<T>(v : T?) -> T</code></li>
      </ul>
    </p>

    <p>Since optional types are part of the
    <a href="#type-simple">type-simple</a> grammar, it is not possible to write
    <code>T[]?</code> for an optional array. Use <code>(T[]?)</code> instead.
    </p>

    <p>Some example optional types:</p>

    <pre class="example">
      <i>// A simple optional declaration:</i>
      var a : Int? // equivalent to Optional&lt;Int&gt;

      <i>// An empty optional:</i>
      var b : Int? = .None

      <i>// Declare an array of optionals:</i>
      var c : [Int?] = [10, nil, 42]
    </pre>

  <!-- _____________________________________________________________________ -->
  <h3 id="type-composition">Protocol Composition Types</h3>
  <pre class="grammar">
   type-composition ::= <a href="#type-identifier">type-identifier</a> ('&amp;' <a href="#type-identifier">type-identifier</a>)*
  </pre>

  <p>A protocol composition type composes together a number of
  protocols to describe a type that meets the requirements of each of
  those protocols. A protocol composition type <code>A &amp; B</code>
  is similar to an explicitly-defined protocol that inherits both
  <code>A</code> and <code>B</code></p>

<pre class="example">
protocol C : A, B { }
</pre>

  <p>but without the need to introduce a new name.</p>

  <div class="commentary">
    If we drop implicit conformance to protocols, protocol composition
    types become much more important, because they allow you to give a
    name to a composition without requiring types to explicitly
    conform to that name.
  </div>

  <p>Each of the types named in the <code>type-composition</code> shall
  refer to either a protocol or to a protocol composition. The empty
  protocol composition is the keyword <code>Any</code> and every
  type conforms to it.

 <pre class="example">
    <i>// A value that represents any type</i>
    var any : Any = 17

    <i>// A value that conforms to both the Document and Enumerator protocols</i>
    var doc : Document &amp; Enumerator
    doc.isEmpty()       <i>// uses Enumerator.isEmpty()</i>
    doc.title = "Hello" <i>// uses Document.title</i>
</pre>

  <!-- _____________________________________________________________________ -->
  <h3 id="inheritance">Type Inheritance</h3>
  <pre class="grammar">
    inheritance ::= ':' <a href="#type-identifier">type-identifier</a> (',' <a href="#type-identifier">type-identifier</a>)*
  </pre>

  <p>A named type (e.g., a class, struct, enum, or protocol) can
 "inherit" some set of protocols, which implies that any object of
that type conforms to each of those protocols. When a protocol
inherits other protocols, the set of requirements from all of those
protocols is effectivel aggregated into the protocol, and a type that
conforms to the current protocol shall conform to each of the
 protocols that it inherits.</p>

  <p>When a non-protocol type inherits a protocol, it is specifying
explicitly that it conforms to that protocol. The program is
ill-formed if the type does not conform to the protocol.</p>

  <pre class="example">
    protocol VersionedDocument : Document { <i>// every VersionedDocument is a Document</i>
      func bumpVersion()
    }

    func print(_ doc : Document) { <i>/* ... */</i> }

    var myDocument : VersionedDocument;
    print(myDocument) <i>// okay: a VersionedDocument is a Document</i>

    class StoredHTML : VersionedDocument { <i>// okay: StoredHTML conforms to VersionedDocument</i>
      var Title : String
      func bumpVersion()
    }
  </pre>

  <!-- 'Patterns' converted to ReST. -->

  <!-- ********************************************************************* -->
  <h2 id="expr">Expressions</h2>
  <!-- ********************************************************************* -->

  <div class="commentary">
    Support for user-defined operators causes some amount of parsing
    to be delayed until after name resolution has occurred.  Other
    restrictions and disambiguations in the grammar permit the parser
    to decide all other aspects of parsing, such as where statements
    must be divided.<br><br>

    Semicolons in C are generally just clutter.  Swift generally tries
    to define away the need for them.
  </div>

  <pre class="grammar">
    expr          ::= expr-basic
    expr          ::= <a
    href="#expr-trailing-closure">expr-trailing-closure</a> <a href="#expr-cast">expr-cast</a>?

    expr-basic    ::= expr-sequence <a href="#expr-cast">expr-cast</a>?

    expr-sequence ::= <a href="#expr-unary">expr-unary</a> <a href="#expr-binary">expr-binary</a>*

    expr-primary  ::= <a href="#expr-literal">expr-literal</a>
    expr-primary  ::= <a href="#expr-identifier">expr-identifier</a>
    expr-primary  ::= <a href="#expr-super">expr-super</a>
    expr-primary  ::= <a href="#expr-closure">expr-closure</a>
    expr-primary  ::= <a href="#expr-anon-closure-arg">expr-anon-closure-arg</a>
    expr-primary  ::= <a href="#expr-paren">expr-paren</a>
    expr-primary  ::= <a href="#expr-delayed-identifier">expr-delayed-identifier</a>

    expr-postfix  ::= expr-primary
    expr-postfix  ::= expr-postfix <a href="#operator">operator-postfix</a>
    expr-postfix  ::= <a href="#expr-new">expr-new</a>
    expr-postfix  ::= <a href="#expr-dot">expr-dot</a>
    expr-postfix  ::= <a href="#expr-metatype">expr-metatype</a>
    expr-postfix  ::= <a href="#expr-init">expr-init</a>
    expr-postfix  ::= <a href="#expr-subscript">expr-subscript</a>
    expr-postfix  ::= <a href="#expr-call">expr-call</a>
    expr-postfix  ::= <a href="#expr-optional">expr-optional</a>
    expr-force-value  ::= <a href="#expr-force-value">expr-force-value</a>

  </pre>

  <p>At the top level of the expression grammar, expressions are a
    sequence of unary expressions joined by binary operators.  When
    parsing an expr, a binary operator immediately following an
    expr-unary continues the expression, and the program is ill-formed
    if it is not then followed by another expr-unary.  This resolves
    an ambiguity which could otherwise arise in statement contexts due
    to semicolon elision.</p>

  <pre class="example">
    5 !- +~123 -+- ~+6
    (foo)(())
    bar(49+1)
    baz()
  </pre>

  <p>A unary or binary expression may optionally be followed by a
    <a href="#expr-cast">cast operator</a>.

  <!-- ===================================================================== -->
  <h3 id="expr-binary">Binary Operators</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    Should this use the expr-identifier production to allow qualified
    identifiers?  This would allow "foo Swift.+ bar".  Is ADL or something
    like it enough?<br><br>
  </div>

  <pre class="grammar">
    expr-binary ::= op-binary-or-ternary <a href="#expr-unary">expr-unary</a> expr-cast?

    op-binary-or-ternary ::= <a href="#operator">operator-binary</a>
    op-binary-or-ternary ::= '='
    op-binary-or-ternary ::= '?'-infix <a href="#expr">expr-sequence</a> ':'

    expr-cast ::= 'is' <a href="#type">type</a>
    expr-cast ::= 'as' <a href="#type">type</a>
  </pre>

  <p>Infix binary expressions are not formed during parsing.  Instead,
  they are formed after name resolution by building a tree from an
  operator-delimited sequence of unary expressions.  Precedence and
  associativity are determined by the <a href="#attribute-infix">infix</a>
  attribute on the resolved names, which must fully agree.</p>

  <p>If an operator is used as a binary operator, but name resolution
  does not find at least one function of binary operator type, the
  expression is ill-formed.</p>

  <p>A simple example is:</p>

  <pre class="example">
    4 + 5 * 123
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-binary-builtin">Builtin Binary Operators</h3>
  <!-- ===================================================================== -->

  <p>In addition to user-defined operators, a handful of builtin operators are
  defined that parse inside binary expressions with predefined precedence and
  associativity.

  <h4 id="expr-assign">Assignment operator</h4>

  <p>The assignment operator <tt>a = b</tt> updates the value of <tt>a</tt> with
  the value of <tt>b</tt>. Its precedence is hardcoded as if declared as
  follows:</p>

  <pre class="example">
    // Not valid Swift code
    infix operator = {
      precedence 90
      associativity right
    }
  </pre>

  The left-hand operand must be an lvalue, or a tuple of lvalues. Assigning to
  a tuple of lvalues performs destructuring reassignment.

  <pre class="example">
    var (a, b) = (1, 2)

    // Swap two values.
    (a, b) = (b, a)

    // Reassign two values.
    (a, b) = (11, 22)

    // Reassign two values by destructuring a tuple.
    var tuple = (111, 222)
    (a, b) = tuple
  </pre>

  <p>An assignment expression evaluates to void. Unlike C, productions such as
  these are invalid:</p>

  <pre class="example">
    // Error: x = y doesn't return Bool
    if x = y { }

    // Error: (y = z) doesn't return Int
    var x, y, z : Int
    x = y = z
  </pre>

  <h4 id="expr-ternary">Ternary operator</h4>

  <p>The ternary operator <tt>a ? b : c</tt> conditionally evaluates its middle
  or right operand based on the value of its left operand. Its precedence is
  hardcoded as if the middle <tt>? b :</tt> subexpression were a binary operator
  declared as follows:</p>

  <pre class="example">
    // Not valid Swift code
    infix operator ?...: {
      precedence 100
      associativity right
    }
  </pre>

  <p>The subexpression to the left of the
  '?' is evaluated, and is converted to 'Bool' using the result's
  'boolValue' property if it is not already 'Bool'. If the condition is
  true, the subexpression to the right of '?' is evaluated, and its result
  becomes the result of the expression. If the
  condition is false, the subexpression to the right of ':' is evaluated, and
  its result becomes the result of the expression. Only one of the
  '?' or ':' subexpressions will be evaluated. The results of the
  '?' and ':' subexpressions must be implicitly convertible to a common type,
  which becomes the type of the ternary expression.

  <pre class="example">
    x += b ? y : z
    x += a ? b ? y : z : w

    for i in 1...101 {
      print(i % 15      ? "fizzbuzz"
            : i %  3 == 0 ? "fizz"
            : i %  5 == 0 ? "buzz"
            : "\(i)")
    }
  </pre>

  <h4 id="expr-cast">Cast operators</h4>

  <p>Cast expressions influence the types of their subexpressions. They can appear
  at the end of a binary operator sequence; their left operand is parsed as if
  the cast operators were declared as follows:</p>

  <pre class="example">
    // Not valid Swift code
    infix operator as {
      precedence 95
      associativity none
    }
  </pre>

  <p>The right operand of all operators is parsed as a type.</p>

  <ul>
  <li><tt>x as T</tt> will try to cast the value of the expression
    <tt>x</tt> to the type <tt>T</tt>. If the type of <tt>x</tt> is
    implicitly convertible to <tt>T</tt>, the conversion is performed
    and the result of the expression is of type <tt>T</tt>. Otherwise,
    the result of the expression is of type <tt>T?</tt>. In this case,
    the type of the operand is checked at runtime, and if it is
    castable to <tt>T</tt>, the <tt>Optional</tt> result contains
    the result of the cast. If the cast fails, the result contains
    <tt>.None</tt>. The latter is only permissible when
    <tt>T</tt> is a subtype of the compile-time type of <tt>x</tt>.
    An example:

  <pre class="example">
    var b: B = D()
    var d: D? = b as D
    var b2 = d! as B
  </pre>

  <li><tt>x is T</tt> will query the type of the value of <tt>x</tt> at runtime.
    <tt>T</tt> must be a subtype of the compile-time type of <tt>x</tt>.
    If the runtime value of <tt>x</tt> is <tt>T</tt>, the <tt>is</tt> expression
    evaluates to true; otherwise, it evaluates to false.

  <pre class="example">
    if b is D {
      var d = (b as D)!
    }
  </pre>
  </ul>

  <p><tt>as</tt> and <tt>is</tt> both parse a type for their
  right-hand argument. They must be parenthesized if followed by subsequent
  operators:

  <pre class="example">
    (b as D)?.derivedMethod()
    ((B as D) as D2)
    (b is D) ? (b as D)! : D()
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-unary">Unary Operators</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-unary   ::= <a href="#operator">operator-prefix</a>* <a href="#expr">expr-postfix</a>
  </pre>

  <p>If an operator is used as a unary operator, but name resolution
  does not find at least one function that takes a single argument, the
  expression is ill-formed.</p>

  <p>Simple examples:</p>

  <pre class="example">
    i = -j
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-literal">Literals</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    The type of a literal is inferred from its context, to allow things like "4"
    to be compatible with any width integer type without 'promotion' rules or
    casting.  In ambiguous cases like "var x = 4", the literals are forced to
    a default type specified by the standard library.
  </div>

  <pre class="grammar">
    expr-literal ::= <a href="#integer_literal">integer_literal</a>
    expr-literal ::= <a href="#floating_literal">floating_literal</a>
    expr-literal ::= <a href="#string_literal">string_literal</a>
    expr-literal ::= expr-array
    expr-literal ::= expr-dictionary
    expr-literal ::= '__FILE__'
    expr-literal ::= '__LINE__'
    expr-literal ::= '__COLUMN__'

    expr-array ::= '[' expr (',' expr)* ','? ']'
    expr-array ::= '[' ']'

    expr-dictionary ::= '[' expr ':' expr (',' expr ':' expr)* ','? ']'
    expr-dictionary ::= '[' ':' ']'
  </pre>

  <p>Numeric literals are either integer, floating point, character, or string
     depending on its lexical form.  The type of the literal is inferred
     based on its context.  If there is no contextual type information for an
     expression, all unresolved types are inferred to 'IntegerLiteralType'
     type, to 'FloatLiteralType', and to
     'StringLiteralType', respectively.
     If a literal is used and these types are not defined, then the code is
     malformed.</p>

  <p>A literal is compatible with its inferred type if that type implements an
     informal protocol required by literals.  This informal protocol requires
     that the type have an unambiguous "type" function defined whose
     result type is the same as the inferred type, and that takes a single
     argument that is either itself literal compatible, or is a <a
     href="#builtin">builtin</a> integer type.</p>

  <p>The '<tt>__FILE__</tt>', '<tt>__LINE__</tt>', and '<tt>__COLUMN__</tt>'
     magic identifiers expand to a literal representation of their position in
     the source code. '<tt>__FILE__</tt>' expands to a string literal;
     '<tt>__LINE__</tt>' and '<tt>__COLUMN__</tt>' each expand to an integer
     literal.</p>

  <pre class="example">
    <i>// File foo.swift</i>

    var file = __FILE__  <i>// file : String = "foo.swift"</i>
    var line = __LINE__  <i>// line : Int = 4</i>
    var col = __COLUMN__ <i>// column : Int = 11</i>
  </pre>

  <p>If '<tt>__FILE__</tt>', '<tt>__LINE__</tt>', and/or '<tt>__COLUMN__</tt>'
    are used as default argument values in a function declaration, they
    instead expand to the source location of each function call that
    instantiates the default argument.</p>

  <pre class="example">
    func log(_ message:String,
             file:String = __FILE__,
             line:Int = __LINE__) {
      print("\(file):\(line): \(message)")
    }

    log("Orders received")
    doIt()
    log("Job's finished")
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-identifier">Identifiers</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-identifier ::= <a href="#identifier">identifier</a> <a href="#generic-args">generic-args</a>?
  </pre>

  <p>A raw identifier refers to a value found via <a
     href="#namebind_value_lookup_unqual">unqualified value lookup</a>, and has
     the type of the declaration returned by name lookup and overload
     resolution.  Value declarations are installed with <a
     href="#decl-var">var</a> and the syntactic sugar forms like <a
     href="decl-func">func</a> declarations.</p>

  <p>If an identifier refers to a generic type, an instance of that generic may
    be referenced by following the identifier with a list of type parameters
    enclosed in angle brackets <tt>&lt;&gt;</tt>:</p>

  <pre class="example">
    <i>// A generic struct.</i>
    struct Dict&lt;K,V&gt; {
      init() {}
      static func fromKeysAndValues(_ keys:K[], values:T[]) -&gt; Dict&lt;K,V&gt; {}
    }

    <i>// Construct an instance of the generic struct.</i>
    var foo = Dict&lt;String, Int&gt;()
    <i>// Invoke a type method of an instance of the generic struct.</i>
    var bar = Dict&lt;String, Int&gt;.fromKeysAndValues(
      ["zim", "zang", "zung"],
      [ 123,    456,    789 ])
  </pre>

  <h4 id="expr-generic-disambiguation">Generic disambiguation</h4>

  <p>Note that <tt>&lt;</tt> and <tt>&gt;</tt> are used as both angle brackets in
  <a href="#expr-identifier">generic identifiers</a> and as characters in
  <a href="#expr-binary">binary operator</a> names. Because of this, there are
  potential parsing ambiguities. Swift uses a context-free heuristic to
  determine whether to parse an expression involving <tt>&lt;</tt> and <tt>&gt;</tt>
  as a generic parameter list or a binary operator:

  <ul>
  <li>When an <a href="#identifier">identifier</a> is followed by <tt>&lt;</tt>,
  Swift attempts to parse starting from the <tt>&lt;</tt> as a
  <a href="#type-identifier">generic parameter list</a>.
  <li>If it succeeds in parsing a generic parameter list, it looks at the
  token after the closing <tt>&gt;</tt>. If it sees one of the following tokens:
  <blockquote>
    <tt>( [ { } ] ) . , ;</tt>
  </blockquote>
  then the expression is parsed as a generic parameter list.
  <li>If Swift cannot parse a generic parameter list after the <tt>&lt;</tt>,
  or the matching <tt>&gt;</tt> is not followed by one of the above tokens,
  the <tt>&lt;</tt> is parsed as an operator character.
  </ul>

  <p>These rules assume that, in most cases, generic type names will be used
  in constructor expressions as in <tt>Foo&lt;T&gt;(x)</tt> or to access type
  members as in <tt>Foo&lt;T&gt;.bar()</tt>. Referring to a generic metatype as a
  value in an expression may require parentheses around the type name.

  <pre class="example">
    <i>// An operator that operates on metatypes.</i>
    infix func +-+ &lt;T, U&gt;(t:T.Type, u:U.Type) -&gt; Foo { }

    var foo = (Dict&lt;String, Int&gt;) +-+ (Array&lt;UnicodeScalar&gt;)
    print(foo)
  </pre>

  <p>On the other hand, some expressions involving <tt>&lt;</tt> and
  <tt>&gt;</tt> operators may misparse as generic arguments as well. These
  can also be corrected by adding or removing parentheses.

  <pre class="example">
    func foo(_ x:Bool, y:Bool)
    var a,b,c,d,e : Int

    foo(a &lt; b, c &gt; (d + e)) // ERROR: Misparses as (a&lt;b,c&gt;)(d + e)
    foo((a &lt; b), c &gt; (d + e)) // Force parsing as (a &lt; b), (c &gt; (d + e))
    foo(a &lt; b, c &gt; d + e) // Also parses as (a &lt; b), (c &gt; (d + e))
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-super">Super</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-super ::= expr-super-method
    expr-super ::= expr-super-subscript
    expr-super ::= expr-super-constructor

    expr-super-method ::= 'super' '.' <a href="#expr-identifier">expr-identifier</a>
    expr-super-subscript ::= 'super' '[' <a href="#expr">expr</a> ']'
    expr-super-constructor ::= 'super' '.' 'init'
  </pre>

  <p>The keyword <tt>super</tt> is used to refer to superclass members from
  a subclass method. This can be used to access members of a superclass
  overridden by the subclass. The following forms are allowed:

  <ul>
  <li>A superclass property or method can be accessed with the form
    <tt>super.name</tt>.</li>
  <li>A superclass subscript accessor can be accessed with the form
    <tt>super[index]</tt>.</li>
  <li>Within a constructor, a superclass constructor can be accessed with the
    form <tt>super.init</tt>.</li>
  </ul>

  <p><tt>super</tt> expressions are invalid outside of a subclass method.
  <tt>super.init</tt> is invalid outside of a subclass constructor.
  <tt>super.init</tt> furthermore may only be called once per derived
  constructor, and must be called before the derived constructor accesses
  <tt>self</tt> or any instance variables.
  </p>

  <!-- ===================================================================== -->
  <h3 id="expr-closure">Closure Expression</h3>
  <!-- ===================================================================== -->
  <pre class="grammar">
    expr-closure ::= '{' closure-signature? <a href="#brace-item-list">brace-item-list</a> '}'

    closure-signature ::= <a href="#pattern-tuple">pattern-tuple</a> <a href="#func-signature">func-signature-result</a>? 'in'
    closure-signature ::= <a href="#identifier">identifier</a> (',' <a href="#identifier">identifier</a>*) <a href="#func-signature">func-signature-result</a>? 'in'

  </pre>

  <p>A closure defines an anonymous function as an expression. Like a
  <a href="#decl-func">func</a> declaration, a closure has parameters,
  a return type, and some number of statements that are executed when
  the closure is called. Like local functions, closures can capture
  values from its enclosing function and closure scopes. Closures are
  often used in lieu of local functions when the function name would
  only be used once, to be called by some other function. As a syntax
  optimization, when the closure contains only a single expression, it's
  value is used as the result of the closure. Thus, the closure <code>{
  5 }</code> is equivalent to <code>{ return 5 }</code>.</p>

  <p>Unlike <a href="#decl-func">func</a>
  declarations, the return type, parameter types, and even the <a
   href="#expr-anon-closure-arg">names of parameters</a> can be
  omitted from the definition of the closure, making it a concise
  syntax for small closures. In such cases, the context in which the
  closure is used must provide information about the parameter and
  return types. In the special case where the closure consists of only
  a single expression, that expression participates in the
  type checking of its context. </p>

  <pre class="example">
    <i>// Takes a closure that it calls to determine an ordering relation.</i>
    func magic(_ val : Int, predicate : (a : Int, b : Int) -> Bool)

    func f() {
      <i>// Compare one way.  Closure is inferred to return Bool and take two ints</i>
      <i>// from the argument context.  This same information infers that $0 and $1</i>
      <i>// both have type 'Int'.</i>
      magic(42, { $0 &lt; $1 })

      <i>// Compare the other way.</i>
      magic(42, { $1 &lt; $0 })

      <i>// Provide parameter names, but infer the types.</i>
      magic(42, { x, y in y &lt; x })

      <i>// Provide parameter names and types.</i>
      magic(42, { (x : Int, y : Int) in y &lt; x })

      <i>// Provide parameter names and types, and return type, with multiple statements.</i>
      magic(42, { (x : Int, y : Int) -> Bool in
        print("Comparing \(x) to \(y).\n")
        return y &lt; x
      })

      <i>// Error, not enough context to infer the type of $0.</i>
      var x = { $0 }
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-anon-closure-arg">Anonymous Closure Arguments</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-anon-closure-arg ::= <a href="#dollarident">dollarident</a>
  </pre>

  <p>A use of an identifier whose name fits the "$[0-9]+" regular
  expression is a reference to an anonymous closure argument that is formed when
  the containing expression is <a href="#typecheck_anon">coerced into a closure
  context</a>.  All other dollar identifiers are invalid.</p>

  <p>This can only be used in the body of a closure (<a
     href="#expr-closure">expr-closure</a>) that does not have explicitly-specified parameters.
  </p>


  <!-- ===================================================================== -->
  <h3 id="expr-delayed-identifier">Delayed Identifier Resolution</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    The ".bar" syntax was picked because it is related to the syntax of a fully
    qualified "foo.bar" reference.
  </div>

  <pre class="grammar">
    expr-delayed-identifier ::= '.' <a href="#identifier">identifier</a> <a href="#expr-paren">expr-paren</a>?
  </pre>

  <p>A delayed identifier expression refers to a case of an <a
    href="type-enum">enum</a> type or a type member of a nominal
    type, without knowing which type it is referring to.  The
    expression is resolved to a member of a concrete type through
    context-sensitive type inference. When it is followed by an <a
    href="#expr-paren">expr-paren</a>, the member must either be an
    enum case that carries a value or a (type) member function.</p>

  <pre class="example">
    enum Direction { case Up, Down }
    func search(_ val : Int, direction : Direction)

    func f() {
      search(42, .Up)
      search(17, .Down)
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-paren">Parenthesized Expressions</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-paren      ::= '(' ')'
    expr-paren      ::= '(' expr-paren-element (',' expr-paren-element)* ')'
    expr-paren-element ::= (<a href="#identifier">identifier</a> ':')? <a href="#expr">expr</a>
  </pre>

  <p>Parentheses expressions contain an (optionally empty) list of optionally
     named values.  Parentheses in an expression context denote one of two
     things: 1) grouping parentheses, or 2) a tuple literal.</p>

  <p>Grouping parentheses occur when there is exactly one value in the list and
     that value does not have a name.  In this case, the type of the parenthesis
     expression is the type of the single value.</p>

  <p>All other cases are tuple literals.  The type of the expression is a tuple
     type whose elements and order match that of the initializer.  If there are
     any named elements, those elements become names for the tuple type.  A
     parenthesis expression with no value has a type of the empty tuple.
  </p>

  <p>Some examples:</p>

  <pre class="example">
    <i>// Simple grouping parenthesis.</i>
    var a = (4)             <i>// Type = Int</i>
    var b = (4+a)           <i>// Type = Int</i>

    <i>// Tuple literals.</i>
    var c = ()               <i>// Type = ()</i>
    var d = (4, 5)           <i>// Type = (Int, Int)</i>
    var e = (c, d)           <i>// Type = ((), (Int, Int))</i>

    var f = (x : 4, y : 5)   <i>// Type = (x : Int, y : Int)</i>
    var g = (4, y : 5, 6)    <i>// Type = (Int, y : Int, Int)</i>

    <i>// Named arguments to functions.</i>
    func foo(_ a : Int, b : Int)
    foo(b = 4, a = 1)
  </pre>


  <!-- ===================================================================== -->
  <h3 id="expr-dot">Dot Expressions</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-dot ::= <a href="#expr">expr-postfix</a> '.' <a href="#integer_literal">integer_literal</a>
  </pre>

  <p>If the base expression has <a href="#type-tuple">tuple type</a>, then the
  magic identifier "[0-9]+" accesses the specified anonymous member of the
  tuple.  Otherwise, this form is invalid.</p>

  <pre class="grammar">
    expr-dot ::= <a href="#expr">expr-postfix</a> '.' <a href="#expr-identifier">expr-identifier</a>
  </pre>

  <p>If the base expression has <a href="#type-tuple">tuple type</a> and if the
  identifier is the name of a field in the tuple, then this is a reference to
  the specified field.</p>

  <p>Otherwise, <a href="#namebind_value_lookup_dot">dot name lookup</a> is
  performed, and this expression is treated as function application.  This
  allows looking up members in modules, metatypes, etc.</p>

  <!-- ===================================================================== -->
  <h3 id="expr-init">Initializer Expressions</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-init ::= <a href="#expr">expr-postfix</a> '.' 'init'
  </pre>

  <p>An initializer reference refers to a set of initializers of the
  base expression. The base expression must be the <code>self</code>
  parameter of an initializer, which is used to delegate the
  initialization of the object to another initializer.</p>

  <pre class="example">
    class X {
      init() {
        self.init(5) // delegate to initializer below
      }

      init(value: Int) { /* ... */ }
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-subscript">Subscript Expressions</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    There is no "built-in" semantics for subscripting. Rather, all
    subscripting semantics is implemented via subscript declarations
    in the library.

    <br/>We require that the '[' not be the first token on a line, so that
    a statement can begin with an array expression.
  </div>

  <pre class="grammar">
    expr-subscript ::= <a href="#expr">expr-postfix</a> '[' <a href="#expr">expr</a> ']'
  </pre>

 <p>A subscript expression invokes a <a
  href="#decl-subscript">subscript getter or setter</a> on the type
  of the <tt>expr-postfix</tt>. The <tt>expr</tt> is used as the
  subscript argument, which will be provided to either the getter or
  setter depending on whether the subscript expression is used as an
  rvalue (reading) or lvalue (writing), respectively. A subscript
  expression that resolves to a subscript declaration with no setter
  cannot be modified.</p>

  <!-- ===================================================================== -->
  <h3 id="expr-new">New Expressions</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    It's not really clear what the behavior of multiple bounds should be.

    <br/><br/>We should probably allow an initializer. The semantics would be
    to evaluate that constructor for each element constructed.
  </div>

  <pre class="grammar">
    expr-new        ::= 'new' <a href="#type">type-identifier</a> expr-new-bounds

    expr-new-bounds ::= expr-new-bound
    expr-new-bounds ::= expr-new-bounds expr-new-bound
    expr-new-bound  ::= '[' <a href="#expr">expr?</a> ']'
  </pre>

  <p>Allocates and initializes a new array of objects. The first clause must
    be an expression; subsequent bounds, if present, must be constant under
    the <a href="#type-array">usual rules for array types</a>. The opening
    square bracket must be on the same line as the type name.</p>

  <!-- ===================================================================== -->
  <h3 id="expr-call">Function Application</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-call ::= <a href="#expr">expr-postfix</a> <a href="#expr-paren">expr-paren</a>
  </pre>

  <p>The leading <tt>'('</tt> of the <tt>expr-paren</tt> must not be
  the first token on a line.  This greatly reduces the likelihood of
  confusion from semicolon elision, without requiring feedback from
  the typechecker or more aggressive whitespace sensitivity.</p>

  <p>If the <tt>expr-postfix</tt> refers to a (possibly
  parenthesized) name of a type, the <tt>expr-paren</tt> is first
  coerced to the type named by <tt>expr-postfix</tt>. If that coercion
  fails, then the <tt>expr-postfix</tt> refers to the set of
  constructors for that type.</p>

 <p>Simple examples:</p>

  <pre class="example">
    <i>// Application of an empty tuple to the function f.</i>
    f()
    <i>// Application of 4 to the function f.</i>
    g(4)

    <i>// Application of 4 to the function returned by h().</i>
    var h : (Int) -&gt; (Int) -&gt; Int
    ...
    h()(4)

    <i>// Two separate statements</i>
    i()
    (j &lt;+ 2)()
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-trailing-closure">Trailing Closures</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    It is possible to model trailing closures as simply another way to
    perform a function call, forgoing the syntactic transformation for
    <a href="#expr-call">expr-call</a>, if functions meant to be used
    with trailing closures are written as curried functions, e.g.,
    <pre>
func map<T, U>(_ array : T[])(fn : (T) -> U) -> U[] { ... }
    </pre>
    There are two problems with this (admittedly simpler) design.
    First, functions imported from C, C++, and Objective-C won't ever
    be written in this curried syntax, so we would have to implement
    redundant entry points to enable this syntax. Second, this design
    forces the idea of currying front and center for Swift programmers
    who otherwise wouldn't care, for mostly theoretical reasons.
  </div>

  <pre class="grammar">
    expr-trailing-closure ::= <a href="#expr">expr-postfix</a> <a href="#expr-closure">expr-closure</a>+
  </pre>

  <p>A postfix expression followed by a closure will be invoked with
  the closure as its argument. This syntax is referred to as a
  "trailing" closure, because the closure itself is outside the
  parentheses used to call the expression. Trailing closures are
  syntactic sugar that eliminates the awkwardness of closing a
  function call with "})", where the "}" ends the closure and the ")"
  ends the call.</p>

  <p>Trailing closures use a simple syntactic translation, making them
  purely syntactic sugar. If the postfix expression preceding the
  trailing closure is an <a href="#expr-call">expr-call</a>, the
  closure is added to the end of the <a
  href="#expr-paren">expr-paren</a> of that call. Otherwise, the
  postfix expression is (implicitly) called with the trailing closure
  as its only argument.</p>

  <pre class="example">
  dispatch_async(q) {
    print("Whenever you get around to it\n")
  }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="expr-optional">Optional Chaining</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-optional ::= <a href="#expr-postfix">expr-postfix</a> '?'-postfix
  </pre>

  <p>The optional-chaining operator provides a convenient syntax for
    dereferencing, calling, or subscripting optional values.</p>

  <p>Informally, the operator attempts to strip one level
    of <code>Optional</code> from its operand, and if that fails, all
    the following postfix operators are skipped and just evaluate
    to <code>None</code>.</p>

  <p>More formally:
    <ul>
      <li>The operand must be convertible to type <code>T?</code> for
        some type <code>T</code>.  As a special rule, the expression
        is ill-formed if the operand is converted to optional type by
        the implicit conversion from <code>T</code>
        to <code>T?</code>.</li>

      <li>The expression itself then has type <code>T</code> (and is
      an r-value).</li>

      <li>A <code>postfix-expression</code> <i>E1</i> is said to
        <i>directly chain</i> to a <code>postfix-expression</code>
        <i>E2</i> if <i>E1</i> is syntactically
        the <code>postfix-expression</code> base of <code>E2</code>;
        note that this does not include any syntactic nesting,
        e.g. via parentheses. <i>E1</i> <i>chains</i> to <i>E2</i>
        if they are the same expression or <i>E1</i> directly chains
        to an expression which chains to <i>E2</i>.  This relation has
        a maximum, called the <i>largest chained expression</i>.</li>

      <li>The largest chained expression of an <code>expr-optional</code>
        must be convertible to an r-value of type <code>U?</code> for
        some type <code>U</code>.  Note that a single expression may
        be the largest chained expression of multiple
        <code>expr-optional</code>s.</li>

      <li>If the operand evaluates to <code>Some(x) : T?</code> for some
        value <code>x : T</code>, the expression yields
        <code>x</code>.</li>

      <li>If the operand evaluates to <code>None : T?</code>, evaluation
        of all the chained expressions immediately terminates, and
        the largest chained expression yields the value
        <code>None : U?</code>.</li>
    </ul></p>

    <!-- ===================================================================== -->
  <h3 id="expr-force-value">Forcing an expression's value</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    expr-force-value ::= <a href="#expr-postfix">expr-postfix</a> '!'
  </pre>

  <p>The postfix '!' forces an optional value to its stored value
  (i.e., the <code>x</code> in <code>.Some(x)</code>), failing at
  runtime if the optional is <code>.None</code>.

  <!-- ********************************************************************* -->
  <h2 id = "stmt">Statements</h2>
  <!-- ********************************************************************* -->

  <div class="commentary">
    Statements can only exist in contexts that are themselves a stmt.
    Statements have no type, they just induce control flow changes.  We choose
    to use constructs that will be familiar to a broad range of C/Java
    programmers.
  </div>

  <pre class="grammar">
    stmt ::= <a href="#stmt-semicolon">stmt-semicolon</a>
    stmt ::= <a href="#stmt-if">stmt-if</a>
    stmt ::= <a href="#stmt-while">stmt-while</a>
    stmt ::= <a href="#stmt-repeat-while">stmt-repeat-while</a>
    stmt ::= <a href="#stmt-for-c-style">stmt-for-c-style</a>
    stmt ::= <a href="#stmt-for-each">stmt-for-each</a>
    stmt ::= <a href="#stmt-switch">stmt-switch</a>
    stmt ::= stmt-control-transfer

    stmt-control-transfer ::= <a href="#stmt-return">stmt-return</a>
    stmt-control-transfer ::= <a href="#stmt-break">stmt-break</a>
    stmt-control-transfer ::= <a href="#stmt-continue">stmt-continue</a>
    stmt-control-transfer ::= <a href="#stmt-fallthrough">stmt-fallthrough</a>
  </pre>

  <p>Statements provide the control flow constructs of function bodies and
  top-level code.</p>

  <pre class="example">
    <i>// A function with some statements.</i>
    func fib(_ v : Int) -&gt; Int {
      if v &lt; 2 {
        return v
      }
      return fib(v-1)+fib(v-2)
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="stmt-semicolon">Semicolon Statement</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    Allowing semicolons as statements causes us to allow semicolons as statement
    separators as well.  This, in turn, means that we don't reject code that has
    semicolons after each statement, which will be common when people first
    start getting used to Swift.
  </div>

  <pre class="grammar">
    stmt-semicolon ::= ';'
  </pre>

  <p>The semicolon statement has no effect.</p>

  <!-- ===================================================================== -->
  <h3 id="stmt-return">'return' Statement</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    stmt-return ::= 'return' <a href="#expr">expr</a>
    stmt-return ::= 'return'
  </pre>

  <p>The return statement sets the return value of the current <a
     href="#decl-func">func declaration</a> or <a href="#expr-closure">closure
     expression</a> and transfers control out of the function.  It sets the
     return value by converting the specified expression result (or '()' if
     none is specified) to the return type of the 'func'.
  </p>

  <p>The stmt-return grammar is ambiguous: "{ return 4 }" could be parsed as
     {"return" "4"} or as a single statement.  Ambiguity here is resolved toward
     the first production, because control flow can't transfer to an
     subexpression.</p>

  <!-- 'break' and 'continue' converted to ReST. -->

  <!-- ===================================================================== -->
  <h3 id="stmt-if">'if' Statement</h3>
  <!-- ===================================================================== -->

  <div class="commentary">
    We require braces around the body of an 'if' for two reasons: first, it
    eliminates the need for parentheses around the condition by making them
    visually distinctive.  Second, it will eliminate all the dithering about
    whether and when people should, or should not, use braces for if bodies.
  </div>

  <pre class="grammar">
    stmt-if      ::= 'if' <a href="#expr">expr-basic</a> <a href="#brace-item-list">brace-item-list</a> stmt-if-else?
    stmt-if-else ::= 'else' <a href="#brace-item-list">brace-item-list</a>
    stmt-if-else ::= 'else' stmt-if
  </pre>

  <p>'if' statements provide a simple control transfer operations that evaluates
  the condition, gets the 'boolValue' property of the result if the result
  not a 'Bool', then determines the direction of the branch based on the result.
  (Internally, the standard library type 'Bool' has a boolValue property that
  returns a 'Builtin.Int1'.)  It is an error if the type of the expression is
  context-dependent or some non-Bool type.
  </p>

  <p>Some examples include:</p>

  <pre class="example">
    if true {
      /*...*/
    }

    if X == 4 {
    } else {
    }

    if X == 4 {
    } else if X == 5 {
    } else {
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="stmt-while">'while' Statement</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    stmt-while ::= 'while' <a href="#expr">expr-basic</a> <a href="#brace-item-list">brace-item-list</a>
  </pre>

  <p>'while' statements provide simple loop construct which (on each iteration
  of the loop) evaluates the condition, gets the 'boolValue' property of
  the result if the result not a 'Bool', then determines whether to keep
  looping. (Internally, the standard library type 'Bool' has a boolValue
  property that yields a 'Builtin.Int1'.)  It is an error if the type of
  the expression is context-dependent or some non-Bool type.
  </p>

  <p>Some examples include:</p>

  <pre class="example">
    while true {
      /*...*/
    }

    while X == 4 {
      X = 3
    }
  </pre>

  <!-- ===================================================================== -->
  <h3 id="stmt-repeat-while">'repeat-while' Statement</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    stmt-repeat-while ::= 'repeat' <a href="#brace-item-list">brace-item-list</a> 'while' '<a href="#expr">expr</a>
  </pre>

  <p>'repeat-while' statements provide simple loop construct which (on each
  iteration of the loop) evaluates the body, then evaluates the condition,
  getting the 'boolValue' property of the result if the result not a 'Bool',
  then determines whether to keep looping. (Internally, the standard library
  type 'Bool' has a boolValue property that yields a 'Builtin.Int1').  It is
  an error if the type of the expression is context-dependent or some non-Bool
  type.
  </p>

  <p>Some examples include:</p>

  <pre class="example">
    repeat {
      /*...*/
    } while true

    repeat {
      X = 3
    } while X == 4
  </pre>

  <!-- ===================================================================== -->
  <h3 id="stmt-for-c-style">C-Style 'for' Statement</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    stmt-for-c-style    ::= 'for'     stmt-for-c-style-init? ';' <a href="#expr">expr</a>? ';' stmt-for-c-style-inc? <a href="#brace-item-list">brace-item-list</a>
    stmt-for-c-style    ::= 'for' '(' stmt-for-c-style-init? ';' <a href="#expr">expr</a>? ';' stmt-for-c-style-inc? ')' <a href="#brace-item-list">brace-item-list</a>
    stmt-for-c-style-init ::= <a href="#decl-var">decl-var</a>
    stmt-for-c-style-init ::= expr (',' expr)*
    stmt-for-c-style-inc ::= expr-basic (',' expr-basic)*
  </pre>

  <p>C-Style 'for' statements provide simple loop construct which evaluates the
  first part (the initializer) before entering the loop, then evaluates the
  second condition as a logic value to determines whether to keep looping.
  The third condition is executed at the end of the loop.  All three are
  evaluated in a new scope that surrounds the for statement.
  </p>

  <p>Some examples include:</p>

  <pre class="example">
    for i = 0; i != 10; ++i {
      /*...*/
    }

    for (i = 0; i != 10; ++i) {
      /*...*/
    }

    for var (i,j) = (0,1); i != 10; ++i {
      /*...*/
    }
</pre>

  <!-- ===================================================================== -->
  <h3 id="stmt-for-each">'for-each' Statement</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    stmt-for-each ::= 'for' <a href="#pattern">pattern</a> 'in' <a href="#expr">expr-basic</a> <a href="#brace-item-list">brace-item-list</a>
  </pre>

  <p>Enumerator-based 'for' statements provide enumeration over the values in a
 container. The <tt>expr</tt> is either a container or an enumerator; and
 respectively, it either conforms to the formal Enumeration or formal Enumerator
 protocol.

  <p>Note that each iteration of the loop declares a distinct variable for each
     variable in the pattern.  For example, in a loop like "for i in 0...10",
     if i is captured inside the loop, each iteration captures a different "i",
     so there would be a total of ten versions generated each time the loop
     runs.</p>

  <p>Some examples include:</p>

  <pre class="example">
    for i in 0...100 {
      print(String(i));
    }
 </pre>

  <!-- 'switch' and 'fallthrough' converted to ReST. -->

  <!-- ********************************************************************* -->
  <h2>Protocols</h2>
  <!-- ********************************************************************* -->

  <!-- ********************************************************************* -->
  <h2>Objects</h2>
  <!-- ********************************************************************* -->

  <!-- ********************************************************************* -->
  <h2>Generics</h2>
  <!-- ********************************************************************* -->

  <!-- ===================================================================== -->
  <h3 id="generic-params">Generic Parameters</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    generic-params ::= '<' generic-param (',' generic-param)* where-clause? '>'

    generic-param ::= identifier
    generic-param ::= identifier ':' <a href="#type-identifier">type-identifier</a>
    generic-param ::= identifier ':' <a href="#type-composition">type-composition</a>

    <a id="where-clause">where-clause</a> ::= 'where' requirement (',' requirement) *

    requirement ::= conformance-requirement
                ::= same-type-requirement

    conformance-requirement ::= <a href="#type-identifier">type-identifier</a> ':' <a href="#type-identifier">type-identifier</a>
    conformance-requirement ::= <a href="#type-identifier">type-identifier</a> ':' <a href="#type-composition">type-composition</a>

    same-type-requirement ::= <a href="#type-identifier">type-identifier</a> '==' <a href="#type-identifier">type-identifier</a>
</pre>

  <p>A generic function or type is parameterized by a given set of
  generic parameters. The generic parameters each have a name as well
  as some set of requirements that specify the capabilities that any
  corresponding generic argument might have. For example, the generic
  parameter <code>T : CustomStringConvertible</code> requires that any generic
  argument substituted for the generic parameter <code>T</code>
  conform to the protocol <code>CustomStringConvertible</code>. Similarly, a generic
  parameter <code>U : SomeClass</code> requires that any generic
  argument substituted for the generic parameter <code>U</code>
  inherit from the class <code>SomeClass</code>.

  <p>Additional requirements on generic parameters and associated types
  of generic parameters can be introduced via the "where" clause,
  which can include additional protocol-conformance requirements
  (e.g., the generic parameter list <code>&lt;T where T :
  CustomStringConvertible&gt;</code>, which is equivalent to <code>&lt;T :
  CustomStringConvertible&gt;</code>), as well as same-type requirements that
  require two types to be identical (e.g., <code>&lt;T : Collection, U
  : Collection where T.Element == U.Element&gt;</code>).

  <!-- ===================================================================== -->
  <h3 id="generic-args">Generic Arguments</h3>
  <!-- ===================================================================== -->

  <pre class="grammar">
    generic-args ::= '<' generic-arg (',' generic-arg)* '>'

    generic-arg ::= <a href="#type">type</a>
  </pre>

  <p>Generic argument lists specify the generic arguments to be provided
  to a generic type or function, which replace the generic parameters
  of that type or function to produce a specialized version of that
  type or function. For example, given a generic class:

  <pre class="example">
    class Dictionary&lt;Key : Hashable, Value&gt; { /* ... */ }
  </pre>

  <p>The type <code>Dictionary&lt;String, Int&gt;</code>, replaces the
  generic parameter <code>Key</code> with <code>String</code> and the
  generic parameter <code>Value</code> with <code>Int</code>. Each
  generic argument must satisfy all of the requirements of its
  corresponding generic parameter (e.g., <code>String</code> must
  conform to the <code>Hashable</code> protocol), and all generic
  arguments, when taken together, must satisfy the additional
  requirements specified in the <code>where</code> clause.

  <!-- ********************************************************************* -->
  <h2 id="namebind">Name Binding</h2>
  <!-- ********************************************************************* -->

  <p>Name binding in swift is performed in different ways depending on what
  language entity is being considered:</p>

  <p>Value names (for <a
  href="#decl-var">var</a> and <a href="#decl-func">func</a> declarations) and
  type names (for <a href="#decl-typealias">typealias</a>, <a
  href="#decl-enum">enum</a>, and <a href="#decl-struct">struct</a>
  declarations) follow the same <a href="#namebind_scope">scope</a> and
  <a href="#namebind_typevalue_lookup">name lookup</a> rules as described below.
  </p>

  <p>tuple element names</p>

  <p>scope within enum decls</p>

  <p>Context sensitive member references are resolved <a
    href="#typecheck_context">during type checking</a>.</p>

  <h3 id="namebind_scope">Scopes for Type and Value Names</h3>


  <h3 id="namebind_value_lookup_unqual">Name Lookup Unqualified Value Names</h3>
  <h3 id="namebind_value_lookup_dot">"dot" Name Lookup Value Names</h3>

  <h3 id="namebind_typevalue_lookup">Name Lookup for Type and Value Names</h3>

  <p>Basic algo:</p>

  <ul>
  <li>Search the current scope tree for a local name.  Local names cannot be
      forward referenced.</li>
  <li>Bind to names defined in the current component, including the current
      module. TODO: is this a good thing?  We could require explicit
      imports if we wanted to.</li>
  <li>Bind to identifiers that are imported with an import directive.  Imports
      are searched in order of introduction (top-down).  The location of an
      import directive in a file (e.g. between func decls) does not affect name
      lookup, but the order of imports w.r.t. each other does.</li>
  </ul>

  <p>Shadowing: Given a ValueDecl D1 in the current module and a ValueDecl D2
  in an imported module with the same name and a member of the same type (if
  relevant): 1. If D1 is a TypeDecl, D2 is shadowed. 2. If neither D1 nor D2
  is a TypeDecl, and they have the same type, D2 is shadowed.  If a
  declaration in an imported module is shadowed by any declaration in the
  current module, it is not found by unqualified global lookup or lookup for
  members of a type.</p>

  <h3 id="namebind_dot">Name Lookup for Dot Expressions</h3>

  <p>
  <a href="#expr-dot">Dot Expressions</a> bind to name of tuple elements.
  </p>

  <!-- ********************************************************************* -->
  <h2 id="typecheck">Type Checking</h2>
  <!-- ********************************************************************* -->

  <p>
  Binary expressions, function application, etc.
  </p>

  <h3 id="typecheck_conversions">Standard Conversions</h3>

  <!--
   Consider foo(4, 5) when foo is declared to take ((Int,Int=3), Int=6).  This
   could be parsed as either ((4,5), 6) or ((4,3),5), but the later one is
   the "right" answer.
  -->

  <h3 id="typecheck_anon">Anonymous Argument Resolution</h3>
  <h3 id="typecheck_context">Context Sensitive Type Resolution</h3>

<!-- *********************************************************************** -->
<hr>
<address>
  <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
  src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
  <a href="http://validator.w3.org/check/referer"><img
    src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>

  <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
</address>

</body>
</html>