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 local clone. Page wiki View or edit the community-maintained wiki page associated with this page.


Functions that manipulate other functions.

Boost License 1.0.

Andrei Alexandrescu


template unaryFun(alias fun, string parmName = "a")
Transforms a string representing an expression into a unary function. The string must either use symbol name a as the parameter or provide the symbol via the parmName argument. If fun is not a string, unaryFun aliases itself away to fun.

alias unaryFun!("(a & 1) == 0") isEven;
assert(isEven(2) && !isEven(1));

template binaryFun(alias fun, string parm1Name = "a", string parm2Name = "b")
Transforms a string representing an expression into a Boolean binary predicate. The string must either use symbol names a and b as the parameters or provide the symbols via the parm1Name and parm2Name arguments. If fun is not a string, binaryFun aliases itself away to fun.

alias less = binaryFun!("a < b");
assert(less(1, 2) && !less(2, 1));
alias greater = binaryFun!("a > b");
assert(!greater("1", "2") && greater("2", "1"));

template reverseArgs(alias pred)
N-ary predicate that reverses the order of arguments, e.g., given pred(a, b, c), returns pred(c, b, a).

template binaryReverseArgs(alias pred)
Binary predicate that reverses the order of arguments, e.g., given pred(a, b), returns pred(b, a).

template not(alias pred)
Negates predicate pred.

string a = "   Hello, world!";
assert(find!(not!isWhite)(a) == "Hello, world!");

template partial(alias fun, alias arg)
Partially evaluates fun by tying its first argument to a particular value.

int fun(int a, int b) { return a + b; }
alias partial!(fun, 5) fun5;
assert(fun5(6) == 11);

Note that in most cases you'd use an alias instead of a value assignment. Using an alias allows you to partially evaluate template functions without committing to a particular type of the function.

deprecated alias curry = partial(alias fun, alias arg);
Deprecated alias for partial, kept for backwards compatibility

template adjoin(F...) if (F.length == 1)
template adjoin(F...) if (F.length > 1)
Takes multiple functions and adjoins them together. The result is a std.typecons.Tuple with one element per passed-in function. Upon invocation, the returned tuple is the adjoined results of all functions.

In the special case where where only a single function is provided (F.length == 1), adjoin simply aliases to the single passed function (F[0]).

import std.functional, std.typecons;
static bool f1(int a) { return a != 0; }
static int f2(int a) { return a / 2; }
auto x = adjoin!(f1, f2)(5);
assert(is(typeof(x) == Tuple!(bool, int)));
assert(x[0] == true && x[1] == 2);

template compose(fun...)
Composes passed-in functions fun[0], fun[1], ... returning a function f(x) that in turn returns fun[0](fun[1](...(x))).... Each function can be a regular functions, a delegate, or a string.

// First split a string in whitespace-separated tokens and then
// convert each token into an integer
assert(compose!(map!(to!(int)), split)("1 2 3") == [1, 2, 3]);

template pipe(fun...)
Pipes functions in sequence. Offers the same functionality as compose, but with functions specified in reverse order. This may lead to more readable code in some situation because the order of execution is the same as lexical order.

// Read an entire text file, split the resulting string in
// whitespace-separated tokens, and then convert each token into an
// integer
int[] a = pipe!(readText, split, map!(to!(int)))("file.txt");

ReturnType!fun memoize(alias fun, uint maxSize = (uint).max)(Args args);
Memoizes a function so as to avoid repeated computation. The memoization structure is a hash table keyed by a tuple of the function's arguments. There is a speed gain if the function is repeatedly called with the same arguments and is more expensive than a hash table lookup. For more information on memoization, refer to this book chapter.

double transmogrify(int a, string b)
   ... expensive computation ...
alias fastTransmogrify = memoize!transmogrify;
    auto slow = transmogrify(2, "hello");
    auto fast = fastTransmogrify(2, "hello");
    assert(slow == fast);

Technically the memoized function should be pure because memoize assumes it will always return the same result for a given tuple of arguments. However, memoize does not enforce that because sometimes it is useful to memoize an impure function, too.

To memoize a recursive function, simply insert the memoized call in lieu of the plain recursive call. For example, to transform the exponential-time Fibonacci implementation into a linear-time computation:

ulong fib(ulong n)
    alias mfib = memoize!fib;
    return n < 2 ? 1 : mfib(n - 2) + mfib(n - 1);
assert(fib(10) == 89);

To improve the speed of the factorial function,

ulong fact(ulong n)
    alias mfact = memoize!fact;
    return n < 2 ? 1 : n * mfact(n - 1);
assert(fact(10) == 3628800);

This memoizes all values of fact up to the largest argument. To only cache the final result, move memoize outside the function as shown below.

ulong factImpl(ulong n)
    return n < 2 ? 1 : n * factImpl(n - 1);
alias fact = memoize!factImpl;
assert(fact(10) == 3628800);

The maxSize parameter is a cutoff for the cache size. If upon a miss the length of the hash table is found to be maxSize, the table is simply cleared.

// Memoize no more than 128 values of transmogrify
alias fastTransmogrify = memoize!(transmogrify, 128);

auto toDelegate(F)(auto ref F fp) if (isCallable!F);
Convert a callable to a delegate with the same parameter list and return type, avoiding heap allocations and use of auxiliary storage.

void doStuff() {
    writeln("Hello, world.");

void runDelegate(void delegate() myDelegate) {

auto delegateToPass = toDelegate(&doStuff);
runDelegate(delegateToPass);  // Calls doStuff, prints "Hello, world."

  • Does not work with @safe functions.
  • Ignores C-style / D-style variadic arguments.