Associative Arrays

Associative arrays have an index that is not necessarily an integer, and can be sparsely populated. The index for an associative array is called the key, and its type is called the KeyType.

Associative arrays are declared by placing the KeyType within the [ ] of an array declaration:

int[string] aa;   // Associative array of ints that are
                  // indexed by string keys.
                  // The KeyType is string.
aa["hello"] = 3;  // set value associated with key "hello" to 3
int value = aa["hello"];  // lookup value from a key
assert(value == 3);

Neither the KeyTypes nor the element types of an associative array can be function types or void.

Literals

auto aa = [21u: "he", 38: "ho", 2: "hi"];
static assert(is(typeof(aa) == string[uint]));
assert(aa[2] == "hi");

See Associative Array Literals.

Removing Keys

Particular keys in an associative array can be removed with the remove function:

aa.remove("hello");

remove(key) does nothing if the given key does not exist and returns false. If the given key does exist, it removes it from the AA and returns true.

All keys can be removed by using the method clear.

Testing Membership

The InExpression yields a pointer to the value if the key is in the associative array, or null if not:

int* p;

p = "hello" in aa;
if (p !is null)
{
    *p = 4;  // update value associated with key
    assert(aa["hello"] == 4);
}

Using Classes as the KeyType

Classes can be used as the KeyType. For this to work, the class definition must override the following member functions of class Object:

• size_t toHash() @trusted nothrow
• bool opEquals(Object)

Note that the parameter to opEquals is of type Object, not the type of the class in which it is defined.

For example:

class Foo
{
    int a, b;

    override size_t toHash() { return a + b; }

    override bool opEquals(Object o)
    {
        Foo foo = cast(Foo) o;
        return foo && a == foo.a && b == foo.b;
    }
}

Using Structs or Unions as the KeyType

If the KeyType is a struct or union type, a default mechanism is used to compute the hash and comparisons of it based on the fields of the struct value. A custom mechanism can be used by providing the following functions as struct members:

size_t toHash() const @safe pure nothrow;
bool opEquals(ref const typeof(this) s) const @safe pure nothrow;

For example:

import std.string;

struct MyString
{
    string str;

    size_t toHash() const @safe pure nothrow
    {
        size_t hash;
        foreach (char c; str)
            hash = (hash * 9) + c;
        return hash;
    }

    bool opEquals(ref const MyString s) const @safe pure nothrow
    {
        return std.string.cmp(this.str, s.str) == 0;
    }
}

The functions can use @trusted instead of @safe.

Construction or Assignment on Setting AA Entries

When an AA indexing access appears on the left side of an assignment operator, it is specially handled for setting an AA entry associated with the key.

string[int] aa;
string s;

//s = aa[1];        // throws RangeError in runtime
aa[1] = "hello";    // handled for setting AA entry
s = aa[1];          // succeeds to lookup
assert(s == "hello");

If the assigned value type is equivalent with the AA element type:

1. If the indexing key does not yet exist in AA, a new AA entry will be allocated, and it will be initialized with the assigned value.
2. If the indexing key already exists in the AA, the setting runs normal assignment.

struct S
{
    int val;
    void opAssign(S rhs) { this.val = rhs.val * 2; }
}
S[int] aa;
aa[1] = S(10);  // first setting initializes the entry aa[1]
assert(aa[1].val == 10);
aa[1] = S(10);  // second setting invokes normal assignment, and
                // operator-overloading rewrites it to member opAssign function.
assert(aa[1].val == 20);

If the assigned value type is not equivalent with the AA element type, the expression could invoke operator overloading with normal indexing access:

struct S
{
    int val;
    void opAssign(int v) { this.val = v * 2; }
}
S[int] aa;
aa[1] = 10;     // is rewritten to: aa[1].opAssign(10), and
                // throws RangeError before opAssign is called

However, if the AA element type is a struct which supports an implicit constructor call from the assigned value, implicit construction is used for setting the AA entry:

struct S
{
    int val;
    this(int v) { this.val = v; }
    void opAssign(int v) { this.val = v * 2; }
}
S s = 1;    // OK, rewritten to: S s = S(1);
s = 1;      // OK, rewritten to: s.opAssign(1);

S[int] aa;
aa[1] = 10; // first setting is rewritten to: aa[1] = S(10);
assert(aa[1].val == 10);
aa[1] = 10; // second setting is rewritten to: aa[1].opAssign(10);
assert(aa[1].val == 20);

import std.bigint;
BigInt[string] aa;
aa["a"] = 10;   // construct BigInt(10) and move it in AA
aa["a"] = 20;   // call aa["a"].opAssign(20)

Inserting if not present

When AA access requires that there must be a value corresponding to the key, a value must be constructed and inserted if not present. The require function provides a means to construct a new value via a lazy argument. The lazy argument is evaluated when the key is not present. The require operation avoids the need to perform multiple key lookups.

class C{}
C[string] aa;

auto a = aa.require("a", new C);   // lookup "a", construct if not present

Sometimes it is necessary to know whether the value was constructed or already exists. The require function doesn't provide a boolean parameter to indicate whether the value was constructed but instead allows the construction via a function or delegate. This allows the use of any mechanism as demonstrated below.

class C{}
C[string] aa;

bool constructed;
auto a = aa.require("a", { constructed=true; return new C;}());
assert(constructed == true);

C newc;
auto b = aa.require("b", { newc = new C; return newc;}());
assert(b is newc);

Advanced updating

Typically updating a value in an associative array is simply done with an assign statement.

int[string] aa;

aa["a"] = 3;  // set value associated with key "a" to 3

Sometimes it is necessary to perform different operations depending on whether a value already exists or needs to be constructed. The update function provides a means to construct a new value via the create delegate or update an existing value via the update delegate. The update operation avoids the need to perform multiple key lookups.

int[string] aa;

// create
aa.update("key",
    () => 1,
    (int) {} // not executed
    );
assert(aa["key"] == 1);

// update value by ref
aa.update("key",
    () => 0, // not executed
    (ref int v) {
        v += 1;
    });
assert(aa["key"] == 2);

For details, see update.

Runtime Initialization of Immutable AAs

Immutable associative arrays are often desirable, but sometimes initialization must be done at runtime. This can be achieved with a constructor (static constructor depending on scope), a buffer associative array and assumeUnique:

immutable long[string] aa;

shared static this()
{
    import std.exception : assumeUnique;
    import std.conv : to;

    long[string] temp; // mutable buffer
    foreach (i; 0 .. 10)
    {
        temp[to!string(i)] = i;
    }
    temp.rehash; // for faster lookups

    aa = assumeUnique(temp);
}

void main()
{
    assert(aa["1"] == 1);
    assert(aa["5"] == 5);
    assert(aa["9"] == 9);
}

Construction and Reference Semantics

An Associative Array defaults to null, and is constructed upon assigning the first key/value pair. However, once constructed, an associative array has reference semantics, meaning that assigning one array to another does not copy the data. This is especially important when attempting to create multiple references to the same array.

int[int] aa;             // defaults to null
int[int] aa2 = aa;       // copies the null reference

assert(aa is null);
aa[1] = 1;
assert(aa2.length == 0); // aa2 still is null
aa2 = aa;
aa2[2] = 2;
assert(aa[2] == 2);      // now both refer to the same instance

A NewExpression allows constructing an associative array instance immediately, rather than when inserting a key.

int[string] a = new int[string];
auto b = a; // a and b point to the same AA instance

assert(b !is null);
a["1"] = 1;
assert(b["1"] == 1);

Properties

Properties for associative arrays are:

Examples

Associative Array Example: word count

import std.algorithm;
import std.stdio;

void main()
{
    ulong[string] dictionary;
    ulong wordCount, lineCount, charCount;

    foreach (line; stdin.byLine(KeepTerminator.yes))
    {
        charCount += line.length;
        foreach (word; splitter(line))
        {
            wordCount += 1;
            if (auto count = word in dictionary)
                *count += 1;
            else
                dictionary[word.idup] = 1;
        }

        lineCount += 1;
    }

    writeln("   lines   words   bytes");
    writefln("%8s%8s%8s", lineCount, wordCount, charCount);

    const char[37] hr = '-';

    writeln(hr);
    foreach (word; sort(dictionary.keys))
    {
        writefln("%3s %s", dictionary[word], word);
    }
}

See wc for the full version.

Associative Array Example: counting pairs

An Associative Array can be iterated in key/value fashion using a foreach statement. As an example, the number of occurrences of all possible substrings of length 2 (aka 2-mers) in a string will be counted:

import std.range : slide;
import std.stdio : writefln;
import std.utf : byCodeUnit; // avoids UTF-8 auto-decoding

int[string] aa;

// The string `arr` has a limited alphabet: {A, C, G, T}
// Thus, for better performance, iteration can be done _without_ decoding
auto arr = "AGATAGA".byCodeUnit;

// iterate over all pairs in the string and count each pair
// ('A', 'G'), ('G', 'A'), ('A', 'T'), ...
foreach (window; arr.slide(2))
    aa[window.source]++; // source unwraps the code unit range

// iterate over all key/value pairs of the Associative Array
foreach (key, value; aa)
{
    writefln("key: %s, value: %d", key, value);
}

> rdmd count.d
key: AT, value: 1
key: GA, value: 2
key: TA, value: 1
key: AG, value: 2