typesystem.pod6

=begin pod :kind("Language") :subkind("Language") :category("fundamental")

=TITLE Type system

=SUBTITLE Introduction to the type system of Raku

=head1 Definition of a Raku type

A type defines a new object by creating a type object that provides an
interface to create instances of objects or to check values against. Any type
object is a subclass of L<Any|/type/Any> or L<Mu|/type/Mu>. Introspection
methods are provided via inheritance from those base classes and the
introspection postfix L<.^|/language/operators#postfix_.^>. A new type is
introduced to the current scope by one of the following type declarators at
compile time or with the L<metaobject protocol|/language/mop> at runtime. All
type names must be unique in their scope.

=head2 Default types

If no type is provided by the user Raku assumes the type to be C<Any>. This
includes L<containers|/language/containers>, base-classes,
L<parameters|/type/Signature#Type_constraints> and return types.

    my $a = 1;
    $a = Nil;
    say $a.^name;
    # OUTPUT: «Any␤»

    class C {};
    say C.^parents(:all);
    # OUTPUT: «((Any) (Mu))␤»

For containers the default type is C<Any> but the default type constraint is
C<Mu>. Please note that binding replaces the container, not just the value. The
type constraint may change in this case.

=head2 Type objects

To test if an object is a type object, use
L<smartmatch|/language/operators#index-entry-smartmatch_operator>
against a type constrained with a
L<type smiley|/type/Signature#Constraining_defined_and_undefined_values> or
L«C<.DEFINITE>|/language/mop#index-entry-syntax_DEFINITE-DEFINITE» method:

=begin code :ok-test<WHAT>
my $a = Int;
say $a ~~ Mu:U;
# OUTPUT: «True␤»
say not $a.DEFINITE;
# OUTPUT: «True␤»
=end code

C<.DEFINITE> will return C<True> if the invocant is an instance. If it returns
C<False>, then the invocant is a type object.

=head3 Undefinedness

Undefined objects maintain type information in Raku. Type objects are used to
represent both undefinedness and the type of the undefined value. To provide a
general undefined value use L<Any|/type/Any>. If differentiation from C<Any>,
the default type for containers and arguments, is required use L<Mu|/type/Mu>.

Instances of objects created by L<.CREATE|/type/Mu#method_CREATE> are by
convention defined. The method L<.defined|/type/Mu#routine_defined> will return
C<Bool::True> to indicate definedness. The exceptions to that rule are
L<Nil|/type/Nil> and L<Failure|/type/Failure>. Please note that any object is
able to overload C<.defined> and as such can carry additional information.
Also, Raku makes a clear distinction between definedness and trueness. Many
values are defined even though they carry the meaning of wrongness or
emptiness. Such values are C<0>, L<Bool::False|/type/Bool>,
L<()|/language/operators#term_(_)> (empty list) and L<NaN|/type/Num#NaN>.

Values can become undefined at runtime via L<mixin|/language/operators#infix_but>.

    my Int $i = 1 but role :: { method defined { False } };
    say $i // "undefined";
    # OUTPUT: «undefined␤»

To test for definedness call C<.defined>, use
L<//|/language/operators#infix_//>,
 L<with/without|/language/control#with,_orwith,_without> and
L<signatures|/type/Signature#Constraining_defined_and_undefined_values>.

=head3 Coercion

Turning one type into another is done with coercion methods that have the same
name as the target type. This convention is made mandatory by
L<Signatures|/type/Signature#Coercion_type>. The source type has to know how to
turn itself into the target type. To allow built-in types to turn themselves
into user defined types use
L<augment|/language/variables#The_augment_declarator> or the
L<MOP|/language/mop>.

    class C {
        has $.int;
        method this-is-c { put 'oi' x $!int ~ '‽' }
    }

    use MONKEY-TYPING;
    augment class Int {
        method C { C.new(:int(self))}
    }

    my $i = 10;
    $i.=C;
    $i.this-is-c();
    # OUTPUT: «oioioioioioioioioioi‽␤»

Raku provides methods defined in L<Cool|/type/Cool> to convert to a target
type before applying further operations. Most built-in types descend from
C<Cool> and as such may provide implicit coercion that may be undesired. It is
the responsibility of the user to care about trap-free usage of those
methods.

    my $whatever = "123.6";
    say $whatever.round;
    # OUTPUT: «124␤»
    say <a b c d>.starts-with("ab");
    # OUTPUT: «False␤»

=head1 Type declarators

Type declarators introduce a new type into the given scope. Nested scopes can
be separated by C<::>. New L<packages|/language/packages> are created
automatically if no such scope exists already.

    class Foo::Bar::C {};
    put Foo::Bar::.keys;
    # OUTPUT: «C␤»

X«|... (forward declaration)»
X«Forward declarations» can be provided with a block containing only C<...>. The
compiler will check at the end of the current scope if the type is defined.

    class C {...}
    # many lines later
    class C { has $.attr }

=head2 C<class>

The C<class> declarator creates a compile time construct that is compiled into a
type object. The latter is a simple Raku object and provides methods to
construct instances by executing initializers and sub methods to fill all
attributes declared in a class, and any parent class, with values. Initializers
can be provided with the declaration of attributes or in constructors. It's the
responsibility of the L<Metamodel::ClassHOW|/type/Metamodel::ClassHOW> to know
how to run them. This is the only magic part of building objects in Raku. The
default parent type is C<Any>, which in turn inherits from C<Mu>. The latter
provides the default constructor C<.new> which is named like this by convention.
Aside from this, C<.new> does not carry any special meaning nor is treated in
any special way.

For more information how to use classes see the L<Classes and objects|/language/classtut>
tutorial.

=head3 Mixins

The type introduced by C<class> can be extended with
L«infix:<but>|/language/operators#infix_but» at runtime. The original type is
not modified, instead a new type object is returned and can be stored in
a container that type checks successful against the original type or the role
that is mixed in.

    class A {}
    role R { method m { say 'oi‽' } }
    my R $A = A but R;
    my $a1 = $A.new;
    $a1.m;
    say [$A ~~ R, $a1 ~~ R];
    # OUTPUT: «oi‽␤[True True]␤»

=head3 Introspection

=head4 Metaclass

To test if a given type object is a class, test the metaobject method C<.HOW>
against L<Metamodel::ClassHOW|/type/Metamodel::ClassHOW>.

    class C {};
    say C.HOW ~~ Metamodel::ClassHOW;
    # OUTPUT: «True␤»

=head3 Private attributes

Private L<attribute|/type/Attribute>s are addressed with any of the twigils
C<$!>, C<@!> and C<%!>. They do not have public accessor methods generated
automatically.  As such they can not be altered from outside the class they are
defined in.

    class C {
        has $!priv;
        submethod BUILD { $!priv = 42 }
    };

    say (.name, .package, .has_accessor) for C.new.^attributes;
    # OUTPUT: «($!priv (C) False)␤»

X«|method (declarator)»
=head3 Methods

The C<method> declarator defines objects of type L<Method|/type/Method> and
binds them to the provided name in the scope of a class. Methods in a class are
C<has> scoped by default. Methods that are C<our> scoped are not added to the
method cache by default and as such can not be called with the accessor sigil
C<$.>. Call them with their fully qualified name and the invocant as the
first argument.

=head4 Inheritance and multis

A normal method in a subclass does not compete with multis of a parent class.

    class A {
        multi method m(Int $i){ say 'Int' }
        multi method m(int $i){ say 'int' }
    }

    class B is A {
        method m(Int $i){ say 'B::Int' }
    }

    my int $i;
    B.new.m($i);
    # OUTPUT: «B::Int␤»

X«|only method»
=head4 Only method

To explicitly state that a method is not a multi method use the C<only> method
declarator.

=for code :skip-test<compile time error>
class C {
    only method m {};
    multi method m {};
};
# OUTPUT: «X::Comp::AdHoc: Cannot have a multi candidate for 'm' when an only method is also in the package 'C'␤»

=head4 submethod BUILD

The L<submethod|/type/Submethod> C<BUILD> is (indirectly) called by
L<.bless|/type/Mu#method_bless>. It
is meant to set private and public attributes of a class and receives all names
attributes passed into C<.bless>. The default
constructor L<.new|/type/Mu#method_new> defined in C<Mu> is the method that
invokes it. Given that public accessor methods are not available in C<BUILD>,
you must use private attribute notation instead.

    class C {
        has $.attr;
        submethod BUILD (:$attr = 42) {
            $!attr = $attr
        };
        multi method new($positional) {
            self.bless(:attr($positional), |%_)
       }
    };

    C.new.say; C.new('answer').say;
    # OUTPUT: «C.new(attr => 42)␤
    #          C.new(attr => "answer")␤»

=head4 Fallback method
X<|FALLBACK (method)>

A method with the special name C<FALLBACK> will be called when other means to
resolve the name produce no result. The first argument holds the name and all
following arguments are forwarded from the original call. Multi methods and
L«sub-signatures|/type/Signature#Sub-signatures» are supported.

    class Magic {
        method FALLBACK ($name, |c(Int, Str)) {
        put "$name called with parameters {c.perl}"  }
    };
    Magic.new.simsalabim(42, "answer");

    # OUTPUT: «simsalabim called with parameters ⌈\(42, "answer")⌋␤»

X<|WHAT (reserved method)>X<|WHO (reserved method)>
X<|HOW (reserved method)>X<|VAR (reserved method)>
=head4 Reserved method names

Some built-in introspection methods are actually special syntax provided by the
compiler, namely C<WHAT>, C<WHO>, C<HOW> and C<VAR>. Declaring methods with
those names will silently fail. A dynamic call will work, what allows to call
methods from foreign objects.

=begin code :ok-test<WHAT>
class A {
    method WHAT { "ain't gonna happen" }
};

say A.new.WHAT;    # OUTPUT: «(A)␤»
say A.new."WHAT"() # OUTPUT: «ain't gonna happen␤»
=end code

=head4 Methods in package scope

Any C<our> scoped method will be visible in the package scope of a class.

    class C {
        our method packaged {};
        method loose {}
    };
    say C::.keys
    # OUTPUT: «(&packaged)␤»

=head4 Setting attributes with namesake variables and methods

Instead of writing C<< attr => $attr >> or C<:attr($attr)>, you can save some
typing if the variable (or method call) you're setting the attribute with
shares the name with the attribute:

    class A { has $.i = 42 };
    class B {
        has $.i = "answer";
        method m() { A.new(:$.i) }
        #                  ^^^^  Instead of i => $.i or :i($.i)
    };
    my $a = B.new.m;
    say $a.i; # OUTPUT: «answer␤»

Since C<$.i> method call is named C<i> and the attribute is also named C<i>,
Raku lets us shortcut. The same applies to C<:$var>, C<:$!private-attribute>,
C<:&attr-with-code-in-it>, and so on.

=head3 trait C<is nodal>

Marks a L<List|/type/List> method to indicate to hyperoperator to not descend into inner
L<Iterables|/type/Iterable> to call this method. This trait generally isn't
something end users would be using, unless they're subclassing or augmenting
core L<List|/type/List> type.

In order to demonstrate the difference consider the following examples, the first
using a method (C<elems>) that C<is nodal> and the second using a method (C<Int>)
which is not nodal.

    say ((1.0, "2", 3e0), [^4], '5')».elems; # OUTPUT: «(3, 4, 1)␤»
    say ((1.0, "2", 3e0), [^4], '5')».Int    # OUTPUT: «((1 2 3) [0 1 2 3] 5)␤»

=head3 trait X<C<handles>|handles trait>

Defined as:

    multi sub trait_mod:<handles>(Attribute:D $target, $thunk)

The L<trait|/type/Sub#Traits> C<handles> applied to an attribute of a class will
delegate all calls to the provided method name to the method with the same name
of the attribute. The object referenced by the attribute must be initialized. A
type constraint for the object that the call is delegated to can be provided.

    class A      { method m(){ 'A::m has been called.' } }
    class B is A { method m(){ 'B::m has been called.' } }
    class C {
        has A $.delegate handles 'm';
        method new($delegate){ self.bless(delegate => $delegate) }
    };
    say C.new(B.new).m(); # OUTPUT: «B::m has been called.␤»

Instead of a method name, a C<Pair> (for renaming), a list of names or
C<Pair>s, a C<Regex> or a C<Whatever> can be provided. In the latter case
existing methods, both in the class itself and its inheritance chain, will take
precedence. If even local X«C<FALLBACK>|FALLBACK (trait handles)»s should be
searched, use a L<C<HyperWhatever>|/type/HyperWhatever>.

    class A {
        method m1(){ 'A::m1 has been called.' }
        method m2(){ 'A::m2 has been called.' }
    }

    class C {
        has $.delegate handles <m1 m2> = A.new()
    }
    say C.new.m2; # OUTPUT: «A::m2 has been called.␤»

    class D {
        has $.delegate handles /m\d/ = A.new()
    }
    say D.new.m1; # OUTPUT: «A::m1 has been called.␤»

    class E {
        has $.delegate handles (em1 => 'm1') = A.new()
    }
    say E.new.em1; # OUTPUT: «A::m1 has been called.␤»

=head3  X<trait C<is>|classes,is>

Defined as:

    multi sub trait_mod:<is>(Mu:U $child, Mu:U $parent)

The L<trait|/type/Sub#Traits> C<is> accepts a type object to be
added as a parent class of a class in its definition. To allow multiple
inheritance the trait can be applied more than once. Adding parents to a class
will import their methods into the target class. If the same method name occurs
in multiple parents, the first added parent will win.

If no C<is> trait is provided the default of L<C<Any>|/type/Any> will be used
as a parent class. This forces all Raku objects to have the same set of basic
methods to provide an interface for introspection and coercion to basic types.

    class A {
        multi method from-a(){ 'A::from-a' }
    }
    say A.new.^parents(:all).perl;
    # OUTPUT: «(Any, Mu)␤»

    class B {
        method from-b(){ 'B::from-b ' }
        multi method from-a(){ 'B::from-A' }
    }

    class C is A is B {}
    say C.new.from-a();
    # OUTPUT: «A::from-a␤»

=head3 trait X«C<is rw>|is rw (class)»

Defined as:

    sub trait_mod:<is>(Mu:U $type, :$rw!)

The L<trait|/type/Sub#Traits> C<is rw> on a class will create writable accessor
methods on all public attributes of that class.

    class C is rw {
        has $.a;
    };
    my $c = C.new.a = 42;
    say $c; # OUTPUT: «42␤»

=head3 trait C<is required>

Defined as:

    multi sub trait_mod:<is>(Attribute $attr, :$required!)
    multi sub trait_mod:<is>(Parameter:D $param, :$required!)

Marks a class or roles attribute as required. If the attribute is not
initialized at object construction time throws
L<X::Attribute::Required|/type/X::Attribute::Required>.

    class Correct {
        has $.attr is required;
        submethod BUILD (:$attr) { $!attr = $attr }
    }
    say Correct.new(attr => 42);
    # OUTPUT: «Correct.new(attr => 42)␤»

    class C {
        has $.attr is required;
    }
    C.new;
    CATCH { default { say .^name => .Str } }
    # OUTPUT: «X::Attribute::Required => The attribute '$!attr' is required, but you did not provide a value for it.␤»

You can provide a reason why it's required as an argument to C<is required>

=for code :skip-test<illustrates error>
class Correct {
    has $.attr is required("it's so cool")
};
say Correct.new();
# OUTPUT: «The attribute '$!attr' is required because it's so cool,␤but you did not provide a value for it.␤»

=head3 trait C<hides>

The trait C<hides> provides inheritance without being subject to
L<re-dispatching|/language/functions#Re-dispatching>.

    class A {
        method m { say 'i am hidden' }
    }
    class B hides A {
        method m { nextsame }
        method n { self.A::m }
    };

    B.new.m;
    B.new.n;
    # OUTPUT: «i am hidden␤»

The trait C<is hidden> allows a class to hide itself from
L<re-dispatching|/language/functions#Re-dispatching>.

    class A is hidden {
        method m { say 'i am hidden' }
    }
    class B is A {
        method m { nextsame }
        method n { self.A::m }
    }

    B.new.m;
    B.new.n;
    # OUTPUT: «i am hidden␤»

=head3 trait C<trusts>

To allow one class to access the private methods of another class use the trait
C<trusts>. A forward declaration of the trusted class may be required.

    class B {...};
    class A {
        trusts B;
        has $!foo;
        method !foo { return-rw $!foo }
        method perl { "A.new(foo => $!foo)" }
    };
    class B {
        has A $.a .= new;
        method change { $!a!A::foo = 42; self }
    };
    say B.new.change;
    # OUTPUT: «B.new(a => A.new(foo => 42))␤»

=head3 Augmenting a class

To add methods and attributes to a class at compile time use C<augment> in
front of a class definition fragment. The compiler will demand the pragmas
C<use MONKEY-TYPING> or C<use MONKEY> early in the same scope. Please note that
there may be performance implications, hence the pragmas.

    use MONKEY; augment class Str {
        method mark(Any :$set){
            state $mark //= $set; $mark
        }
    };
    my $s = "42";
    $s.mark(set => "answer");
    say $s.mark
    # OUTPUT: «answer␤»

There are few limitations of what can be done inside the class fragment. One of
them is the redeclaration of a method or sub into a multi. Using added
attributes is not yet implemented. Please note that adding a multi candidate
that differs only in its named parameters will add that candidate behind the
already defined one and as such it won't be picked by the dispatcher.

=head3 Versioning and authorship

Versioning and authorship can be applied via the adverbs
X«C«:ver<>»|:ver<> (class)» and X«C«:auth<>»|:auth<> (class)».
Both take a string as argument, for C<:ver> the string is converted to a
L<Version|/type/Version> object. To query a class version and author use
C<.^ver> and C<^.auth>.

    class C:ver<4.2.3>:auth<me@here.local> {}
    say [C.^ver, C.^auth];
    # OUTPUT: «[v4.2.3 me@here.local]␤»

=head2 C<role>

X<|declarator,role (typesystem)>
Roles are class fragments, which allow the definition of interfaces that are
shared by classes. The C<role> declarator also introduces a type object that
can be used for type checks. Roles can be mixed into classes and objects at
runtime and compile time. The C<role> declarator returns the created type
object thus allowing the definition of anonymous roles and in-place mixins.

    role Serialize {
        method to-string { self.Str }
        method to-number { self.Num }
    }

    class A does Serialize {}
    class B does Serialize {}

    my Serialize @list;
    @list.push: A.new;
    @list.push: B.new;

    say @list».to-string;
    # OUTPUT: «[A<57192848> B<57192880>]␤»

Use C<...> as the only element of a method body to declare a method to be
abstract. Any class getting such a method mixed in has to overload it. If the
method is not overloaded before the end of the compilation unit
C<X::Comp::AdHoc> will be thrown.

    EVAL 'role R { method overload-this(){...} }; class A does R {}; ';
    CATCH { default { say .^name, ' ', .Str } }
    # OUTPUT: «X::Comp::AdHoc Method 'overload-this' must be implemented by A because it is required by roles: R.␤»

=head3 Auto-punning

A role can be used instead of a class to create objects. Since roles can't
exist at runtime, a class of the same name is created that will type check
successful against the role.

    role R { method m { say 'oi‽' } };
    R.new.^mro.say;
    # OUTPUT: «((R) (Any) (Mu))␤»
    say R.new.^mro[0].HOW.^name;
    # OUTPUT: «Perl6::Metamodel::ClassHOW␤»
    say R.new ~~ R;
    # OUTPUT: «True␤»

=head3 trait C<does>

The trait C<does> can be applied to roles and classes providing compile time
mixins. To refer to a role that is not defined yet, use a forward declaration.
The type name of the class with mixed in roles does not reflect the mixin, a
type check does. If methods are provided in more than one mixed in role, the
method that is defined first takes precedence. A list of roles separated by
comma can be provided. In this case conflicts will be reported at compile time.

    role R2 {...};
    role R1 does R2 {};
    role R2 {};
    class C does R1 {};

    say [C ~~ R1, C ~~ R2];
    # OUTPUT: «[True True]␤»

For runtime mixins see L<but|/language/operators#infix_but> and L<does|/language/operators#infix_does>.

=head3 Parameterized

Roles can be provided with parameters in-between C<[]> behind a roles name.
X<|Type Capture (role)>L<Type captures|/type/Signature#Type_captures> are supported.

    role R[$d] { has $.a = $d };
    class C does R["default"] { };

    my $c = C.new;
    say $c;
    # OUTPUT: «C.new(a => "default")␤»

Parameters can have type constraints, C<where> clauses are not supported for
types but can be implemented via L<C<subset>s|#subset>.

    class A {};
    class B {};
    subset A-or-B where * ~~ A|B;
    role R[A-or-B ::T] {};
    R[A.new].new;

Default parameters can be provided.

    role R[$p = fail("Please provide a parameter to role R")] {};
    my $i = 1 does R;
    CATCH { default { say .^name, ': ', .Str} }
    # OUTPUT: «X::AdHoc: Could not instantiate role 'R':␤Please provide a parameter to role R␤»

=head3 As type constraints

Roles can be used as type constraints wherever a type is expected. If a role is
mixed in with C<does> or C<but>, its type-object is added to the type-object
list of the object in question. If a role is used instead of a class (using
auto-punning), the auto-generated class' type-object, of the same name as the
role, is added to the inheritance chain.

=begin code
role Unitish[$unit = fail('Please provide a SI unit quantifier as a parameter to the role Unitish')] {
    has $.SI-unit-symbol = $unit;
    method gist {
        given self {
            # ...
            when * < 1 { return self * 1000 ~ 'm' ~ $.SI-unit-symbol }
            when * < 1000 { return self ~ $.SI-unit-symbol }
            when * < 1_000_000 { return self / 1_000 ~ 'k' ~ $.SI-unit-symbol }
            # ...
        }
    }
}

role SI-second   does Unitish[<s>] {}
role SI-meter    does Unitish[<m>] {}
role SI-kilogram does Unitish[<g>] {}

sub postfix:<s>(Numeric $num) { ($num) does SI-second }
sub postfix:<m>(Numeric $num) { ($num) does SI-meter }
sub postfix:<g>(Numeric $num) { ($num) does SI-kilogram }
sub postfix:<kg>(Numeric $num){ ($num * 1000) does SI-kilogram }

constant g = 9.806_65;

role SI-Newton does Unitish[<N>] {}

multi sub N(SI-kilogram $kg, SI-meter $m, SI-second $s --> SI-Newton ){ ($kg * ($m / $s²)) does SI-Newton }
multi sub N(SI-kilogram $kg --> SI-Newton)                            { ($kg * g) does SI-Newton }

say [75kg, N(75kg)];
# OUTPUT: «[75kg 735.49875kN]␤»
say [(75kg).^name, N(75kg).^name];
# OUTPUT: «[Int+{SI-kilogram} Rat+{SI-Newton}]␤»
=end code

=head3 Versioning and authorship

Versioning and authorship can be applied via the adverbs
X«C«:ver<>»|:ver<> (role)» and X«C«:auth<>»|:auth<> (role)».
Both take a string as argument, for C<:ver> the string is converted to a
L<Version|/type/Version> object. To query a role's version and author use
C<.^ver> and C<^.auth>.

    role R:ver<4.2.3>:auth<me@here.local> {}
    say [R.^ver, R.^auth];
    # OUTPUT: «[v4.2.3 me@here.local]␤»

X<|Enumeration; Enums; enum>
=head2 C<enum>

Enumerations provide constant key-value-pairs with an associated type. Any key
is of that type and injected as a symbol into the current scope. If the symbol
is used, it is treated as a constant expression and the symbol is replaced with
the value of the enum-pair. Any Enumeration inherits methods from the role
L<C<Enumeration>|/type/Enumeration>. Complex expressions for generating
key-value pairs are not supported. In general, an C<enum> is a L<Map|/type/Map>
whose elements have the C<Enumeration> role mixed in; this role includes, for
each element, an index which creates an order on the map.

Stringification of the symbol, which is done automatically in string context and
is exactly equal to its name, which is also the key of the enum-pair.

    enum Names ( name1 => 1, name2 => 2 );
    say name1, ' ', name2; # OUTPUT: «name1 name2␤»
    say name1.value, ' ', name2.value; # OUTPUT: «1 2␤»

Comparing symbols will use type information and the value of the enum-pair. As
value types C<Num> and C<Str> are supported.

    enum Names ( name1 => 1, name2 => 2 );
    sub same(Names $a, Names $b){
       $a eqv $b
    }

    say same(name1, name1); # OUTPUT: «True␤»
    say same(name1, name2); # OUTPUT: «False␤»
    my $a = name1;
    say $a ~~ Names; # OUTPUT: «True␤»
    say $a.^name;    # OUTPUT: «Names␤»

All keys have to be of the same type.

    enum Mass ( mg => 1/1000, g => 1/1, kg => 1000/1 );

    say Mass.enums;
    # OUTPUT: «Map.new((g => 1, kg => 1000, mg => 0.001))␤»

And you can use any kind of symbol:

    enum Suit <♣ ♦ ♥ ♠>;

As long as you refer to that symbol using the full syntax:

=for code :preamble<< enum Suit<♣> >>
say Suit::<♣>; # OUTPUT: «♣␤»

Attempting to access unicode enum keys without said syntax will result in an
error:

=for code :skip-test<example of bad code>
say ♣ ; # OUTPUT: «(exit code 1) ===SORRY!===␤Argument to "say" seems to be    malformed…

If no value is given C<Int> will be assumed as the values type and incremented
by one per key starting at zero. As enum key types C<Int>, C<Num>, C<Rat> and
C<Str> are supported.

    enum Numbers <one two three four>;

    say Numbers.enums;
    # OUTPUT: «Map.new((four => 3, one => 0, three => 2, two => 1))␤»

A different starting value can be provided.

    enum Numbers «:one(1) two three four»;

    say Numbers.enums;
    # OUTPUT: «Map.new((four => 4, one => 1, three => 3, two => 2))␤»

You can also do this with the B<()> form of the initializer, but will need to
quote keys that do not have a value:

    enum Numbers (
      one => 1,
      'two',
      'three',
      'four'
    );

Enums can also be anonymous, with the only difference with named C<enum>s being
that you cannot use it in C<Signature>s or to declare variables.

    my $e = enum <one two three>;
    say two;       # OUTPUT: «two␤»
    say one.^name; # OUTPUT: «␤»
    say $e.^name;  # OUTPUT: «Map␤»

There are various methods to get access to the keys and values of the symbols
that have been defined. All of them turn the values into C<Str>, which may not
be desirable. By treating the enum as a package, we can get a list of types for
the keys.

    enum E <one two>;
    my @keys = E::.values;
    say @keys.map: *.perl;
    # OUTPUT: «(E::one E::two)␤»

With the use of B<()> parentheses, an enum can be defined using any
arbitrary dynamically defined list. The list should consist of Pair
objects:

For example, in file C<config> we have:

=for code :lang<text>
a 1
b 2

We can create an enum using it with this code:

=for code :skip-test<needs filesystem>
enum ConfigValues ('config'.IO.lines.map({ my ($key, $value) = $_.words; $key => $value }));
say ConfigValues.enums;          # OUTPUT: «Map.new((a => 1, b => 2))␤»

Firstly, we read lines from C<config> file, split every line using
C<words> method and return resulting pair for every line, thus
creating a List of Pairs.

=head3 Metaclass

To test if a given type object is an C<enum>, test the metaobject method
C<.HOW> against L<Metamodel::EnumHOW|/type/Metamodel::EnumHOW> or simply test
against the C<Enumeration> role.

    enum E(<a b c>);
    say E.HOW ~~ Metamodel::EnumHOW; # OUTPUT: «True␤»
    say E ~~ Enumeration;            # OUTPUT: «True␤»

=head3 Methods

=head4 method enums

Defined as:

    method enums()

Returns the list of enum-pairs.

    enum Mass ( mg => 1/1000, g => 1/1, kg => 1000/1 );
    say Mass.enums; # OUTPUT: «{g => 1, kg => 1000, mg => 0.001}␤»


=head3 Coercion

If you want to coerce the value of an enum element to its proper
enum object, use the coercer with the name of the enum:

    my enum A (sun => 42, mon => 72);
    A(72).pair.say;   # OUTPUT: «mon => 72␤»
    A(1000).say; # OUTPUT: «(A)␤»

The last example shows what happens if there is no enum-pair that includes that
as a value.

=head2 C<module>

Modules are usually one or more source files that expose Raku constructs,
such as classes, roles, grammars, subroutines and variables. Modules are
usually used for distributing Raku code as libraries which can be used in
another Raku program.

For a full explanation see L<Modules|/language/modules>.

=head3 Versioning and authorship

Versioning and authorship can be applied via the adverbs C«:ver<>» and C«:auth<>».
Both take a string as argument, for C<:ver> the string is converted to a
L<Version|/type/Version> object. To query a modules version and author use
C<.^ver> and C<^.auth>.

    module M:ver<4.2.3>:auth<me@here.local> {}
    say [M.^ver, M.^auth];
    # OUTPUT: «[v4.2.3 me@here.local]␤»

=head2 C<package>

Packages are nested namespaces of named program elements. Modules, classes and
grammars are all types of package.

For a full explanation see L<Packages|/language/packages>.

=head2 C<grammar>

Grammars are a specific type of class intended for parsing text. Grammars are
composed of rules, tokens and regexes which are actually methods, since grammars
are classes.

For a full explanation see L<Grammars|/language/grammars>.

=head3 Versioning and authorship

Versioning and authorship can be applied via the adverbs
X«C«:ver<>»|:ver<> (grammar)» and X«C«:auth<>»|:auth<> (grammar)».
Both take a string as argument, for C<:ver> the string is converted to a
L<Version|/type/Version> object. To query a grammars version and author use
C<.^ver> and C<^.auth>.

    grammar G:ver<4.2.3>:auth<me@here.local> {}
    say [G.^ver, G.^auth];
    # OUTPUT: «[v4.2.3 me@here.local]␤»

=head2 C<subset>

A X<C<subset>|subset> declares a new type that will re-dispatch to its base
type. If a L<C<where>|/type/Signature#where> clause is supplied any assignment
will be checked against the given code object.

    subset Positive of Int where * > -1;
    my Positive $i = 1;
    $i = -42;
    CATCH { default { put .^name,': ', .Str } }
    # OUTPUT: «X::TypeCheck::Assignment: Type check failed in assignment to $i; expected Positive but got Int (-42)␤»

Subsets can be used in signatures, e.g. by typing the output:

    subset Foo of List where (Int,Str);
    sub a($a, $b, --> Foo) { $a, $b }
    # Only a List with the first element being an Int and the second a Str will pass the type check.
    a(1, "foo");  # passes
    a("foo", 1);  # fails

Subsets can be anonymous, allowing inline placements where a subset is required
but a name is neither needed nor desirable.

    my enum E1 <A B>;
    my enum E2 <C D>;
    sub g(@a where { .all ~~ subset :: where E1|E2 } ) {
        say @a
    }
    g([A, C]);
    # OUTPUT: «[A C]␤»

Subsets can be used to check types dynamically, which can be useful in conjunction
with L<require|/language/modules#require>.
X<|dynamic subset>

=for code
require ::('YourModule');
subset C where ::('YourModule::C');

=end pod

# vim: expandtab softtabstop=4 shiftwidth=4 ft=perl6