The D Programming Language

This module implements a variety of type constructors, i.e., templates that allow construction of new, useful general-purpose types.

// value tuples

alias Tuple!(float, "x", float, "y", float, "z") Coord;
Coord c;
c[1] = 1;       // access by index

c.z = 1;        // access by given name

alias Tuple!(string, string) DicEntry; // names can be omitted

// Rebindable references to const and immutable objects

void bar()
    const w1 = new Widget, w2 = new Widget;;
    // w1 = w2 would not work; can't rebind const object

    auto r = Rebindable!(const Widget)(w1);
    // invoke method as if r were a Widget object;
    // rebind r to refer to another object

    r = w2;
Boost License 1.0.
Andrei Alexandrescu, Bartosz Milewski, Don Clugston, Shin Fujishiro, Kenji Hara

struct  Unique(T);

Encapsulates unique ownership of a resource. Resource of type T is deleted at the end of the scope, unless it is transferred. The transfer can be explicit, by calling release, or implicit, when returning  Unique from a function. The resource can be a polymorphic class object, in which case  Unique behaves polymorphically too.


this(RefT p);

Constructor that takes an rvalue. It will ensure uniqueness, as long as the rvalue isn't just a view on an lvalue (e.g., a cast) Typical usage:

Unique!(Foo) f = new Foo;

this(ref RefT p);

Constructor that takes an lvalue. It nulls its source. The nulling will ensure uniqueness as long as there are no previous aliases to the source.

Unique  release();

Returns a unique rvalue. Nullifies the current contents

RefT  opDot();

Forwards member access to contents

struct  Tuple(Specs...);

 Tuple of values, for example  Tuple!(int, string) is a record that stores an int and a string.  Tuple can be used to bundle values together, notably when returning multiple values from a function. If obj is a tuple, the individual members are accessible with the syntax obj[0] for the first field, obj[1] for the second, and so on.

The choice of zero-based indexing instead of one-base indexing was motivated by the ability to use value tuples with various compile-time loop constructs (e.g. type tuple iteration), all of which use zero-based indexing.

Tuple!(int, int) point;
// assign coordinates

point[0] = 5;
point[1] = 6;
// read coordinates

auto x = point[0];
auto y = point[1];

 Tuple members can be named. It is legal to mix named and unnamed members. The method above is still applicable to all fields.
alias Tuple!(int, "index", string, "value") Entry;
Entry e;
e.index = 4;
e.value = "Hello";
assert(e[1] == "Hello");
assert(e[0] == 4);

Tuples with named fields are distinct types from tuples with unnamed fields, i.e. each naming imparts a separate type for the tuple. Two tuple differing in naming only are still distinct, even though they might have the same structure.
Tuple!(int, "x", int, "y") point1;
Tuple!(int, int) point2;
assert(!is(typeof(point1) == typeof(point2))); // passes

alias  Types = staticMap!(extractType, fieldSpecs);

The type of the tuple's components.

Types  expand;

Use t. expand for a tuple t to  expand it into its components. The result of  expand acts as if the tuple components were listed as a list of values. (Ordinarily, a Tuple acts as a single value.)

auto t = tuple(1, " hello ", 2.3);
writeln(t);        // Tuple!(int, string, double)(1, " hello ", 2.3)

writeln(t.expand); // 1 hello 2.3

this()(Types values);

Constructor taking one value for each field. Each argument must be implicitly assignable to the respective element of the target.

this(U, size_t n)(U[n] values) if (n == Types.length && is(typeof(() { foreach (i, _; Types) { field[i] = values[i]; } } )));

Constructor taking a compatible array. The array element type must be implicitly assignable to each element of the target.

int[2] ints;
Tuple!(int, int) t = ints;

this(U)(U another) if (areCompatibleTuples!(typeof(this), U, "="));

Constructor taking a compatible tuple. Each element of the source must be implicitly assignable to the respective element of the target.

bool  opEquals(R)(R rhs) if (areCompatibleTuples!(typeof(this), R, "=="));
const bool  opEquals(R)(R rhs) if (areCompatibleTuples!(typeof(this), R, "=="));

Comparison for equality.

int  opCmp(R)(R rhs) if (areCompatibleTuples!(typeof(this), R, "<"));
const int  opCmp(R)(R rhs) if (areCompatibleTuples!(typeof(this), R, "<"));

Comparison for ordering.

void  opAssign(R)(auto ref R rhs) if (areCompatibleTuples!(typeof(this), R, "="));

Assignment from another tuple. Each element of the source must be implicitly assignable to the respective element of the target.

@trusted Tuple!(sliceSpecs!(from, to))  slice(size_t from, size_t to)() if (from <= to && to <= Types.length);

Takes a  slice of the tuple.

Tuple!(int, string, float, double) a;
a[1] = "abc";
a[2] = 4.5;
auto s = a.slice!(1, 3);
static assert(is(typeof(s) == Tuple!(string, float)));
assert(s[0] == "abc" && s[1] == 4.5);

string  toString();

Converts to string.

Tuple!T  tuple(T...)(T args);

Returns a Tuple object instantiated and initialized according to the arguments.

auto value = tuple(5, 6.7, "hello");
assert(value[0] == 5);
assert(value[1] == 6.7);
assert(value[2] == "hello");

template  isTuple(T)

Returns true if and only if T is an instance of the Tuple struct template.

template  Rebindable(T) if (is(T == class) || is(T == interface) || isArray!T)

 Rebindable!(T) is a simple, efficient wrapper that behaves just like an object of type T, except that you can reassign it to refer to another object. For completeness,  Rebindable!(T) aliases itself away to T if T is a non-const object type. However,  Rebindable!(T) does not compile if T is a non-class type.

Regular const object references cannot be reassigned:

class Widget { int x; int y() const { return x; } }
const a = new Widget;
a.y();          // fine

a.x = 5;        // error! can't modify const a

a = new Widget; // error! can't modify const a

However,  Rebindable!(Widget) does allow reassignment, while otherwise behaving exactly like a const Widget:

auto a = Rebindable!(const Widget)(new Widget);
a.y();          // fine

a.x = 5;        // error! can't modify const a

a = new Widget; // fine

You may want to use  Rebindable when you want to have mutable storage referring to const objects, for example an array of references that must be sorted in place.  Rebindable does not break the soundness of D's type system and does not incur any of the risks usually associated with cast.

Rebindable!T  rebindable(T)(T obj) if (is(T == class) || is(T == interface) || isArray!T);

Convenience function for creating a Rebindable using automatic type inference.

Rebindable!T  rebindable(T)(Rebindable!T obj);

This function simply returns the Rebindable object passed in. It's useful in generic programming cases when a given object may be either a regular class or a Rebindable.

string  alignForSize(E...)(string[] names...);

Order the provided members to minimize size while preserving alignment. Returns a declaration to be mixed in.

struct Banner {
  mixin(alignForSize!(byte[6], double)(["name", "height"]));

Alignment is not always optimal for 80-bit reals, nor for structs declared as align(1).

struct  Nullable(T);

Defines a value paired with a distinctive "null" state that denotes the absence of a value. If default constructed, a  Nullable!T object starts in the null state. Assigning it renders it non-null. Calling nullify can nullify it again.

Nullable!int a;
a = 5;
assert(a == 5);

Practically  Nullable!T stores a T and a bool.

inout this(U : T)(U value);

Constructor initializing this with value.

const pure nothrow @safe bool  isNull();

Returns true if and only if this is in the null state.

void  nullify()();

Forces this to the null state.

void  opAssign()(T value);

Assigns value to the internally-held state. If the assignment succeeds, this becomes non-null.

inout pure nothrow @safe inout(T)  get();

Gets the value. this must not be in the null state. This function is also called for the implicit conversion to T.

struct  Nullable(T, T nullValue);

Just like  Nullable!T, except that the null state is defined as a particular value. For example,  Nullable!(uint, uint.max) is an uint that sets aside the value uint.max to denote a null state.  Nullable!(T, nullValue) is more storage-efficient than  Nullable!T because it does not need to store an extra bool.

this()(T value);

Constructor initializing this with value.

const bool  isNull()();

Returns true if and only if this is in the null state.

void  nullify()();

Forces this to the null state.

void  opAssign()(T value);

Assigns value to the internally-held state. No null checks are made. Note that the assignment may leave this in the null state.

inout inout(T)  get()();

Gets the value. this must not be in the null state. This function is also called for the implicit conversion to T.

struct  NullableRef(T);

Just like Nullable!T, except that the object refers to a value sitting elsewhere in memory. This makes assignments overwrite the initially assigned value. Internally  NullableRef!T only stores a pointer to T (i.e., Nullable!T.sizeof == (T*).sizeof).

pure nothrow @safe this(T* value);

Constructor binding this with value.

pure nothrow @safe void  bind(T* value);

Binds the internal state to value.

const pure nothrow @safe bool  isNull();

Returns true if and only if this is in the null state.

pure nothrow @safe void  nullify();

Forces this to the null state.

void  opAssign()(T value) if (isAssignable!T);

Assigns value to the internally-held state.

inout pure nothrow @safe inout(T)  get();

Gets the value. this must not be in the null state. This function is also called for the implicit conversion to T.

template  BlackHole(Base)

 BlackHole!Base is a subclass of Base which automatically implements all abstract member functions in Base as do-nothing functions. Each auto-implemented function just returns the default value of the return type without doing anything.

The name came from Class::BlackHole Perl module by Sean M. Burke.

abstract class C
    int m_value;
    this(int v) { m_value = v; }
    int value() @property { return m_value; }

    abstract real realValue() @property;
    abstract void doSomething();

void main()
    auto c = new BlackHole!C(42);
    writeln(c.value);     // prints "42"

    // Abstract functions are implemented as do-nothing:

    writeln(c.realValue); // prints "NaN"

    c.doSomething();      // does nothing

See Also
AutoImplement, generateEmptyFunction

template  WhiteHole(Base)

 WhiteHole!Base is a subclass of Base which automatically implements all abstract member functions as throw-always functions. Each auto-implemented function fails with throwing an Error and does never return. Useful for trapping use of not-yet-implemented functions.

The name came from Class::WhiteHole Perl module by Michael G Schwern.

class C
    abstract void notYetImplemented();

void main()
    auto c = new WhiteHole!C;
    c.notYetImplemented(); // throws an Error

Known Bugs
Nothrow functions cause program to abort in release mode because the trap is implemented with assert(0) for nothrow functions.
See Also
AutoImplement, generateAssertTrap

class  AutoImplement(Base, alias how, alias what = isAbstractFunction): Base;

 AutoImplement automatically implements (by default) all abstract member functions in the class or interface Base in specified way.

how template which specifies how functions will be implemented/overridden.

Two arguments are passed to how: the type Base and an alias to an implemented function. Then how must return an implemented function body as a string.

The generated function body can use these keywords:
  • a0, a1, …: arguments passed to the function;
  • args: a tuple of the arguments;
  • self: an alias to the function itself;
  • parent: an alias to the overridden function (if any).

You may want to use templated property functions (instead of Implicit Template Properties) to generate complex functions:
// Prints log messages for each call to overridden functions.

string generateLogger(C, alias fun)() @property
    import std.traits;
    enum qname = C.stringof ~ "." ~ __traits(identifier, fun);
    string stmt;

    stmt ~= q{ struct Importer { import std.stdio; } };
    stmt ~= `Importer.writeln$(LPAREN)"Log: ` ~ qname ~ `(", args, ")"$(RPAREN);`;
    static if (!__traits(isAbstractFunction, fun))
        static if (is(ReturnType!fun == void))
            stmt ~= q{ parent(args); };
            stmt ~= q{
                auto r = parent(args);
                Importer.writeln("--> ", r);
                return r;
    return stmt;
what template which determines what functions should be implemented/overridden.

An argument is passed to what: an alias to a non-final member function in Base. Then what must return a boolean value. Return true to indicate that the passed function should be implemented/overridden.

// Sees if fun returns something.

template hasValue(alias fun)
    enum bool hasValue = !is(ReturnType!(fun) == void);
Generated code is inserted in the scope of std.typecons module. Thus, any useful functions outside std.typecons cannot be used in the generated code. To workaround this problem, you may import necessary things in a local struct, as done in the generateLogger() template in the above example.

Known Bugs
  • Variadic arguments to constructors are not forwarded to super.
  • Deep interface inheritance causes compile error with messages like "Error: function std.typecons.AutoImplement!(Foo) does not override any function". [Bugzilla 2525, Bugzilla 3525]
  • The parent keyword is actually a delegate to the super class' corresponding member function. [Bugzilla 2540]
  • Using alias template parameter in how and/or what may cause strange compile error. Use template tuple parameter instead to workaround this problem. [Bugzilla 4217]

template  generateEmptyFunction(C, func...)
template  generateAssertTrap(C, func...)

Predefined how-policies for AutoImplement. These templates are used by BlackHole and WhiteHole, respectively.

template  wrap(Targets...) if (Targets.length >= 1 && allSatisfy!(isMutable, Targets))
template  wrap(Targets...) if (Targets.length >= 1 && !allSatisfy!(isMutable, Targets))

Supports structural based typesafe conversion.

If Source has structural conformance with the interface Targets,  wrap creates internal wrapper class which inherits Targets and  wrap src object, then return it.

template  unwrap(Target) if (isMutable!Target)
template  unwrap(Target) if (!isMutable!Target)

Extract object which wrapped by wrap.

    interface Quack
        int quack();
        @property int height();
    interface Flyer
        @property int height();
    class Duck : Quack
        int quack() { return 1; }
        @property int height() { return 10; }
    class Human
        int quack() { return 2; }
        @property int height() { return 20; }

    Duck d1 = new Duck();
    Human h1 = new Human();

    interface Refleshable
        int reflesh();
    // does not have structural conformance

    static assert(!__traits(compiles, d1.wrap!Refleshable));
    static assert(!__traits(compiles, h1.wrap!Refleshable));

    // strict upcast

    Quack qd = d1.wrap!Quack;
    assert(qd is d1);
    assert(qd.quack() == 1);    // calls Duck.quack

    // strict downcast

    Duck d2 = qd.unwrap!Duck;
    assert(d2 is d1);

    // structural upcast

    Quack qh = h1.wrap!Quack;
    assert(qh.quack() == 2);    // calls Human.quack

    // structural downcast

    Human h2 = qh.unwrap!Human;
    assert(h2 is h1);

    // structural upcast (two steps)

    Quack qx = h1.wrap!Quack;   // Human -> Quack

    Flyer fx = qx.wrap!Flyer;   // Quack -> Flyer

    assert(fx.height == 20);    // calls Human.height

    // strucural downcast (two steps)

    Quack qy = fx.unwrap!Quack; // Flyer -> Quack

    Human hy = qy.unwrap!Human; // Quack -> Human

    assert(hy is h1);
    // strucural downcast (one step)

    Human hz = fx.unwrap!Human; // Flyer -> Human

    assert(hz is h1);
    interface A { int run(); }
    interface B { int stop(); @property int status(); }
    class X
        int run() { return 1; }
        int stop() { return 2; }
        @property int status() { return 3; }

    auto x = new X();
    auto ab = x.wrap!(A, B);
    A a = ab;
    B b = ab;
    assert( == 1);
    assert(b.stop() == 2);
    assert(b.status == 3);
    static assert(functionAttributes!(typeof(ab).status) &;

Options regarding auto-initialization of a RefCounted object (see the definition of RefCounted below).


Do not auto-initialize the object


Auto-initialize the object

struct  RefCounted(T, RefCountedAutoInitialize autoInit = RefCountedAutoInitialize.yes) if (!is(T == class));

Defines a reference-counted object containing a T value as payload.  RefCounted keeps track of all references of an object, and when the reference count goes down to zero, frees the underlying store.  RefCounted uses malloc and free for operation.

 RefCounted is unsafe and should be used with care. No references to the payload should be escaped outside the  RefCounted object.

The autoInit option makes the object ensure the store is automatically initialized. Leaving autoInit == RefCountedAutoInitialize.yes (the default option) is convenient but has the cost of a test whenever the payload is accessed. If autoInit ==, user code must call either refCountedStore.isInitialized or refCountedStore.ensureInitialized before attempting to access the payload. Not doing so results in null pointer dereference.

// A pair of an $(D int) and a $(D size_t) - the latter being the

// reference count - will be dynamically allocated

auto rc1 = RefCounted!int(5);
assert(rc1 == 5);
// No more allocation, add just one extra reference count

auto rc2 = rc1;
// Reference semantics

rc2 = 42;
assert(rc1 == 42);
// the pair will be freed when rc1 and rc2 go out of scope

struct  RefCountedStore;

RefCounted storage implementation.

const bool  isInitialized();

Returns true if and only if the underlying store has been allocated and initialized.

const size_t  refCount();

Returns underlying reference count if it is allocated and initialized (a positive integer), and 0 otherwise.

Makes sure the payload was properly initialized. Such a call is typically inserted before using the payload.

inout inout(RefCountedStore)  refCountedStore();

Returns storage implementation struct.

this(A...)(auto ref A args) if (A.length > 0);

Constructor that initializes the payload.


void  opAssign(typeof(this) rhs);
void  opAssign(T rhs);

Assignment operators

T  refCountedPayload();
inout inout(T)  refCountedPayload();

Returns a reference to the payload. If (autoInit == RefCountedAutoInitialize.yes), calls refCountedStore.ensureInitialized. Otherwise, just issues assert(refCountedStore.isInitialized). Used with alias  refCountedPayload this;, so callers can just use the RefCounted object as a T.

The first overload exists only if autoInit == RefCountedAutoInitialize.yes. So if autoInit == or called for a constant or immutable object, then  refCountedPayload will also be qualified as safe and nothrow (but will still assert if not initialized).

template  Proxy(alias a)

Make proxy for a.

struct MyInt
    private int value;
    mixin Proxy!value;

    this(int n){ value = n; }

MyInt n = 10;

// Enable operations that original type has.

assert(n == 11);
assert(n * 2 == 22);

void func(int n) { }

// Disable implicit conversions to original type.

//int x = n;


template  Typedef(T)
struct  Typedef(T, T init, string cookie = null);

Library typedef.

template  scoped(T) if (is(T == class))

Allocates a class object right inside the current scope, therefore avoiding the overhead of new. This facility is unsafe; it is the responsibility of the user to not escape a reference to the object outside the scope.

it's illegal to move a class reference even if you are sure there are no pointers to it. As such, it is illegal to move a  scoped object.
    class A
        int x;
        this()     {x = 0;}
        this(int i){x = i;}

    // Standard usage

    auto a1 = scoped!A();
    auto a2 = scoped!A(1);
    a1.x = 42;
    assert(a1.x == 42);
    assert(a2.x ==  1);

    // Restrictions

    static assert(!is(typeof({
        auto e1 = a1; // illegal, scoped objects can't be copied

        assert([a2][0].x == 42); // ditto

        alias ScopedObject = typeof(a1);
        auto e2 = ScopedObject();  //Illegal, must be built via scoped!A

        auto e3 = ScopedObject(1); //Illegal, must be built via scoped!A


    // Use with alias

    alias makeScopedA = scoped!A;
    auto a6 = makeScopedA();
    auto a7 = makeScopedA();

@system auto  scoped(Args...)(auto ref Args args);

Returns the  scoped object

template  Flag(string name)

Defines a simple, self-documenting yes/no flag. This makes it easy for APIs to define functions accepting flags without resorting to bool, which is opaque in calls, and without needing to define an enumerated type separately. Using  Flag!"Name" instead of bool makes the flag's meaning visible in calls. Each yes/no flag has its own type, which makes confusions and mix-ups impossible.

// Before

string getLine(bool keepTerminator)
    if (keepTerminator) ...
// Code calling getLine (usually far away from its definition) can't

// be understood without looking at the documentation, even by users

// familiar with the API. Assuming the reverse meaning

// (i.e. "ignoreTerminator") and inserting the wrong code compiles and

// runs with erroneous results.

auto line = getLine(false);

// After

string getLine(Flag!"KeepTerminator" keepTerminator)
    if (keepTerminator) ...
// Code calling getLine can be easily read and understood even by

// people not fluent with the API.

auto line = getLine(Flag!"KeepTerminator".yes);

Passing categorical data by means of unstructured bool parameters is classified under "simple-data coupling" by Steve McConnell in the Code Complete book, along with three other kinds of coupling. The author argues citing several studies that coupling has a negative effect on code quality.  Flag offers a simple structuring method for passing yes/no flags to APIs.

As a perk, the flag's name may be any string and as such can include characters not normally allowed in identifiers, such as spaces and dashes.

enum  Flag: bool;


When creating a value of type Flag!"Name", use Flag!"Name". no for the negative option. When using a value of type Flag!"Name", compare it against Flag!"Name". no or just false or 0.


When creating a value of type Flag!"Name", use Flag!"Name". yes for the affirmative option. When using a value of type Flag!"Name", compare it against Flag!"Name". yes.

struct  Yes;
struct  No;

Convenience names that allow using e.g.  Yes.encryption instead of Flag!"encryption".yes and No.encryption instead of Flag!"encryption".no.