Subscripts

Accessing data structure elements by index or key

Table of Contents
1	Basics
1.1	Positional subscripting
1.2	Associative subscripting
1.3	Applying subscripts
2	Nonexistent elements
3	From the end
4	Slices
4.1	Truncating slices
4.2	Zen slices
5	Multiple dimensions
6	Modifying elements
7	Autovivification
8	Binding
9	Adverbs
9.1	`:exists`
9.2	`:delete`
9.3	`:p`
9.4	`:kv`
9.5	`:k`
9.6	`:v`
10	Custom types
10.1	Custom type example
10.2	Methods to implement for positional subscripting
10.2.1	method elems
10.2.2	method AT-POS
10.2.3	method EXISTS-POS
10.2.4	method DELETE-POS
10.2.5	method ASSIGN-POS
10.2.6	method BIND-POS
10.2.7	method STORE
10.3	Methods to implement for associative subscripting
10.3.1	method AT-KEY
10.3.2	method EXISTS-KEY
10.3.3	method DELETE-KEY
10.3.4	method ASSIGN-KEY
10.3.5	method BIND-KEY
10.3.6	method STORE

One often needs to refer to a specific element (or slice of elements) from a collection or data structure. Borrowing from mathematical notation where the components of a vector v would be referred to as v₁, v₂, v₃, this concept is called "subscripting" (or "indexing") in Perl 6.

Basics

Perl 6 provides two universal subscripting interfaces:

	elements are identified by	interface name	supported by
[ ]	zero-based indices	Positional	Array, List, Buf, Match, ...
{ }	string or object keys	Associative	Hash, Bag, Mix, Match, ...

Positional subscripting

Positional subscripting(via postcircumfix [ ]) addresses elements of an ordered collection by their position. Index 0 refers to the first element, index 1 to the second, and so on:

my @chores = "buy groceries", "feed dog", "wash car";
say @chores[0];  # OUTPUT: «buy groceries␤» 
say @chores[1];  # OUTPUT: «feed dog␤» 
say @chores[2];  # OUTPUT: «wash car␤»

Associative subscripting

Associative subscripting (via postcircumfix { }), does not require the collection to keep its elements in any particular order - instead, it uses a unique key to address each value. The nature of the keys depends on the collection in question: For example a standard Hash uses string keys, whereas a Mix allows arbitrary objects as keys, etc.:

my %grade = Zoe => "C", Ben => "B+";
say %grade{"Zoe"};  # OUTPUT: «C␤» 
say %grade{"Ben"};  # OUTPUT: «B+␤» 
 
my $stats = ( Date.today => 4.18, Date.new(2015,  4,  5) => 17.253 ).Mix;
say $stats{ Date.new(2015, 4, 4) + 1 };  # OUTPUT: «17.253␤»

For passing single-word string keys to { }, you can also use the angle bracketed word quoting constructs as if they were postcircumfix operators:

my %grade = Zoe => "C", Ben => "B+";
say %grade<Zoe>;    # OUTPUT: «C␤» 
say %grade<Ben>;    # OUTPUT: «B+␤»

This is really just syntactic sugar that gets turned into the corresponding { } form at compile-time:

%hash<foo bar>;       # same as %hash{ <foo bar> } 
%hash«foo "$var"»;    # same as %hash{ «foo "$var"» } 
%hash<<foo "$var">>;  # same as %hash{ <<foo "$var">> }

You may have noted above that we avoided having to quote Zoe by using the => operator, but that same operator did not just put invisible quotes around Date.new(2015, 4, 5), and we were able to find the same element using $stats{ Date.new(2015, 4, 4) + 1 }. This is because => only puts invisible quotes around single words, and by "word" we mean an identifier/name. The => operator is there to prevent us from accidentally calling functions or using constants with that name.

Hash subscripts do not do the same thing as =>. The default Hash has been made to behave the way new users have come to expect from using other languages, and for general ease of use. On a default Hash, subscripts coerce keys into strings, as long as those keys produce something Cool. You can use .perl on a collection to be sure whether the keys are strings or objects:

( 1  => 1 ).perl.say;            # OUTPUT: «1 => 1␤» 
my %h; %h{1}   = 1; say %h.perl; # OUTPUT: «{ "1" => 1 }␤» 
( 1/2 => 1 ).perl.say;           # OUTPUT: «0.5 => 1␤» 
my %h; %h{1/2} = 1; say %h.perl; # OUTPUT: «{ "0.5" => 1 }␤» 
( pi => 1 ).perl.say;            # OUTPUT: «:pi(1)␤» 
my %h; %h{pi}  = 1; say %h.perl; # OUTPUT: «{ "3.14159265358979" => 1 }␤»

While the invisible quotes around single names is built into =>, string conversion is not built into the curly braces: it is a behavior of the default Hash. Not all types of hashes or collections do so:

my %h := MixHash.new;
%h{pi} = 1; %h.perl.say;         # OUTPUT: «(3.14159265358979e0=>1).MixHash␤»

(Any name that => would convert to a string can also be used to build a pair using "adverbial notation" and will appear that way when viewed through .perl, which is why we see :pi(1) above.)

Applying subscripts

Subscripts can be applied to any expression that returns a subscriptable object, not just to variables:

say "__Hello__".match(/__(.*)__/)[0];   # OUTPUT: «｢Hello｣␤» 
say "__Hello__".match(/__(.*)__/).[0];  # same, in method notation

Positional and associative subscripting are not mutually exclusive - for example, Match objects support both (each accessing a different set of data). Also, to make list processing more convenient, class Any provides a fallback implementation for positional subscripts which simply treats the invocant as a list of one element. (But there's no such fallback for associative subscripts, so they throw a runtime error when applied to an object that does not implement support for them.)

say 42[0];    # OUTPUT: «42␤» 
say 42<foo>;  # ERROR: Type Int does not support associative indexing.

Nonexistent elements

What happens when a nonexistent element is addressed by a subscript is up to the collection type in question. Standard Array and Hash collections return the type object of their value type constraint (which, by default, is Any) unless the collection has been declared with the is default trait in which case the returned value will be whatever the programmer declared it to be:

# no default values specified 
my @array1;     say @array1[10];                   # OUTPUT: «(Any)␤» 
my Int @array2; say @array2[10];                   # OUTPUT: «(Int)␤» 
 
my %hash1;      say %hash1<foo>;                   # OUTPUT: «(Any)␤» 
my Int %hash2;  say %hash2<foo>;                   # OUTPUT: «(Int)␤» 
 
# default values specified 
my @array3 is default('--');     say @array3[10];  # OUTPUT: «--␤» 
my Int @array4 is default(17);   say @array4[10];  # OUTPUT: «17␤» 
 
my %hash3 is default('Empty');   say %hash3<foo>;  # OUTPUT: «Empty␤» 
my Int %hash4 is default(4711);  say %hash4<foo>;  # OUTPUT: «4711␤»

However, other types of collections may react differently to subscripts that address nonexistent elements:

say (0, 10, 20)[3];           # OUTPUT: «Nil␤» 
say bag(<a a b b b>)<c>;      # OUTPUT: «0␤» 
say array[uint8].new(1, 2)[2] # OUTPUT: «0␤»

To silently skip nonexistent elements in a subscripting operation, see Truncating slices and the :v adverb.

From the end

Positional indices are counted from the start of the collection, but there's also a notation for addressing elements by their position relative to the end: *-1 refers to the last element, *-2 to the second-to-last element, and so on.

my @alphabet = 'A' .. 'Z';
say @alphabet[*-1];  # OUTPUT: «Z␤» 
say @alphabet[*-2];  # OUTPUT: «Y␤» 
say @alphabet[*-3];  # OUTPUT: «X␤»

Note: The asterisk is important. Passing a bare negative integer (e.g. @alphabet[-1]) like you would do in many other programming languages, throws an error in Perl 6.

What actually happens here, is that an expression like *-1 declares a code object via Whatever-currying - and the [ ] subscript reacts to being given a code object as an index, by calling it with the length of the collection as argument and using the result value as the actual index. In other words, @alphabet[*-1] becomes @alphabet[@alphabet.elems - 1].

This means that you can use arbitrary expressions which depend on the size of the collection:

say @array[* div 2];  # select the middlemost element 
say @array[$i % *];   # wrap around a given index ("modular arithmetic") 
say @array[ -> $size { $i % $size } ];  # same as previous

Slices

When multiple elements of a collection need to be accessed, there's a shortcut to doing multiple separate subscripting operations: Simply specify a list of indices/keys in the subscript, to get back a list of elements - also called a "slice" - in the same order.

For positional slices, you can mix normal indices with from-the-end ones:

my @alphabet = 'a' .. 'z';
say @alphabet[15, 4, *-9, 11].perl;  # OUTPUT: «("p", "e", "r", "l")␤»

In the *-number construct above, * indicates the end of the array as explained above in the From the end section. So if you want to take the last N elements of the array, you will have to create a Range that includes it.

(5802..5830).map( {.chr} )[*-10..*-5] # OUTPUT:  «(ᚽ ᚾ ᚿ ᛀ ᛁ ᛂ)␤»

Using * as the last element of the range will, effectively, return the last elements of the collection.

For associative slices, the angle brackets form often comes in handy:

my %color = kiwi => "green", banana => "yellow", cherry => "red";
say %color{"cherry", "kiwi"};  # OUTPUT: «(red green)␤» 
say %color<cherry kiwi>;       # OUTPUT: «(red green)␤» 
say %color{*};                 # OUTPUT: «(red yellow green)␤»

Be aware that slices are controlled by the type of what is passed to (one dimension of) the subscript, not its length. In particular the type can be any of the following:

a lazy Iterable, that truncates in [ ]
accordingly, an infinite Range will truncate, but a finite one produces a normal slice
'*' (whatever-star), that returns the full slice (as if all keys/indices were specified)
any other object, that provides a single-element access rather than a slice
Callable, whatever is returned by the callable (this can lead to recursion)
empty, the full slice known as Zen slice
any iterable different from the above ones, normal slice

The notable difference between * and Zen slice (empty) is that the Whatever star will cause full reification or itemization, while Zen slice won't. Both versions also de-cont.

So even a one-element list returns a slice, whereas a bare scalar value doesn't:

say @alphabet[2,];        # OUTPUT: «(c)␤» 
say @alphabet[2,].^name;  # OUTPUT: «List␤» 
 
say @alphabet[2];         # OUTPUT: «c␤» 
say @alphabet[2].^name;   # OUTPUT: «Str␤»

(The angle bracket form for associative subscripts works out because word quoting conveniently returns a Str in case of a single word, but a List in case of multiple words.)

In fact, the list structure of (the current dimension of) the subscript is preserved across the slice operation (but the kind of Iterable is not – the result is always just lists.)

say @alphabet[0, (1..2, (3,))];       # OUTPUT: «(a ((b c) (d)))␤» 
say @alphabet[0, (1..2, [3,])];       # OUTPUT: «(a ((b c) (d)))␤» 
say @alphabet[flat 0, (1..2, (3,))];  # OUTPUT: «(a b c d)␤» 
say flat @alphabet[0, (1..2, (3,))];  # OUTPUT: «(a b c d)␤»

Truncating slices

Referring to nonexistent elements in a slice subscript causes the output List to contain undefined values (or whatever else the collection in question chooses to return for nonexistent elements):

my  @letters = <a b c d e f>;
say @letters[3..7];  # OUTPUT: «(d e f (Any) (Any))␤»

This behavior, while at first glance may seem unintuitive, is desirable in instances when you want to assign a value at an index in which a value does not currently exist.

my  @letters;
say @letters; # OUTPUT: «[]␤» 
 
@letters[^10] = 'a'..'z';
say @letters; # OUTPUT: «[a b c d e f g h i j]␤»

If you want the resulting slice to only include existing elements, you can silently skip the non-existent elements using the :v adverb.

my  @letters = <a b c d e f>;
say @letters[3..7]:v;  # OUTPUT: «(d e f)␤»

The behavior when indexing a collection via lazy subscripts is different than when indexing with their eager counterparts. When accessing via a lazy subscript, the resulting slice will be truncated.

say @letters[lazy 3..7]; # OUTPUT: «(d e f)␤» 
say @letters[     3..*]; # OUTPUT: «(d e f)␤»

This behavior exists as a precaution to prevent runaway generation of massive, potentially infinite Lists and the out-of-memory issues that occur as a result.

Zen slices

If you put the subscript operator behind an object without specifying any indices/keys at all, it simply returns the subscripted object itself. Since it is empty but returns everything, it is known as a Zen slice.

Zen slicing is different from passing a Whatever-star (which, like a normal slice, always returns a List of elements no matter the type of the original object) or an empty list (which returns an empty slice):

my %bag := (orange => 1, apple => 3).Bag;
say %bag<>;    # OUTPUT: «Bag(apple(3), orange)␤» 
say %bag{};    # OUTPUT: «Bag(apple(3), orange)␤» 
say %bag{*};   # OUTPUT: «(1 3)␤» 
say %bag{()};  # OUTPUT: «()␤»

Zen slicing does not reify or cache and merely returns the invocant. It is usually used to interpolate entire arrays / hashes into strings or to decont.

my @words = "cruel", "world";
say "Hello, @words[]!";  # OUTPUT: «Hello, cruel world!␤» 
 
my $list = <a b c>;
.say for $list;   # OUTPUT: «(a b c)␤» 
.say for $list<>; # OUTPUT: «a␤b␤c␤» 
.say for $list[]; # OUTPUT: «a␤b␤c␤»

Multiple dimensions

Dimensions in subscripts are separated by a semicolon, allowing to mix lists of elements and dimensions.

my @twodim = (<a b c>, (1, 2, 3));
say @twodim;
# OUTPUT: «[(a b c) (1 2 3)]␤» 
 
say @twodim[0,1;1]; # 2nd element of both lists 
# OUTPUT: «(b 2)␤» 
 
my %pantheon = %('Baldr' => 'consort' => 'Nanna' ,
                 'Bragi' => 'consort' => 'Iðunn' ,
                 'Nótt'  => 'consort' => 'Dellingr' );
say %pantheon{'Bragi','Nótt';'consort'}; # 'consort' value for both keys 
# OUTPUT: «(Iðunn Dellingr)␤»

Multidimensional subscripts can be used to flatten nested lists when combined with Whatever.

my @toomany = [[<a b>], [1, 2]];
say @toomany;
# OUTPUT: «[[a b] [1 2]]␤» 
 
say @toomany[*;*];
# OUTPUT: «(a b 1 2)␤»

You can use as many flattening semicolons as you want; there will be, at most, as many nesting levels flattened as the number of semicolons:

say [[1,2,[3,4]],[4,5]][*;*];     # OUTPUT: «(1 2 [3 4] 4 5)␤» 
say [[1,2,[3,4]],[4,5]][*;*;*;*]; # OUTPUT: «(1 2 3 4 4 5)␤»

In the first example, with one Whatever less than the number of levels, the deepest one will not be flattened; in the second case it is, since it's greater than the number of levels.

You can use Whatever to select ranges or "rows" in multidimensional subscripts.

my @a = [[1,2], [3,4]];
say @a[*;1]; # 2nd element of each sub list 
# OUTPUT: «(2 4)␤» 
my @a = (<1 c 6>, <2 a 4>, <5 b 3>);
say @a.sort( { $_[1] } ); # sort by 2nd column 
# OUTPUT: «((2 a 4) (5 b 3) (1 c 6))␤»

Modifying elements

Autovivification

Subscripts participate in "autovivification", i.e. the process by which arrays and hashes automatically spring into existence when needed, so that you can build nested data structures without having to pre-declare the collection type at each level:

my $beatles;
 
$beatles{"White Album"}[0] = "Back in the U.S.S.R.";  # autovivification! 
 
say $beatles.perl;  # OUTPUT: «${"White Album" => $["Back in the U.S.S.R."]}␤»

$beatles started out undefined, but became a Hash object because it was subscripted with { } in the assignment. Similarly, $beatles{"White Album"} became an Array object due to being subscripted with [ ] in the assignment.

Note that the subscripting itself does not cause autovivification: It only happens when the result of the subscripting chain is assigned to (or otherwise mutated).

Binding

A subscripting expression may also be used as the left-hand-side of a binding statement. If supported by the subscripted collection's type, this replaces whatever value container would be naturally found at that "slot" of the collection, with the specified container.

The built-in Array and Hash types support this in order to allow building complex linked data structures:

my @a = 10, 11, 12, 13;
my $x = 1;
 
@a[2] := $x;  # Bound! (@a[2] and $x refer to the same container now.) 
 
$x++; @a[2]++;
 
say @a;  # OUTPUT: «[10 11 3 13]␤» 
say $x;  # OUTPUT: «3␤»

This can be specially useful when lazy data structures are part of a bigger one.

my @fib = 1,1, * + * … ∞;
my @lucas = 1,3, * + * … ∞;
my %sequences;
%sequences<f> := @fib;
%sequences<l> := @lucas;
for %sequences.keys -> $s {
    for ^10 -> $n {
        say %sequences{$s}[100+$n*10]/%sequences{$s}[101+$n*10];
    }
}
# OUTPUT: 0.6180339887498949 times 20.

In this case, hash keys are bound to lazily generated sequences. The fact that they are bound means that whatever state has been computed is shared by the hash value and the sequence it's bound to, making computations of subsequent elements faster.

See method BIND-POS and method BIND-KEY for the underlying mechanism.

Adverbs

The return value and possible side-effect of a subscripting operation can be controlled using adverbs; these are defined on the relevant subscript operators.

Beware of the relatively loose precedence of operator adverbs, which may require you to add parentheses in compound expressions:

if $foo || %hash<key>:exists { ... }    # WRONG, tries to adverb the || op

if $foo || (%hash<key>:exists) { ... }  # correct 
if $foo or %hash<key>:exists { ... }    # also correct

The supported adverbs are:

`:exists`

Returns whether or not the requested element exists, instead of returning the element's actual value. This can be used to distinguish between elements with an undefined value, and elements that aren't part of the collection at all:

my @foo = Any, 10;
say @foo[0].defined;    # OUTPUT: «False␤» 
say @foo[0]:exists;     # OUTPUT: «True␤» 
say @foo[2]:exists;     # OUTPUT: «False␤» 
say @foo[0, 2]:exists;  # OUTPUT: «(True False)␤» 
 
my %fruit = apple => Any, orange => 10;
say %fruit<apple>.defined;       # OUTPUT: «False␤» 
say %fruit<apple>:exists;        # OUTPUT: «True␤» 
say %fruit<banana>:exists;       # OUTPUT: «False␤» 
say %fruit<apple banana>:exists; # OUTPUT: «(True False)␤»

May also be negated to test for non-existence:

say %fruit<apple banana>:!exists; # OUTPUT: «(False True)␤»

To check if all elements of a slice exist, use an all junction:

if all %fruit<apple orange banana>:exists { ... }

It can be used on multi-dimensional arrays and hashes:

my @multi-dim = 1, [2, 3, [4, 5]];
say @multi-dim[1;2;0]:exists;                # OUTPUT: «True␤» 
say @multi-dim[1;2;5]:exists;                # OUTPUT: «False␤» 
 
my %multi-dim = 1 => { foo => { 3 => 42 } };
say %multi-dim{1;'foo';3}:exists;            # OUTPUT: «True␤» 
say %multi-dim{1;'bar';3}:exists;            # OUTPUT: «False␤»

:exists can be combined with the :delete and :p/:kv adverbs - in which case the behavior is determined by those adverbs, except that any returned element value is replaced with the corresponding Bool indicating element existence.

See method EXISTS-POS and method EXISTS-KEY for the underlying mechanism.

`:delete`

Delete the element from the collection or, if supported by the collection, creates a hole at the given index, in addition to returning its value.

my @tens = 0, 10, 20, 30;
say @tens[3]:delete;     # OUTPUT: «30␤» 
say @tens;               # OUTPUT: «[0 10 20]␤» 
 
my %fruit = apple => 5, orange => 10, banana => 4, peach => 17;
say %fruit<apple>:delete;         # OUTPUT: «5␤» 
say %fruit<peach orange>:delete;  # OUTPUT: «(17 10)␤» 
say %fruit;                       # OUTPUT: «{banana => 4}␤»

Note that assigning Nil will revert the container at the given index to its default value. It will not create a hole. The created holes can be tested for with :exists but iteration will not skip them and produce undefined values instead.

my @a = 1, 2, 3;
@a[1]:delete;
say @a[1]:exists;
# OUTPUT: «False␤» 
.say for @a;
# OUTPUT: «1␤(Any)␤3␤»

With the negated form of the adverb, the element is not actually deleted. This means you can pass a flag to make it conditional:

say %fruit<apple> :delete($flag);  # deletes the element only if $flag is 
                                   # true, but always returns the value.

Can be combined with the :exists and :p/:kv/:k/:v adverbs - in which case the return value will be determined by those adverbs, but the element will at the same time also be deleted.

See method DELETE-POS and method DELETE-KEY for the underlying mechanism.

`:p`

Return both the index/key and the value of the element, in the form of a Pair, and silently skip nonexistent elements:

my  @tens = 0, 10, 20, 30;
say @tens[1]:p;        # OUTPUT: «1 => 10␤» 
say @tens[0, 4, 2]:p;  # OUTPUT: «(0 => 0 2 => 20)␤» 
 
my  %month = Jan => 1, Feb => 2, Mar => 3;
say %month<Feb>:p;          # OUTPUT: «Feb => 2␤» 
say %month<Jan Foo Mar>:p;  # OUTPUT: «(Jan => 1 Mar => 3)␤»

If you don't want to skip nonexistent elements, use the negated form:

say %month<Jan Foo Mar>:!p;  # OUTPUT: «(Jan => 1 Foo => (Any) Mar => 3)␤»

Can be combined with the :exists and :delete adverbs.

`:kv`

Return both the index/key and the value of the element, in the form of a List, and silently skip nonexistent elements. When used on a slice, the return value is a single flat list of interleaved keys and values:

my  @tens = 0, 10, 20, 30;
say @tens[1]:kv;        # OUTPUT: «(1 10)␤» 
say @tens[0, 4, 2]:kv;  # OUTPUT: «(0 0 2 20)␤» 
 
my  %month = Jan => 1, Feb => 2, Mar => 3;
say %month<Feb>:kv;          # OUTPUT: «(Feb 2)␤» 
say %month<Jan Foo Mar>:kv;  # OUTPUT: «(Jan 1 Mar 3)␤»

If you don't want to skip nonexistent elements, use the negated form:

say %month<Jan Foo Mar>:!kv;  # OUTPUT: «(Jan 1 Foo (Any) Mar 3)␤»

This adverb is commonly used to iterate over slices:

for %month<Feb Mar>:kv -> $month, $i {
    say "$month had {Date.new(2015, $i, 1).days-in-month} days in 2015"
}

Can be combined with the :exists and :delete adverbs.

`:k`

Return only the index/key of the element, rather than its value, and silently skip nonexistent elements:

my @tens = 0, 10, 20, 30;
say @tens[1]:k;        # OUTPUT: «1␤» 
say @tens[0, 4, 2]:k;  # OUTPUT: «(0 2)␤» 
 
my %month = Jan => 1, Feb => 2, Mar => 3;
say %month<Feb>:k;          # OUTPUT: «Feb␤» 
say %month<Jan Foo Mar>:k;  # OUTPUT: «(Jan Mar)␤»

If you don't want to skip nonexistent elements, use the negated form:

say %month<Jan Foo Mar>:!k;  # OUTPUT: «(Jan Foo Mar)␤»

`:v`

Return the bare value of the element (rather than potentially returning a mutable value container), and silently skip nonexistent elements:

my @tens = 0, 10, 20, 30;
say @tens[1]:v;        # OUTPUT: «10␤» 
say @tens[0, 4, 2]:v;  # OUTPUT: «(0, 20)␤» 
@tens[3] = 31;         # OK 
@tens[3]:v = 31;       # ERROR, Cannot modify an immutable Int (31) 
 
my %month = Jan => 1, Feb => 2, Mar => 3;
say %month<Feb>:v;          # OUTPUT: «2␤» 
say %month<Jan Foo Mar>:v;  # OUTPUT: «(1 3)␤»

If you don't want to skip nonexistent elements, use the negated form:

say %month<Jan Foo Mar>:!v;  # OUTPUT: «(1 (Any) 3)␤»

Custom types

The subscripting interfaces described on this page are not meant to be exclusive to Perl 6's built-in collection types - you can (and should) reuse them for any custom type that wants to provide access to data by index or key.

You don't have to manually overload the postcircumfix [ ] and postcircumfix { } operators and re-implement all their magic, to achieve that - instead, you can rely on the fact that their standard implementation dispatches to a well-defined set of low-level methods behind the scenes. For example:

when you write:	this gets called behind the scenes:
%foo<aa>	%foo.AT-KEY("aa")
%foo<aa>:delete	%foo.DELETE-KEY("aa")
@foo[3, 4, 5]	@foo.AT-POS(3), @foo.AT-POS(4), @foo.AT-POS(5)
@foo[*-1]	@foo.AT-POS(@foo.elems - 1)

So in order to make subscripting work, you only have to implement or delegate those low-level methods (detailed below) for your custom type.

If you do, you should also let your type compose the Positional or Associative role, respectively. This doesn't add any functionality per se, but announces (and may be used to check) that the type implements the corresponding subscripting interface.

Custom type example

Imagine a HTTP::Header type which, despite being a custom class with special behavior, can be indexed like a hash:

my $request = HTTP::Request.new(GET => "perl6.org");
say $request.header.^name;  # OUTPUT: «HTTP::Header␤» 
 
$request.header<Accept> = "text/plain";
$request.header{'Accept-' X~ <Charset Encoding Language>} = <utf-8 gzip en>;
$request.header.push('Accept-Language' => "fr");  # like .push on a Hash 
 
say $request.header<Accept-Language>.perl;  # OUTPUT: «["en", "fr"]␤» 
 
my $rawheader = $request.header.Str;  # stringify according to HTTP spec

A simple way to implement this class would be to give it an attribute of type Hash, and delegate all subscripting and iterating related functionality to that attribute (using a custom type constraint to make sure users don't insert anything invalid into it):

class HTTP::Header does Associative {
    subset StrOrArrayOfStr where Str | ( Array & {.all ~~ Str} );
 
    has %!fields of StrOrArrayOfStr
                 handles <AT-KEY EXISTS-KEY DELETE-KEY push
                          iterator list kv keys values>;
 
    method Str { #`[not shown, for brevity] }
}

However, HTTP header field names are supposed to be case-insensitive (and preferred in camel-case). We can accommodate this by taking the *-KEY and push methods out of the handles list, and implementing them separately like this:

method AT-KEY     ($key) is rw { %!fields{normalize-key $key}        }
method EXISTS-KEY ($key)       { %!fields{normalize-key $key}:exists }
method DELETE-KEY ($key)       { %!fields{normalize-key $key}:delete }
method push(*@_) { #`[not shown, for brevity] }
 
sub normalize-key ($key) { $key.subst(/\w+/, *.tc, :g) }

Note that subscripting %!fields returns an appropriate rw container, which our AT-KEY can simply pass on.

However, we may prefer to be less strict about user input and instead take care of sanitizing the field values ourselves. In that case, we can remove the StrOrArrayOfStr type constraint on %!fields, and replace our AT-KEY implementation with one that returns a custom Proxy container which takes care of sanitizing values on assignment:

multi method AT-KEY (::?CLASS:D: $key) is rw {
    my $element := %!fields{normalize-key $key};
 
    Proxy.new(
        FETCH => method () { $element },
 
        STORE => method ($value) {
            $element = do given $value».split(/',' \s+/).flat {
                when 1  { .[0] }    # a single value is stored as a string 
                default { .Array }  # multiple values are stored as an array 
            }
        }
    );
}

Note that declaring the method as multi and restricting it to :D (defined invocants) makes sure that the undefined case is passed through to the default implementation provided by Any (which is involved in auto-vivification).

Methods to implement for positional subscripting

In order to make index-based subscripting via postcircumfix [ ] work for your custom type, you should implement at least elems, AT-POS and EXISTS-POS - and optionally others as detailed below.

method elems

multi method elems(::?CLASS:D:)

Expected to return a number indicating how many subscriptable elements there are in the object. May be called by users directly, and is also called by postcircumfix [ ] when indexing elements from the end, as in @foo[*-1].

If not implemented, your type will inherit the default implementation from Any that always returns 1 for defined invocants - which is most likely not what you want. So if the number of elements cannot be known for your positional type, add an implementation that fails or dies, to avoid silently doing the wrong thing.

method AT-POS

multi method AT-POS (::?CLASS:D: $index)

Expected to return the element at position $index. This is what postcircumfix [ ] normally calls.

If you want an element to be mutable (like they are for the built-in Array type), you'll have to make sure to return it in the form of an item container that evaluates to the element's value when read, and updates it when assigned to. (Remember to use return-rw or the is rw routine trait to make that work; see the example.)

method EXISTS-POS

multi method EXISTS-POS (::?CLASS:D: $index)

Expected to return a Bool indicating whether or not there is an element at position $index. This is what postcircumfix [ ] calls when invoked like @foo[42]:exists.

What "existence" of an element means, is up to your type.

If you don't implement this, your type will inherit the default implementation from Any, which returns True for 0 and False for any other index - which is probably not what you want. So if checking for element existence cannot be done for your type, add an implementation that fails or dies, to avoid silently doing the wrong thing.

method DELETE-POS

multi method DELETE-POS (::?CLASS:D: $index)

Expected to delete the element at position $index, and return the value it had. This is what postcircumfix [ ] calls when invoked like @foo[42]:delete.

What "deleting" an element means, is up to your type.

Implementing this method is optional; if you don't, users trying to delete elements from an object of this type will get an appropriate error message.

method ASSIGN-POS

multi method ASSIGN-POS (::?CLASS:D: $index, $new)

Expected to set the element at position $index to the value $new. Implementing this is entirely optional; if you don't, self.AT-POS($index) = $new is used instead, and if you do, you should make sure it has the same effect.

This is meant as an opt-in performance optimization, so that simple assignments like @numbers[5] = "five" can operate without having to call AT-POS (which would have to create and return a potentially expensive container object).

Note that implementing ASSIGN-POS does not relieve you from making AT-POS an rw method though, because less trivial assignments/modifications such as @numbers[5]++ will still use AT-POS.

method BIND-POS

multi method BIND-POS (::?CLASS:D: $index, \new)

Expected to bind the value or container new to the slot at position $index, replacing any container that would be naturally found there. This is what is called when you write:

my $x = 10;
@numbers[5] := $x;

The generic Array class supports this in order to allow building complex linked data structures, but for more domain-specific types it may not make sense, so don't feel compelled to implement it. If you don't, users will get an appropriate error message when they try to bind to a positional slot of an object of this type.

method STORE

method STORE (::?CLASS:D: \values, :$initialize)

This method should only be supplied if you want to support this syntax:

my @a is Foo = 1,2,3;

Which is used for binding your implementation of the Positional role.

STORE should accept the values to (re-)initialize the object with. The optional named parameter will contain a True value when the method is called on the object for the first time. It should return the invocant.

class DNA {
    has $.chain is rw;
 
    method STORE ($chain where {
                         $chain ~~ /^^ <[ACGT]>+ $$ / and
                         $chain.chars %% 3
                     }, :$initialize --> DNA) {
        if $initialize {
            self= DNA.new( chain => $chain )
        } else {
            self.chain = $chain;
            self
        }
    }
 
    method Str(::?CLASS:D:) { return $.chain.comb.rotor(3) }
};
 
my @string is DNA = 'GAATCC';
say @string.Str;    # OUTPUT: «((G A A) (T C C))␤» 
@string = 'ACGTCG';
say @string.Str;    # OUTPUT: «((A C G) (T C G))␤»

This code takes into account the value of $initialize, which is set to True only if we are assigning a value to a variable declared using the is syntax for the first time. The STORE method should set the self variable and return it in all cases, including when the variable has already been initialized.

Methods to implement for associative subscripting

In order to make key-based subscripting via postcircumfix { } work for your custom type, you should implement at least AT-KEY and EXISTS-KEY - and optionally others as detailed below.

method AT-KEY

multi method AT-KEY (::?CLASS:D: $key)

Expected to return the element associated with $key. This is what postcircumfix { } normally calls.

If you want an element to be mutable (like they are for the built-in Hash type), you'll have to make sure to return it in the form of an item container that evaluates to the element's value when read, and updates it when assigned to. (Remember to use return-rw or the is rw routine trait to make that work; see the example.)

On the other hand if you want your collection to be read-only, feel free to return non-container values directly.

method EXISTS-KEY

multi method EXISTS-KEY (::?CLASS:D: $key)

Expected to return a Bool indicating whether or not there is an element associated with $key. This is what postcircumfix { } calls when invoked like %foo<aa>:exists.

What "existence" of an element means, is up to your type.

If you don't implement this, your type will inherit the default implementation from Any, which always returns False - which is probably not what you want. So if checking for element existence cannot be done for your type, add an implementation that fails or dies, to avoid silently doing the wrong thing.

method DELETE-KEY

multi method DELETE-KEY (::?CLASS:D: $key)

Expected to delete the element associated with $key, and return the value it had. This is what postcircumfix { } calls when invoked like %foo<aa>:delete.

What "deleting" an element means, is up to your type - though it should usually cause EXISTS-KEY to become False for that key.

Implementing this method is optional; if you don't, users trying to delete elements from an object of this type will get an appropriate error message.

method ASSIGN-KEY

multi method ASSIGN-KEY (::?CLASS:D: $key, $new)

Expected to set the element associated with $key to the value $new. Implementing this is entirely optional; if you don't, self.AT-KEY($key) = $new is used instead, and if you do, you should make sure it has the same effect.

This is meant as an opt-in performance optimization, so that simple assignments %age<Claire> = 29 can operate without having to call AT-KEY (which would have to create and return a potentially expensive container object).

Note that implementing ASSIGN-KEY does not relieve you from making AT-KEY an rw method though, because less trivial assignments/modifications such as %age<Claire>++ will still use AT-KEY.

method BIND-KEY

multi method BIND-KEY (::?CLASS:D: $key, \new)

Expected to bind the value or container new to the slot associated with $key, replacing any container that would be naturally found there. This is what is called when you write:

my $x = 10;
%age<Claire> := $x;

The generic Hash class supports this in order to allow building complex linked data structures, but for more domain-specific types it may not make sense, so don't feel compelled to implement it. If you don't, users will get an appropriate error message when they try to bind to an associative slot of an object of this type.

method STORE

method STORE (::?CLASS:D: \values, :$initialize)

This method should only be supplied if you want to support the:

my %h is Foo = a => 42, b => 666;

syntax for binding your implementation of the Associative role.

Should accept the values to (re-)initialize the object with, which either could consist of Pairs, or separate key/value pairs. The optional named parameter will contain a True value when the method is called on the object for the first time. Should return the invocant.