Report a bug
If you spot a problem with this page, click here to create a Bugzilla issue.
Improve this page
Quickly fork, edit online, and submit a pull request for this page. Requires a signed-in GitHub account. This works well for small changes. If you'd like to make larger changes you may want to consider using a local clone.

Functions

Grammar

FuncDeclaration:
    StorageClassesopt BasicType FuncDeclarator FunctionBody
    AutoFuncDeclaration
AutoFuncDeclaration: StorageClasses Identifier FuncDeclaratorSuffix FunctionBody
FuncDeclarator: BasicType2opt Identifier FuncDeclaratorSuffix
FuncDeclaratorSuffix: Parameters MemberFunctionAttributesopt TemplateParameters Parameters MemberFunctionAttributesopt Constraintopt
Parameters:
    ( ParameterListopt )
ParameterList: Parameter Parameter , ParameterList ...
Parameter: InOutopt BasicType Declarator InOutopt BasicType Declarator ... InOutopt BasicType Declarator = AssignExpression InOutopt Type InOutopt Type ...
InOut: InOutX InOut InOutX
InOutX: auto TypeCtor final in lazy out ref return ref scope
FunctionAttributes: FunctionAttribute FunctionAttribute FunctionAttributes
FunctionAttribute: nothrow pure Property
MemberFunctionAttributes: MemberFunctionAttribute MemberFunctionAttribute MemberFunctionAttributes
MemberFunctionAttribute: const immutable inout return shared FunctionAttribute
FunctionBody:
    BlockStatement
    FunctionContractsopt BodyStatement
    FunctionContracts
FunctionContracts: InStatement OutStatementopt OutStatement InStatementopt
InStatement: in BlockStatement
OutStatement: out BlockStatement out ( Identifier ) BlockStatement
BodyStatement: body BlockStatement

Contracts

The in and out blocks of a function declaration specify the pre- and post-conditions of the function. They are used in Contract Programming. The code inside these blocks should not have any side-effects, including modifying function parameters and/or return values.

Function Return Values

Function return values are considered to be rvalues. This means they cannot be passed by reference to other functions.

Functions Without Bodies

Functions without bodies:

int foo();

that are not declared as abstract are expected to have their implementations elsewhere, and that implementation will be provided at the link step. This enables an implementation of a function to be completely hidden from the user of it, and the implementation may be in another language such as C, assembler, etc.

Pure Functions

Pure functions are functions that cannot directly access global or static mutable state. pure guarantees that a pure function call won't access or modify any implicit state in the program.

Unlike other functional programming languages, D's pure functions allow modification of the caller state through their mutable parameters.

pure int foo(int[] arr) { arr[] += 1; return arr.length; }
int[] a = [1, 2, 3];
foo(a);
assert(a == [2, 3, 4]);

A pure function accepting parameters with mutable indirections offers what's called "weak purity" because it can change program state transitively through its arguments. A pure function that has no parameter with mutable indirections is called "strongly pure" and fulfills the purity definition in traditional functional languages. Weakly pure functions are useful as reusable building blocks for strongly pure functions.

To control mutations, D has the immutable type qualifier. If all of a pure function's parameters are immutable or copied values without any indirections, it can guarantee that the pure function has no side effects.

To prevent mutation, D offers the immutable type qualifier. If all of a pure function's parameters are immutable or copied values without any indirections (e.g. int), the type system guarantees no side effects.

struct S { double x; }
pure int foo(immutable(int)[] arr, int num, S val)
{
    //arr[num] = 1; // compile error
    num = 2;        // has no side effect to the caller side
    val.x = 3.14;   // ditto
    return arr.length;
}

The maximum guarantee of pure is called "strong purity". It can enable optimizations based on the fact that a function is guaranteed to not mutate anything which isn't passed to it. For cases where the compiler can guarantee that a pure function cannot alter its arguments, it can enable full, functional purity (i.e. the guarantee that the function will always return the same result for the same arguments). To that end, a pure function:

This definition of mutable functions is more general than the one traditionally employed by pure functional languages because it allows a D pure function to use state mutation, as long as all state is created internally or reachable through its arguments. In particular, a pure function may allocate memory by means of e.g. new or malloc without these being special cases. A pure function is allowed to loop indefinitely or terminate the program.

As a concession to practicality, a pure function can also:

A pure function can throw exceptions.

import std.stdio;
int x;
immutable int y;
const int* pz;

pure int foo(int i,
             char* p,
             const char* q,
             immutable int* s)
{
    debug writeln("in foo()"); // ok, impure code allowed in debug statement
    x = i;   // error, modifying global state
    i = x;   // error, reading mutable global state
    i = y;   // ok, reading immutable global state
    i = *pz; // error, reading const global state
    return i;
}

An implementation may assume that a pure function that (a) accepts only parameters without mutable indirections, and (b) returns a result without mutable indirections, will have the same effect for all invocation with equivalent arguments, and is allowed to memoize the result of the function under the assumption that equivalent parameters always produce equivalent results. Such functions are termed strongly pure functions in this document. Note that a strongly pure function may still have behavior inconsistent with memoization by e.g. using casts or by changing behavior depending on the address of its parameters. An implementation is currently not required to enforce validity of memoization in all cases.

A pure function that accepts only parameters without mutable indirections and returns a result that has mutable indirections is called a pure factory function. An implementation may assume that all mutable memory returned by the call is not referenced by any other part of the program, i.e. it is newly allocated by the function. Conversely, the mutable references of the result may be assumed to not refer to any object that existed before the function call. For example:

struct List { int payload; List* next; }
pure List* make(int a, int b)
{
    auto result = new List(a, null);
    result.next = new List(b, result);
    return result;
}

Here, an implementation may assume (without having knowledge of the body of make) that all references in make's result refer to other List objects created by make, and that no other part of the program refers to any of these objects.

Any pure function that is not strongly pure cannot be assumed to be memoizable, and calls to it may not be elided even if it returns void (save for compiler optimizations that prove the function has no effect). Function calls may still be elided, or results be memoized, by means of traditional inlining and optimization techniques available for all functions.

If a strongly pure function throws an exception or an error, the assumptions related to memoization and references do not carry to the thrown exception.

Pure destructors do not benefit of special elision.

Nothrow Functions

Nothrow functions can only throw exceptions derived from class Error.

Nothrow functions are covariant with throwing ones.

Ref Functions

Ref functions allow functions to return by reference. This is analogous to ref function parameters.

ref int foo()
{
    auto p = new int;
    return *p;
}
...
foo() = 3;  // reference returns can be lvalues

Auto Functions

Auto functions have their return type inferred from any ReturnStatements in the function body.

An auto function is declared without a return type. If it does not already have a storage class, use the auto storage class.

If there are multiple ReturnStatements, the types of them must be implicitly convertible to a common type. If there are no ReturnStatements, the return type is inferred to be void.

auto foo(int x) { return x + 3; }          // inferred to be int
auto bar(int x) { return x; return 2.5; }  // inferred to be double

Auto Ref Functions

Auto ref functions infer their return type just as auto functions do. In addition, they become ref functions if all return expressions are lvalues, and it would not be a reference to a local or a parameter.

auto ref f1(int x)     { return x; }  // value return
auto ref f2()          { return 3; }  // value return
auto ref f3(ref int x) { return x; }  // ref return
auto ref f4(out int x) { return x; }  // ref return
auto ref f5() { static int x; return x; }  // ref return

The ref-ness of a function is determined from all ReturnStatements in the function body:

auto ref f1(ref int x) { return 3; return x; }  // ok, value return
auto ref f2(ref int x) { return x; return 3; }  // ok, value return
auto ref f3(ref int x, ref double y)
{
    return x; return y;
    // The return type is deduced to double, but cast(double)x is not an lvalue,
    // then become a value return.
}

Auto ref function can have explicit return type.

auto ref int (ref int x) { return x; }  // ok, ref return
auto ref int foo(double x) { return x; }   // error, cannot convert double to int

Inout Functions

Functions that deal with mutable, const, or immutable types with equanimity often need to transmit their type to the return value:

int[] f1(int[] a, int x, int y) { return a[x .. y]; }

const(int)[] f2(const(int)[] a, int x, int y) { return a[x .. y]; }

immutable(int)[] f3(immutable(int)[] a, int x, int y) { return a[x .. y]; }

The code generated by these three functions is identical. To indicate that these can be one function, the inout type constructor is employed:

inout(int)[] foo(inout(int)[] a, int x, int y) { return a[x .. y]; }

The inout forms a wildcard that stands in for any of mutable, const, immutable, inout, or inout const. When the function is called, the inout of the return type is changed to whatever the mutable, const, immutable, inout, or inout const status of the argument type to the parameter inout was.

Inout types can be implicitly converted to const or inout const, but to nothing else. Other types cannot be implicitly converted to inout. Casting to or from inout is not allowed in @safe functions.

A set of arguments to a function with inout parameters is considered a match if any inout argument types match exactly, or:

  1. No argument types are composed of inout types.
  2. A mutable, const or immutable argument type can be matched against each corresponding parameter inout type.

If such a match occurs, the inout is considered the common qualifier of the matched qualifiers. If more than two parameters exist, the common qualifier calculation is recursively applied.

Common qualifier of the two type qualifiers
mutableconstimmutableinoutinout const
mutable (= m)mcccc
const (= c)ccccc
immutable (= i)cciwcwc
inout (= w)ccwcwwc
inout const (= wc)ccwcwcwc

The inout in the return type is then rewritten to be the inout matched qualifiers:

int[] ma;
const(int)[] ca;
immutable(int)[] ia;

inout(int)[] foo(inout(int)[] a) { return a; }
void test1()
{
    // inout matches to mutable, so inout(int)[] is
    // rewritten to int[]
    int[] x = foo(ma);

    // inout matches to const, so inout(int)[] is
    // rewritten to const(int)[]
    const(int)[] y = foo(ca);

    // inout matches to immutable, so inout(int)[] is
    // rewritten to immutable(int)[]
    immutable(int)[] z = foo(ia);
}

inout(const(int))[] bar(inout(int)[] a) { return a; }
void test2()
{
    // inout matches to mutable, so inout(const(int))[] is
    // rewritten to const(int)[]
    const(int)[] x = bar(ma);

    // inout matches to const, so inout(const(int))[] is
    // rewritten to const(int)[]
    const(int)[] y = bar(ca);

    // inout matches to immutable, so inout(int)[] is
    // rewritten to immutable(int)[]
    immutable(int)[] z = bar(ia);
}

Note: Shared types are not overlooked. Shared types cannot be matched with inout.

Nested functions inside pure function are implicitly marked as pure.

pure int foo(int x, immutable int y)
{
    int bar()
    // implicitly marked as pure, to be "weak purity"
    // hidden context pointer is mutable
    {
        x = 10;     // can access states in enclosing scope
                    // through the mutable context pointer
        return x;
    }
    pragma(msg, typeof(&bar));  // int delegate() pure

    int baz() immutable
    // qualify hidden context pointer with immutable,
    // and has no other parameters, make "strong purity"
    {
        //return x; // error, cannot access mutable data
                    // through the immutable context pointer
        return y;   // ok
    }

    // can call pure nested functions
    return bar() + baz();
}

Optional Parentheses

If a function call passes no explicit argument, i.e. it would syntactically use (), then these parentheses may be omitted, similar to a getter invocation of a property function.

void foo() {}   // no arguments
void fun(int x = 10) { }
void bar(int[] arr) {}  // for UFCS

void main()
{
    foo();      // OK
    foo;        // also OK
    fun;        // OK

    int[] arr;
    arr.bar();  // OK
    arr.bar;    // also OK
}

Optional parentheses are not applied to delegates or function pointers.

void main()
{
    int function() fp;

    assert(fp == 6);    // Error, incompatible types int function() and int
    assert(*fp == 6);   // Error, incompatible types int() and int

    int delegate() dg;
    assert(dg == 6);    // Error, incompatible types int delegate() and int
}

If a function returns a delegate or function pointer, the parantheses are required if the returned value is to be called.

struct S {
    int function() callfp() { return &numfp; }
    int delegate() calldg() { return &numdg; }
    int numdg() { return 6; }
}

int numfp() { return 6; }

void main()
{
    S s;
    int function() fp;

    fp = s.callfp;
    assert(fp() == 6);

    fp = s.callfp();
    assert(fp() == 6);

    int x = s.callfp()();
    assert(x == 6);

    int delegate() dg;

    dg = s.calldg;
    assert(dg() == 6);

    dg = s.calldg();
    assert(dg() == 6);

    int y = s.calldg()();
    assert(y == 6);
}

Property Functions

WARNING: The definition and usefulness of property functions is being reviewed, and the implementation is currently incomplete. Using property functions is not recommended until the definition is more certain and implementation more mature.

Properties are functions that can be syntactically treated as if they were fields or variables. Properties can be read from or written to. A property is read by calling a method or function with no arguments; a property is written by calling a method or function with its argument being the value it is set to.

Simple getter and setter properties can be written using UFCS. These can be enhanced with the additon of the @property attribute to the function, which adds the following behaviors:

A simple property would be:

struct Foo
{
    @property int data() { return m_data; } // read property

    @property int data(int value) { return m_data = value; } // write property

  private:
    int m_data;
}

To use it:

int test()
{
    Foo f;

    f.data = 3;        // same as f.data(3);
    return f.data + 3; // same as return f.data() + 3;
}

The absence of a read method means that the property is write-only. The absence of a write method means that the property is read-only. Multiple write methods can exist; the correct one is selected using the usual function overloading rules.

In all the other respects, these methods are like any other methods. They can be static, have different linkages, have their address taken, etc.

The built in properties .sizeof, .alignof, and .mangleof may not be declared as fields or methods in structs, unions, classes or enums.

If a property function has no parameters, it works as a getter. If has exactly one parameter, it works as a setter.

Virtual Functions

Virtual functions are functions that are called indirectly through a function pointer table, called a vtbl[], rather than directly. All public and protected member functions which are non-static and are not templatized are virtual unless the compiler can determine that they will never be overridden (e.g. they are marked with final and do not override any functions in a base class), in which case, it will make them non-virtual. This results in fewer bugs caused by not declaring a function virtual and then overriding it anyway.

Member functions which are private or package are never virtual, and hence cannot be overridden.

Functions with non-D linkage cannot be virtual and hence cannot be overridden.

Member template functions cannot be virtual and hence cannot be overridden.

Functions marked as final may not be overridden in a derived class, unless they are also private. For example:

class A
{
    int def() { ... }
    final int foo() { ... }
    final private int bar() { ... }
    private int abc() { ... }
}

class B : A
{
    override int def() { ... }  // ok, overrides A.def
    override int foo() { ... }  // error, A.foo is final
    int bar() { ... }  // ok, A.bar is final private, but not virtual
    int abc() { ... }  // ok, A.abc is not virtual, B.abc is virtual
}

void test(A a)
{
    a.def();    // calls B.def
    a.foo();    // calls A.foo
    a.bar();    // calls A.bar
    a.abc();    // calls A.abc
}

void func()
{
    B b = new B();
    test(b);
}

Covariant return types are supported, which means that the overriding function in a derived class can return a type that is derived from the type returned by the overridden function:

class A { }
class B : A { }

class Foo
{
    A test() { return null; }
}

class Bar : Foo
{
    // overrides and is covariant with Foo.test()
    override B test() { return null; }
}

Virtual functions all have a hidden parameter called the this reference, which refers to the class object for which the function is called.

To avoid dynamic binding on member function call, insert base class name before the member function name. For example:

class B
{
    int foo() { return 1; }
}
class C : B
{
    override int foo() { return 2; }

    void test()
    {
        assert(B.foo() == 1);  // translated to this.B.foo(), and
                               // calls B.foo statically.
        assert(C.foo() == 2);  // calls C.foo statically, even if
                               // the actual instance of 'this' is D.
    }
}
class D : C
{
    override int foo() { return 3; }
}
void main()
{
    auto d = new D();
    assert(d.foo() == 3);    // calls D.foo
    assert(d.B.foo() == 1);  // calls B.foo
    assert(d.C.foo() == 2);  // calls C.foo
    d.test();
}

Function Inheritance and Overriding

A function in a derived class with the same name and parameter types as a function in a base class overrides that function:

class A
{
    int foo(int x) { ... }
}

class B : A
{
    override int foo(int x) { ... }
}

void test()
{
    B b = new B();
    bar(b);
}

void bar(A a)
{
    a.foo(1);   // calls B.foo(int)
}

However, when doing overload resolution, the functions in the base class are not considered:

class A
{
    int foo(int x) { ... }
    int foo(long y) { ... }
}

class B : A
{
    override int foo(long x) { ... }
}

void test()
{
    B b = new B();
    b.foo(1);  // calls B.foo(long), since A.foo(int) not considered
    A a = b;

    a.foo(1);  // issues runtime error (instead of calling A.foo(int))
}

To consider the base class's functions in the overload resolution process, use an AliasDeclaration:

class A
{
    int foo(int x) { ... }
    int foo(long y) { ... }
}

class B : A
{
    alias foo = A.foo;
    override int foo(long x) { ... }
}

void test()
{
    B b = new B();
    bar(b);
}

void bar(A a)
{
    a.foo(1);      // calls A.foo(int)
    B b = new B();
    b.foo(1);      // calls A.foo(int)
}

If such an AliasDeclaration is not used, the derived class's functions completely override all the functions of the same name in the base class, even if the types of the parameters in the base class functions are different. If, through implicit conversions to the base class, those other functions do get called, a compile-time error will be given:

class A
{
    void set(long i) { }
    void set(int i)  { }
}
class B : A
{
    void set(long i) { }
}

void foo(A a)
{
    int i;
    a.set(3);   // error, use of A.set(int) is hidden by B
                // use 'alias set = A.set;' to introduce base class overload set.
    assert(i == 1);
}

void main()
{
    foo(new B);
}

If an error occurs during the compilation of your program, the use of overloads and overrides needs to be reexamined in the relevant classes.

The compiler will not give an error if the hidden function is disjoint, as far as overloading is concerned, from all the other virtual functions is the inheritance hierarchy.

A function parameter's default value is not inherited:

class A
{
    void foo(int x = 5) { ... }
}

class B : A
{
    void foo(int x = 7) { ... }
}

class C : B
{
    void foo(int x) { ... }
}

void test()
{
    A a = new A();
    a.foo();       // calls A.foo(5)

    B b = new B();
    b.foo();       // calls B.foo(7)

    C c = new C();
    c.foo();       // error, need an argument for C.foo
}

If a derived class overrides a base class member function with diferrent FunctionAttributes, the missing attributes will be automatically compensated by the compiler.

class B
{
    void foo() pure nothrow @safe {}
}
class D : B
{
    override void foo() {}
}
void main()
{
    auto d = new D();
    pragma(msg, typeof(&d.foo));
    // prints "void delegate() pure nothrow @safe" in compile time
}

It's not allowed to mark an overridden method with the attributes @disable or deprecated. To stop the compilation or to output the deprecation message, the compiler must be able to determine the target of the call, which can't be guaranteed when it is virtual.

class B
{
    void foo() {}
}

class D : B
{
    @disable override void foo() {}
}

void main()
{
    B b = new D;
    b.foo(); // compiles and calls the most derived even if disabled.
}

Inline Functions

The compiler makes the decision whether to inline a function or not. This decision may be controlled by pragma(inline), assuming that the compiler implements it, which is not mandatory.

Note that any FunctionLiteral should be inlined when used in its declaration scope.

Function Overloading

Functions are overloaded based on how well the arguments to a function can match up with the parameters. The function with the best match is selected. The levels of matching are:

  1. no match
  2. match with implicit conversions
  3. match with conversion to const
  4. exact match

Each argument (including any this pointer) is compared against the function's corresponding parameter, to determine the match level for that argument. The match level for a function is the worst match level of each of its arguments.

Literals do not match ref or out parameters.

If two or more functions have the same match level, then partial ordering is used to try to find the best match. Partial ordering finds the most specialized function. If neither function is more specialized than the other, then it is an ambiguity error. Partial ordering is determined for functions f() and g() by taking the parameter types of f(), constructing a list of arguments by taking the default values of those types, and attempting to match them against g(). If it succeeds, then g() is at least as specialized as f(). For example:

class A { }
class B : A { }
class C : B { }
void foo(A);
void foo(B);

void test()
{
    C c;
    /* Both foo(A) and foo(B) match with implicit conversion rules.
     * Applying partial ordering rules,
     * foo(B) cannot be called with an A, and foo(A) can be called
     * with a B. Therefore, foo(B) is more specialized, and is selected.
     */
    foo(c); // calls foo(B)
}

A function with a variadic argument is considered less specialized than a function without.

Functions defined with non-D linkage cannot be overloaded. This is because the name mangling might not take the parameter types into account.

Overload Sets

Functions declared at the same scope overload against each other, and are called an Overload Set. A typical example of an overload set are functions defined at module level:

module A;
void foo() { }
void foo(long i) { }

A.foo() and A.foo(long) form an overload set. A different module can also define functions with the same name:

module B;
class C { }
void foo(C) { }
void foo(int i) { }

and A and B can be imported by a third module, C. Both overload sets, the A.foo overload set and the B.foo overload set, are found. An instance of foo is selected based on it matching in exactly one overload set:

import A;
import B;

void bar(C c)
{
    foo();    // calls A.foo()
    foo(1L);  // calls A.foo(long)
    foo(c);   // calls B.foo(C)
    foo(1,2); // error, does not match any foo
    foo(1);   // error, matches A.foo(long) and B.foo(int)
    A.foo(1); // calls A.foo(long)
}

Even though B.foo(int) is a better match than A.foo(long) for foo(1), it is an error because the two matches are in different overload sets.

Overload sets can be merged with an alias declaration:

import A;
import B;

alias foo = A.foo;
alias foo = B.foo;

void bar(C c)
{
    foo();    // calls A.foo()
    foo(1L);  // calls A.foo(long)
    foo(c);   // calls B.foo(C)
    foo(1,2); // error, does not match any foo
    foo(1);   // calls B.foo(int)
    A.foo(1); // calls A.foo(long)
}

Function Parameters

Parameter storage classes are in, out, ref, lazy, const, immutable, shared, inout or scope. For example:

int foo(in int x, out int y, ref int z, int q);

x is in, y is out, z is ref, and q is none.

Parameter Storage Classes
Storage ClassDescription
noneparameter becomes a mutable copy of its argument
indefined as scope const. However in has not yet been properly implemented so it's current implementation is equivalent to const. It is recommended to avoid using in until it is properly defined and implemented. Use scope const or const explicitly instead.
outparameter is initialized upon function entry with the default value for its type
ref parameter is passed by reference
scopereferences in the parameter cannot be escaped (e.g. assigned to a global variable). Ignored for parameters with no references
lazyargument is evaluated by the called function and not by the caller
constargument is implicitly converted to a const type
immutableargument is implicitly converted to an immutable type
sharedargument is implicitly converted to a shared type
inoutargument is implicitly converted to an inout type
void foo(out int x)
{
    // x is set to int.init,
    // which is 0, at start of foo()
}

int a = 3;
foo(a);
// a is now 0

void abc(out int x)
{
    x = 2;
}

int y = 3;
abc(y);
// y is now 2

void def(ref int x)
{
    x += 1;
}

int z = 3;
def(z);
// z is now 4

For dynamic array and object parameters, which are passed by reference, in/out/ref apply only to the reference and not the contents.

lazy arguments are evaluated not when the function is called, but when the parameter is evaluated within the function. Hence, a lazy argument can be executed 0 or more times. A lazy parameter cannot be an lvalue.

void dotimes(int n, lazy void exp)
{
    while (n--)
        exp();
}

void test()
{
    int x;
    dotimes(3, writeln(x++));
}

prints to the console:

0
1
2

A lazy parameter of type void can accept an argument of any type.

Function Default Arguments

Function parameter declarations can have default values:

void foo(int x, int y = 3)
{
    ...
}
...
foo(4);   // same as foo(4, 3);

Default parameters are evaluated in the context of the function declaration. If the default value for a parameter is given, all following parameters must also have default values.

Return Ref Parameters

Note: The return attribute is currently only enforced by dmd when the -dip25 switch is passed.

Return ref parameters are used with ref functions to ensure that the returned reference will not outlive the matching argument's lifetime.

ref int identity(return ref int x) {
  return x; // pass-through function that does nothing
}

ref int fun() {
  int x;
  return identity(x); // Error: escaping reference to local variable x
}

ref int gun(return ref int x) {
  return identity(x); // OK
}

Ref methods marked with the return attribute ensure the returned reference will not outlive the respective aggregate instance.

struct S
{
    private int x;
    ref int get() return { return x; }
}

ref int escape()
{
    S s;
    return s.get(); // Error: escaping reference to local variable s
}

Template functions and lambdas can deduce the return attribute.

inout ref parameters imply the return attribute.

Variadic Functions

Functions taking a variable number of arguments are called variadic functions. A variadic function can take one of three forms:

  1. C-style variadic functions
  2. Variadic functions with type info
  3. Typesafe variadic functions

C-style Variadic Functions

A C-style variadic function is declared as taking a parameter of ... after the required function parameters. It has non-D linkage, such as extern (C):

extern (C) void foo(int x, int y, ...);

foo(3, 4);      // ok
foo(3, 4, 6.8); // ok, one variadic argument
foo(2);         // error, y is a required argument

There must be at least one non-variadic parameter declared.

extern (C) int def(...); // error, must have at least one parameter

C-style variadic functions match the C calling convention for variadic functions, and is most useful for calling C library functions like printf.

C-style variadic functions cannot be marked as @safe.

Access to variadic arguments is done using the standard library module core.stdc.stdarg.

import core.stdc.stdarg;

void test()
{
    foo(3, 4, 5);   // first variadic argument is 5
}

void foo(int x, int y, ...)
{
    va_list args;

    va_start(args, y);  // y is the last named parameter

    int z;
    va_arg(args, z);  // z is set to 5
}

D-style Variadic Functions

Variadic functions with argument and type info are declared as taking a parameter of ... after the required function parameters. It has D linkage, and need not have any non-variadic parameters declared:

int abc(char c, ...);   // one required parameter: c
int def(...);           // ok

To access them, the following import is required:

import core.vararg;

These variadic functions have a special local variable declared for them, _argptr, which is a core.vararg reference to the first of the variadic arguments. To access the arguments, _argptr must be used in conjuction with va_arg:

import core.vararg;

void test()
{
    foo(3, 4, 5);   // first variadic argument is 5
}

void foo(int x, int y, ...)
{
    int z;

    z = va_arg!int(_argptr); // z is set to 5
}

An additional hidden argument with the name _arguments and type TypeInfo[] is passed to the function. _arguments gives the number of arguments and the type of each, enabling type safety to be checked at run time.

import std.stdio;
import core.vararg;

class Foo { int x = 3; }
class Bar { long y = 4; }

void printargs(int x, ...)
{
    writefln("%d arguments", _arguments.length);
    for (int i = 0; i < _arguments.length; i++)
    {
        writeln(_arguments[i]);

        if (_arguments[i] == typeid(int))
        {
            int j = va_arg!(int)(_argptr);
            writefln("\t%d", j);
        }
        else if (_arguments[i] == typeid(long))
        {
            long j = va_arg!(long)(_argptr);
            writefln("\t%d", j);
        }
        else if (_arguments[i] == typeid(double))
        {
            double d = va_arg!(double)(_argptr);
            writefln("\t%g", d);
        }
        else if (_arguments[i] == typeid(Foo))
        {
            Foo f = va_arg!(Foo)(_argptr);
            writefln("\t%s", f);
        }
        else if (_arguments[i] == typeid(Bar))
        {
            Bar b = va_arg!(Bar)(_argptr);
            writefln("\t%s", b);
        }
        else
            assert(0);
    }
}

void main()
{
    Foo f = new Foo();
    Bar b = new Bar();

    writefln("%s", f);
    printargs(1, 2, 3L, 4.5, f, b);
}
which prints:
0x00870FE0
5 arguments
int
        2
long
        3
double
        4.5
Foo
        0x00870FE0
Bar
        0x00870FD0

D-style variadic functions cannot be marked as @safe.

Typesafe Variadic Functions

Typesafe variadic functions are used when the variable argument portion of the arguments are used to construct an array or class object.

For arrays:

int main()
{
    return sum(1, 2, 3) + sum(); // returns 6+0
}

int func()
{
    int[3] ii = [4, 5, 6];
    return sum(ii);             // returns 15
}

int sum(int[] ar ...)
{
    int s;
    foreach (int x; ar)
        s += x;
    return s;
}

For static arrays:

int test()
{
    return sum(2, 3);   // error, need 3 values for array
    return sum(1, 2, 3); // returns 6
}

int func()
{
    int[3] ii = [4, 5, 6];
    int[] jj = ii;
    return sum(ii); // returns 15
    return sum(jj); // error, type mismatch
}

int sum(int[3] ar ...)
{
    int s;
    foreach (int x; ar)
        s += x;
    return s;
}

For class objects:

class Foo
{
    int x;
    string s;

    this(int x, string s)
    {
        this.x = x;
        this.s = s;
    }
}

void test(int x, Foo f ...);

...

Foo g = new Foo(3, "abc");
test(1, g);         // ok, since g is an instance of Foo
test(1, 4, "def");  // ok
test(1, 5);         // error, no matching constructor for Foo

An implementation may construct the object or array instance on the stack. Therefore, it is an error to refer to that instance after the variadic function has returned:

Foo test(Foo f ...)
{
    return f;   // error, f instance contents invalid after return
}

int[] test(int[] a ...)
{
    return a;       // error, array contents invalid after return
    return a[0..1]; // error, array contents invalid after return
    return a.dup;   // ok, since copy is made
}

For other types, the argument is built with itself, as in:

int test(int i ...)
{
    return i;
}

...
test(3);    // returns 3
test(3, 4); // error, too many arguments
int[] x;
test(x);    // error, type mismatch

Lazy Variadic Functions

If the variadic parameter is an array of delegates with no parameters:

void foo(int delegate()[] dgs ...);

Then each of the arguments whose type does not match that of the delegate is converted to a delegate.

int delegate() dg;
foo(1, 3+x, dg, cast(int delegate())null);

is the same as:

foo( { return 1; }, { return 3+x; }, dg, null );

Local Variables

It is an error to use a local variable without first assigning it a value. The implementation may not always be able to detect these cases. Other language compilers sometimes issue a warning for this, but since it is always a bug, it should be an error.

It is an error to declare a local variable that hides another local variable in the same function:

void func(int x)
{
    int x;       // error, hides previous definition of x
    double y;
    ...
    {
        char y;  // error, hides previous definition of y
        int z;
    }
    {
        wchar z; // legal, previous z is out of scope
    }
}

While this might look unreasonable, in practice whenever this is done it either is a bug or at least looks like a bug.

It is an error to return the address of or a reference to a local variable.

It is an error to have a local variable and a label with the same name.

Local Static Variables

Local variables in functions can be declared as static or __gshared in which case they are statically allocated rather than being allocated on the stack. As such, their value persists beyond the exit of the function.

void foo()
{
    static int n;
    if (++n == 100)
        writeln("called 100 times");
}

The initializer for a static variable must be evaluatable at compile time, and they are initialized upon the start of the thread (or the start of the program for __gshared). There are no static constructors or static destructors for static local variables.

Although static variable name visibility follows the usual scoping rules, the names of them must be unique within a particular function.

void main()
{
    { static int x; }
    { static int x; } // error
    { int i; }
    { int i; } // ok
}

Nested Functions

Functions may be nested within other functions:

int bar(int a)
{
    int foo(int b)
    {
        int abc() { return 1; }

        return b + abc();
    }
    return foo(a);
}

void test()
{
    int i = bar(3); // i is assigned 4
}

Nested functions can be accessed only if the name is in scope.

void foo()
{
    void A()
    {
        B(); // error, B() is forward referenced
        C(); // error, C undefined
    }
    void B()
    {
        A(); // ok, in scope
        void C()
        {
            void D()
            {
                A();      // ok
                B();      // ok
                C();      // ok
                D();      // ok
            }
        }
    }
    A(); // ok
    B(); // ok
    C(); // error, C undefined
}
and:
int bar(int a)
{
    int foo(int b) { return b + 1; }
    int abc(int b) { return foo(b); }   // ok
    return foo(a);
}

void test()
{
    int i = bar(3);     // ok
    int j = bar.foo(3); // error, bar.foo not visible
}

Nested functions have access to the variables and other symbols defined by the lexically enclosing function. This access includes both the ability to read and write them.

int bar(int a)
{
    int c = 3;

    int foo(int b)
    {
        b += c;       // 4 is added to b
        c++;          // bar.c is now 5
        return b + c; // 12 is returned
    }
    c = 4;
    int i = foo(a); // i is set to 12
    return i + c;   // returns 17
}

void test()
{
    int i = bar(3); // i is assigned 17
}

This access can span multiple nesting levels:

int bar(int a)
{
    int c = 3;

    int foo(int b)
    {
        int abc()
        {
            return c;   // access bar.c
        }
        return b + c + abc();
    }
    return foo(3);
}

Static nested functions cannot access any stack variables of any lexically enclosing function, but can access static variables. This is analogous to how static member functions behave.

int bar(int a)
{
    int c;
    static int d;

    static int foo(int b)
    {
        b = d;          // ok
        b = c;          // error, foo() cannot access frame of bar()
        return b + 1;
    }
    return foo(a);
}

Functions can be nested within member functions:

struct Foo
{
    int a;

    int bar()
    {
        int c;

        int foo()
        {
            return c + a;
        }
        return 0;
    }
}

Nested functions always have the D function linkage type.

Unlike module level declarations, declarations within function scope are processed in order. This means that two nested functions cannot mutually call each other:

void test()
{
    void foo() { bar(); } // error, bar not defined
    void bar() { foo(); } // ok
}

There are several workarounds for this limitation:

Nested functions cannot be overloaded.

Delegates, Function Pointers, and Closures

A function pointer can point to a static nested function:

int function() fp;

void test()
{
    static int a = 7;
    static int foo() { return a + 3; }

    fp = &foo;
}

void bar()
{
    test();
    int i = fp();       // i is set to 10
}

Note: Two functions with identical bodies, or two functions that compile to identical assembly code, are not guaranteed to have distinct function pointer values. The compiler is free to merge functions bodies into one if they compile to identical code.

int abc(int x) { return x + 1; }
int def(int y) { return y + 1; }

int delegate(int) fp1 = &abc;
int delegate(int) fp2 = &def;
// Do not rely on fp1 and fp2 being different values; the compiler may merge
// them.

A delegate can be set to a non-static nested function:

int delegate() dg;

void test()
{
    int a = 7;
    int foo() { return a + 3; }

    dg = &foo;
    int i = dg(); // i is set to 10
}

The stack variables referenced by a nested function are still valid even after the function exits (this is different from D 1.0). This is called a closure. Returning addresses of stack variables, however, is not a closure and is an error.

int* bar()
{
    int b;
    test();
    int i = dg(); // ok, test.a is in a closure and still exists
    return &b;    // error, bar.b not valid after bar() exits
}

Delegates to non-static nested functions contain two pieces of data: the pointer to the stack frame of the lexically enclosing function (called the frame pointer) and the address of the function. This is analogous to struct/class non-static member function delegates consisting of a this pointer and the address of the member function. Both forms of delegates are interchangeable, and are actually the same type:

struct Foo
{
    int a = 7;
    int bar() { return a; }
}

int foo(int delegate() dg)
{
    return dg() + 1;
}

void test()
{
    int x = 27;
    int abc() { return x; }
    Foo f;
    int i;

    i = foo(&abc);   // i is set to 28
    i = foo(&f.bar); // i is set to 8
}

This combining of the environment and the function is called a dynamic closure.

The .ptr property of a delegate will return the frame pointer value as a void*.

The .funcptr property of a delegate will return the function pointer value as a function type.

Functions and delegates declared at module scope are zero-initialized by default. However both can be initialized to any function pointer (including a function literal). For delegates, the context pointer .ptr is initialized to null.

int function() foo = { return 42; };
int delegate() bar = { return 43; };
int delegate() baz;
void main()
{
    assert(foo() == 42);
    assert(bar() == 43);
    assert(baz is null);
}

Future directions: Function pointers and delegates may merge into a common syntax and be interchangeable with each other.

Anonymous Functions and Anonymous Delegates

See FunctionLiterals.

main() Function

For console programs, main() serves as the entry point. It gets called after all the module initializers are run, and after any unittests are run. After it returns, all the module destructors are run. main() must be declared using one of the following forms:

void main() { ... }
void main(string[] args) { ... }
int main() { ... }
int main(string[] args) { ... }

Function Templates

Template functions are useful for avoiding code duplication - instead of writing several copies of a function, each with a different parameter type, a single function template can be sufficient. For example:

// Only one copy of func needs to be written
void func(T)(T x)
{
    writeln(x);
}
void main()
{
    func!(int)(1); // pass an int
    func(1);    // pass an int, inferring T = int
    func("x");  // pass a string
    func(1.0);  // pass a float
    struct S {}
    S s;
    func(s);    // pass a struct
}

func takes a template parameter T and a runtime parameter, x. T is a placeholder identifier that can accept any type. In this case T can be inferred from the runtime argument type.

Note: Using the name T is just a convention. The name TypeOfX could have been used instead.

For more information, see function templates.

Compile Time Function Execution (CTFE)

Functions which are both portable and free of global side-effects can be executed at compile time. In certain contexts, such compile time execution is guaranteed. It is called Compile Time Function Execution (CTFE) then. The contexts that trigger CTFE are:

enum eval(Args...) = Args[0];

int square(int i)
{
    return i * i;
}

void foo()
{
    static j = square(3);      // CTFE
    writeln(j);
    assert(square(4));         // run time
    writeln(eval!(square(5))); // CTFE
}

CTFE is subject to the following restrictions:

  1. The function source code must be available to the compiler. Functions which exist in the source code only as extern declarations cannot be executed in CTFE.
  2. Executed expressions may not reference any global or local static variables.
  3. asm statements are not permitted
  4. Non-portable casts (eg, from int[] to float[]), including casts which depend on endianness, are not permitted. Casts between signed and unsigned types are permitted
  5. Reinterpretation of overlapped fields in a Union is not permitted.

Pointers are permitted in CTFE, provided they are used safely:

Note that the above restrictions apply only to expressions which are actually executed. For example:

static int y = 0;

int countTen(int x)
{
    if (x > 10)
        ++y;
    return x;
}

static assert(countTen(6) == 6);    // OK
static assert(countTen(12) == 12);  // invalid, modifies y.

The __ctfe boolean pseudo-variable, which evaluates to true in CTFE, but false otherwise, can be used to provide an alternative execution path to avoid operations which are forbidden in CTFE. Every usage of __ctfe is evaluated before code generation and therefore has no run-time cost, even if no optimizer is used.

Executing functions via CTFE can take considerably longer than executing it at run time. If the function goes into an infinite loop, it will hang at compile time (rather than hanging at run time).

Non-recoverable errors (such as assert failures) do not throw exceptions; instead, they end interpretation immediately.

Functions executed via CTFE can give different results from run time in the following scenarios:

These are the same kinds of scenarios where different optimization settings affect the results.

String Mixins and Compile Time Function Execution

Any functions that execute in CTFE must also be executable at run time. The compile time evaluation of a function does the equivalent of running the function at run time. This means that the semantics of a function cannot depend on compile time values of the function. For example:

int foo(char[] s)
{
    return mixin(s);
}

const int x = foo("1");
is illegal, because the runtime code for foo cannot be generated. A function template would be the appropriate method to implement this sort of thing.

No-GC Functions

No-GC functions are functions marked with the @nogc attribute. Those functions do not allocate memory on the GC heap, through the following language features:

@nogc void foo()
{
    auto a = ['a'];    // error, allocates
    a.length = 1;      // error, array resizing allocates
    a = a ~ a;         // error, arrays concatenation allocates
    a ~= 'c';          // error, appending to arrays allocates

    auto aa = ["x":1]; // error, allocates
    aa["abc"];         // error, indexing may allocate and throws

    auto p = new int;  // error, operator new allocates
}

No-GC functions cannot call functions that are not @nogc.

@nogc void foo()
{
    bar();             // error, bar() may allocate
}

void bar() { }

No-GC functions cannot be closures.

@nogc int delegate() foo()
{
    int n;              // error, variable n cannot be allocated on heap
    return (){ return n; }
}

@nogc affects the type of the function. A @nogc function is covariant with a non-@nogc function.

void function() fp;
void function() @nogc gp;  // pointer to @nogc function

void foo();
@nogc void bar();

void test()
{
    fp = &foo; // ok
    fp = &bar; // ok, it's covariant
    gp = &foo; // error, not contravariant
    gp = &bar; // ok
}

To ease debugging, in a ConditionalStatement controlled by a DebugCondition @nogc functions can call functions that are not @nogc.

Function Safety

Safe functions are functions that are statically checked to exhibit no possibility of undefined behavior. Undefined behavior is often used as a vector for malicious attacks.

Safe Functions

Safe functions are marked with the @safe attribute.

The following operations are not allowed in safe functions:

When indexing and slicing an array, an out of bounds access will cause a runtime error, in order to prevent undefined behavior.

Functions nested inside safe functions default to being safe functions.

Safe functions are covariant with trusted or system functions.

Note: The verifiable safety of functions may be compromised by bugs in the compiler and specification. Please report all such errors so they can be corrected.

Trusted Functions

Trusted functions are marked with the @trusted attribute.

Trusted functions are guaranteed by the programmer to not exhibit any undefined behavior if called by a safe function. Generally, trusted functions should be kept small so that they are easier to manually verify.

Trusted functions may call safe, trusted, or system functions.

Trusted functions are covariant with safe or system functions.

System Functions

System functions are functions not marked with @safe or @trusted and are not nested inside @safe functions. System functions may be marked with the @system attribute. A function being system does not mean it actually is unsafe, it just means that the compiler is unable to verify that it cannot exhibit undefined behavior.

System functions are not covariant with trusted or safe functions.

Function Attribute Inference

FunctionLiterals and function templates, since their function bodies are always present, infer the pure, nothrow, @safe, and @nogc attributes unless specifically overridden.

Attribute inference is not done for other functions, even if the function body is present.

The inference is done by determining if the function body follows the rules of the particular attribute.

Cyclic functions (i.e. functions that wind up directly or indirectly calling themselves) are inferred as being impure, throwing, and @system.

If a function attempts to test itself for those attributes, then the function is inferred as not having those attributes.

Uniform Function Call Syntax (UFCS)

A free function can be called with a syntax that looks as if the function were a member function of its first parameter type.

void func(X thisObj);

X obj;
obj.func();
// If 'obj' does not have regular member 'func',
// it's automatically rewritten to 'func(obj)'

This provides a way to add functions to a class externally as if they were public final member functions, which enables function chaining and component programming.

stdin.byLine(KeepTerminator.yes)
    .map!(a => a.idup)
    .array
    .sort
    .copy(stdout.lockingTextWriter());

It also works with @property functions:

@property prop(X thisObj);
@property prop(X thisObj, int value);

X obj;
obj.prop;      // Rewrites to: prop(obj);
obj.prop = 1;  // Rewrites to: prop(obj, 1);

Syntactically parenthesis-less check for @property functions is done at the same time as UFCS rewrite.

When UFCS rewrite is necessary, compiler searches the name on accessible module level scope, in order from the innermost scope.

module a;
void foo(X);
alias boo = foo;
void main()
{
    void bar(X);
    import b : baz;  // void baz(X);

    X obj;
    obj.foo();    // OK, calls a.foo;
    //obj.bar();  // NG, UFCS does not see nested functions
    obj.baz();    // OK, calls b.baz, because it is declared at the
                  // top level scope of module b

    import b : boo = baz;
    obj.boo();    // OK, calls aliased b.baz instead of a.boo (== a.foo),
                  // because the declared alias name 'boo' in local scope
                  // overrides module scope name
}
class C
{
    void mfoo(X);
    static void sbar(X);
    import b : ibaz = baz;  // void baz(X);
    void test()
    {
        X obj;
        //obj.mfoo();  // NG, UFCS does not see member functions
        //obj.sbar();  // NG, UFCS does not see static member functions
        obj.ibaz();    // OK, ibaz is an alias of baz which declared at
                       //     the top level scope of module b
    }
}

The reason why local symbols are not considered by UFCS, is to avoid unexpected name conflicts. See below problematic examples.

int front(int[] arr) { return arr[0]; }

void main()
{
    int[] a = [1,2,3];
    auto x = a.front();   // call .front by UFCS

    auto front = x;       // front is now a variable
    auto y = a.front();   // Error, front is not a function
}

class C
{
    int[] arr;
    int front()
    {
        return arr.front(); // Error, C.front is not callable
                            // using argument types (int[])
    }
}
Type Qualifiers
Operator Overloading