Signature literals

A guide to signatures in Raku

Signatures appear inside parentheses after subroutine and method names, on blocks after a -> or <-> arrow, as the input to variable declarators like my, or as a separate term starting with a colon.

sub f($x) { }
    #    ^^^^ Signature of sub f
    my method x() { }
    #          ^^ Signature of a method
    my $s = sub (*@a) { }
    #           ^^^^^ Signature of an anonymous function
for <a b c> -> $x { }
    #              ^^   Signature of a Block
my ($a, @b) = 5, (6, 7, 8);
    #  ^^^^^^^^ Signature of a variable declarator
my $sig = :($a, $b);
    #          ^^^^^^^^ Standalone Signature object

Signature literals can be used to define the signature of a callback or a closure.

sub f(&c:(Int)) { }
    sub will-work(Int) { }
    sub won't-work(Str) { }
    f(&will-work);
f(&won't-work);
    CATCH { default { put .^name, ': ', .Str } };
    # OUTPUT: «X::TypeCheck::Binding::Parameter: Constraint type check failed in binding to parameter '&c'␤»
f(-> Int { 'this works too' } );

You can use any kind of literal, including numeric ones, as part of a signature; this is generally used in conjunction with multis

proto stuff(|) {*}
multi stuff(33) { 58 }
multi stuff(⅓) { 43 }
multi stuff(Int)  { 3 }
multi stuff(Complex)  { 66 }
say stuff($_) for (33, ⅓, i, 48); # OUTPUT: «58␤43␤66␤3␤»

However, you can't use True or False as literals in signatures since they will always succeed (or fail). A warning will be issued if you do so:

sub foo(True) {};
my $sig =  :( True );

They will both warn "Literal values in signatures are smartmatched against and smartmatch with True will always succeed. Use the where clause instead.". Use of False will produce a similar warning.

Smartmatching signatures against a List is supported.

my $sig = :(Int $i, Str $s);
    say (10, 'answer') ~~ $sig;
    # OUTPUT: «True␤»
    my $sub = sub ( Str $s, Int $i ) { return $s xx $i };
    say $sub.signature ~~ :( Str, Int );
    # OUTPUT: «True␤»
    given $sig {
        when :(Str, Int) { say 'mismatch' }
        when :($, $)     { say 'match' }
        default          { say 'no match' }
    }
    # OUTPUT: «match␤»

It matches the second when clause since :($, $) represents a Signature with two scalar, anonymous, arguments, which is a more general version of $sig.

When smartmatching against a Hash, the signature is assumed to consist of the keys of the Hash.

my %h = left => 1, right => 2;
    say %h ~~ :(:$left, :$right);
    # OUTPUT: «True␤»

Signature literals can contain string/numeric literals

my $sig = :('Þor', Str, Int);
    say <Þor Hammer 1> ~~ $sig; # OUTPUT: «True␤»

And they can also contain the invocant marker

class Foo {
        method bar( $self: ){ "baz" }
    };
    say Foo.^methods.first(*.name eq 'bar').signature ~~ :($: *%) ;
    # OUTPUT: «True␤»

Parameter separators

A signature consists of zero or more Parameters, separated by commas.

my $sig = :($a, @b, %c);
    sub add($a, $b) { $a + $b };

As an exception the first parameter may be followed by a colon instead of a comma to mark the invocant of a method. This is done in order to distinguish it from what would then be a regular positional parameter. The invocant is the object that was used to call the method, which is usually bound to self. By specifying it in the signature, you can change the variable name it is bound to.

method ($a: @b, %c) {};       # first argument is the invocant
class Foo {
        method whoami($me:) {
            "Well I'm class $me.^name(), of course!"
        }
    }
    say Foo.whoami; # OUTPUT: «Well I'm class Foo, of course!␤»

Type constraints

Parameters can optionally have a type constraint (the default is Any). These can be used to restrict the allowed input to a function.

my $sig = :(Int $a, Str $b);

Type constraints can have any compile-time defined value

subset Positive-integer of Int where * > 0;
sub divisors(Positive-integer $n) { $_ if $n %% $_ for 1..$n };
CATCH { default { put .^name, ': ', .Str; .resume } };

divisors 2.5;
# OUTPUT: «X::TypeCheck::Binding::Parameter: Type check failed in binding to parameter '$n'; expected Positive-integer but got Rat (2.5)␤»

divisors -3;
# OUTPUT: «X::TypeCheck::Binding::Parameter: Constraint type check failed in binding to parameter '$n'; expected Positive-integer but got Int (-3)␤»

Please note that in the code above type constraints are enforced at two different levels: the first level checks if it belongs to the type in which the subset is based, in this case Int. If it fails, a Type check error is produced. Once that filter is cleared, the constraint that defined the subset is checked, producing a Constraint type check error if it fails.

Type constraints can define multiple allowable types

sub abbrev($arg where Str|List|Hash) {...} # throws if $arg is not one of those types

Anonymous arguments are fine too, if you don't actually need to refer to a parameter by name, for instance to distinguish between different signatures in a multi or to check the signature of a Callable.

my $sig = :($, @, %a);          # two anonymous and a "normal" parameter
$sig = :(Int, Positional);      # just a type is also fine (two parameters)
sub baz(Str) { "Got passed a Str" }

Type constraints may also be type captures.

In addition to those nominal types, additional constraints can be placed on parameters in the form of code blocks which must return a true value to pass the type check

sub f(Real $x where { $x > 0 }, Real $y where { $y >= $x }) { }

The code in where clauses has some limitations: anything that produces side-effects (e.g., printing output, pulling from an iterator, or increasing a state variable) is not supported and may produce surprising results if used. Also, the code of the where clause may run more than once for a single typecheck in some implementations.

The where clause doesn't need to be a code block, anything on the right of the where-clause will be used to smartmatch the argument against it. So you can also write:

multi factorial(Int $ where 0) { 1 }
    multi factorial(Int $x)        { $x * factorial($x - 1) }

The first of those can be shortened to

multi factorial(0) { 1 }

i.e., you can use a literal directly as a type and value constraint on an anonymous parameter.

Tip: pay attention to not accidentally leave off a block when you, say, have several conditions:

-> $y where   .so && .name    {}( sub one   {} ); # WRONG!!
  -> $y where { .so && .name }  {}( sub two   {} ); # OK!
  -> $y where   .so &  .name.so {}( sub three {} ); # Also good

The first version is wrong and will issue a warning about a sub object coerced to string. The reason is the expression is equivalent to ($y ~~ ($y.so && $y.name)); that is "call .so, and if that is True, call .name; if that is also True use its value for smartmatching…". It's the result of (.so && .name) it will be smartmatched against, but we want to check that both .so and .name are truthy values. That is why an explicit Block or a Junction is the right version.

All previous arguments that are not part of a sub-signature in a Signature are accessible in a where-clause that follows an argument. Therefore, the where-clause of the last argument has access to all arguments of a signature that are not part of a sub-signature. For a sub-signature place the where-clause inside the sub-signature.

sub foo($a, $b where * == $a ** 2) { say "$b is a square of $a" }
    foo 2, 4; # OUTPUT: «4 is a square of 2␤»»
    # foo 2, 3;
    # OUTPUT: «Constraint type check failed in binding to parameter '$b'…»

Constraining optional arguments

Optional arguments can have constraints, too. Any where clause on any parameter will be executed, even if it's optional and not provided by the caller. In that case you may have to guard against undefined values within the where clause.

sub f(Int $a, UInt $i? where { !$i.defined or $i > 5 }) { ... }

Constraining slurpy arguments

Slurpy arguments can not have type constraints. A where-clause in conjunction with a Junction can be used to that effect.

sub f(*@a where {$_.all ~~ Int}) { say @a };
f(42);
f(<a>);
CATCH { default { say .^name, ' ==> ', .Str }  }
# OUTPUT: «[42]␤Constraint type check failed in binding to parameter '@a' ...»

Constraining named arguments

Constraints against named arguments apply to the value part of the colon-pair.

sub f(Int :$i){};
    f :i<forty-two>;
    CATCH { default { say .^name, ' ==> ', .Str }  }
    # OUTPUT: «X::TypeCheck::Binding::Parameter ==> Type check failed in
    # binding to parameter '$i'; expected Int but got Str ("forty-two")␤»

Constraining argument definiteness

Normally, a type constraint only checks whether the value of the parameter is of the correct type. Crucially, both object instances and type objects will satisfy such a constraint as illustrated below:

say  42.^name;    # OUTPUT: «Int␤»
    say  42 ~~ Int;   # OUTPUT: «True␤»
    say Int ~~ Int;   # OUTPUT: «True␤»

Note how both 42 and Int satisfy the match.

Sometimes we need to distinguish between these object instances (42) and type objects (Int). Consider the following code:

sub limit-lines(Str $s, Int $limit) {
        my @lines = $s.lines;
        @lines[0 .. min @lines.elems, $limit].join("\n")
    }
    say (limit-lines "a \n b \n c \n d \n", 3).raku; # "a \n b \n c \n d "
    say limit-lines Str, 3;
    CATCH { default { put .^name, ': ', .Str } };
    # OUTPUT: «X::Multi::NoMatch: Cannot resolve caller lines(Str: );
    # none of these signatures match:
    #     (Str:D $: :$count!, *%_)
    #     (Str:D $: $limit, *%_)
    #     (Str:D $: *%_)»
    say limit-lines "a \n b", Int; # Always returns the max number of lines

Here we really only want to deal with string instances, not type objects. To do this, we can use the :D type constraint. This constraint checks that the value passed is an object instance, in a similar fashion to calling its DEFINITE (meta)method.

To warm up, let's apply :D to the right-hand side of our humble Int example:

say  42 ~~ Int:D;  # OUTPUT: «True␤»
    say Int ~~ Int:D;  # OUTPUT: «False␤»

Note how only 42 matches Int:D in the above.

Returning to limit-lines, we can now amend its signature to catch the error early:

sub limit-lines(Str:D $s, Int $limit) { };
    say limit-lines Str, 3;
    CATCH { default { put .^name ~ '--' ~ .Str } };
    # OUTPUT: «Parameter '$s' of routine 'limit-lines' must be an object instance of type 'Str',
    #          not a type object of type 'Str'.  Did you forget a '.new'?»

This is much better than the way the program failed before, since here the reason for failure is clearer.

It's also possible that type objects are the only ones that make sense for a routine to accept. This can be done with the :U type constraint, which checks whether the value passed is a type object rather than an object instance. Here's our Int example again, this time with :U applied:

say  42 ~~ Int:U;  # OUTPUT: «False␤»
    say Int ~~ Int:U;  # OUTPUT: «True␤»

Now 42 fails to match Int:U while Int succeeds.

Here's a more practical example:

sub can-turn-into(Str $string, Any:U $type) {
       return so $string.$type;
    }
    say can-turn-into("3", Int);        # OUTPUT: «True␤»
    say can-turn-into("6.5", Int);      # OUTPUT: «True␤»
    say can-turn-into("6.5", Num);      # OUTPUT: «True␤»
    say can-turn-into("a string", Num); # OUTPUT: «False␤»

Calling can-turn-into with an object instance as its second parameter will yield a constraint violation as intended:

say can-turn-into("a string", 123);
# OUTPUT: «Parameter '$type' of routine 'can-turn-into' must be a type object
# of type 'Any', not an object instance of type 'Int'...»

For explicitly indicating the normal behavior, that is, not constraining whether the argument will be an instance or a type object, :_ can be used but this is unnecessary since this is the default constraint (of this kind) on arguments. Thus, :(Num:_ $) is the same as :(Num $).

To recap, here is a quick illustration of these type constraints, also known collectively as type smileys:

# Checking a type object
    say Int ~~ Any:D;    # OUTPUT: «False␤»
    say Int ~~ Any:U;    # OUTPUT: «True␤»
    say Int ~~ Any:_;    # OUTPUT: «True␤»
# Checking a subset
    subset Even of Int where * // 2;
    say 3 ~~ Even:D;     # OUTPUT: «True␤»
    say 3 ~~ Even:U;     # OUTPUT: «False␤»
    say Int ~~ Even:U;   # OUTPUT: «True␤»
# Checking an object instance
    say 42 ~~ Any:D;     # OUTPUT: «True␤»
    say 42 ~~ Any:U;     # OUTPUT: «False␤»
    say 42 ~~ Any:_;     # OUTPUT: «True␤»
# Checking a user-supplied class
    class Foo {};
    say Foo ~~ Any:D;    # OUTPUT: «False␤»
    say Foo ~~ Any:U;    # OUTPUT: «True␤»
    say Foo ~~ Any:_;    # OUTPUT: «True␤»
# Checking an instance of a class
    my $f = Foo.new;
    say $f  ~~ Any:D;    # OUTPUT: «True␤»
    say $f  ~~ Any:U;    # OUTPUT: «False␤»
    say $f  ~~ Any:_;    # OUTPUT: «True␤»

The Classes and Objects document further elaborates on the concepts of instances and type objects and discovering them with the .DEFINITE method.

Keep in mind all parameters have values; even optional ones have default values that are the type object of the constrained type for explicit type constraints. If no explicit type constraint exists, the default value is an Any type object for methods, submethods, and subroutines, and a Mu type object for blocks. This means that if you use the :D type smiley, you'd need to provide a default value or make the parameter required. Otherwise, the default value would be a type object, which would fail the definiteness constraint.

sub divide (Int:D :$a = 2, Int:D :$b!) { say $a/$b }
    divide :1a, :2b; # OUTPUT: «0.5␤»

The default value will kick in when that particular parameter, either positional or named, gets no value at all.

sub f($a = 42){
      my $b is default('answer');
      say $a;
      $b = $a;
      say $b
    };
    f;     # OUTPUT: «42␤42␤»
    f Nil; # OUTPUT: «Nil␤answer␤»

$a has 42 as its default value. With no value, $a will be assigned the default value declared in the Signature. However, in the second case, it does receive a value, which happens to be Nil. Assigning Nil to any variable resets it to its default value, which has been declared as 'answer' by use of the default trait. That explains what happens the second time we call f. Routine parameters and variables deal differently with default value, which is in part clarified by the different way default values are declared in each case (using = for parameters, using the default trait for variables).

Note: in 6.c language, the default value of :U/:D constrained variables was a type object with such a constraint, which is not initializable, thus you cannot use the .= operator, for example.

use v6.c;
my Int:D $x .= new: 42;
# OUTPUT: You cannot create an instance of this type (Int:D)
# in block <unit> at -e line 1

In the 6.d language, the default default is the type object without the smiley constraint:

use v6.d;
my Int:D $x .= new: 42; # OUTPUT: «42␤»

A closing remark on terminology: this section is about the use of the type smileys :D and :U to constrain the definiteness of arguments. Occasionally definedness is used as a synonym for definiteness; this may be confusing, since the terms have subtly different meanings.

As explained above, definiteness is concerned with the distinction between type objects and object instances. A type object is always indefinite, while an object instance is always definite. Whether an object is a type object/indefinite or an object instance/definite can be verified using the DEFINITE (meta)method.

Definiteness should be distinguished from definedness, which is concerned with the difference between defined and undefined objects. Whether an object is defined or undefined can be verified using the defined-method, which is implemented in class Mu. By default a type object is considered undefined, while an object instance is considered defined; that is: .defined returns False on a type object, and True otherwise. But this default behavior may be overridden by subclasses. An example of a subclass that overrides the default .defined behavior is Failure, so that even an instantiated Failure acts as an undefined value:

my $a = Failure;                # Initialize with type object
    my $b = Failure.new("foo");     # Initialize with object instance
    say $a.DEFINITE;                # OUTPUT: «False␤» : indefinite type object
    say $b.DEFINITE;                # OUTPUT: «True␤»  : definite object instance
    say $a.defined;                 # OUTPUT: «False␤» : default response
    say $b.defined;                 # OUTPUT: «False␤» : .defined override

Constraining signatures of Callables

:u

whitespace allowed):

sub apply(&l:(Int:D --> Int:D), Int:D \n) {
        l(n)
    }
sub identity(Int:D \i --> Int:D) { i }
    sub double(Int:D \x --> Int:D) { 2 * x }
say apply &identity, 10; # OUTPUT: «10␤»
    say apply &double, 10;   # OUTPUT: «20␤»

Typed lambdas also work with constrained callable parameters.

say apply -> Int:D \x --> Int:D { 2 * x }, 3;  # OUTPUT: «6␤»
say apply -> Int:D \x --> Int:D { x ** 3 }, 3; # OUTPUT: «27␤»

Note that this shorthand syntax is available only for parameters with the & sigil. For others, you need to use the long version:

sub play-with-tens($c where .signature ~~ :(Int, Str)) { say $c(10, 'ten') }
    sub by-joining-them(Int $i, Str $s) { $s ~ $i }
    play-with-tens &by-joining-them;                              # OUTPUT: «ten10␤»
    play-with-tens -> Int \i, Str \s { s x (1..10).roll mod i };  # OUTPUT: «tenten␤»
sub g(Num $i, Str $s) { $s ~ $i }
    # play-with-tens(&g); # Constraint type check failed

Constraining return types

There are multiple ways to constrain return types on a Routine. All versions below are currently valid and will force a type check on successful execution of a routine.

Nil and Failure are always allowed as return types, regardless of any type constraint. This allows Failure to be returned and passed on down the call chain.

sub foo(--> Int) { Nil };
    say foo.raku; # OUTPUT: «Nil␤»

Type captures are not supported.

|Syntax,--> |Syntax,Return type arrow

Return type arrow: -->

This form of indicating return types (or constants) in the signature is preferred, since it can handle constant values while the others can't. For consistency, it is the only form accepted on this site.

The return type arrow has to be placed at the end of the parameter list, with or without a , before it.

sub greeting1(Str $name  --> Str) { say "Hello, $name" } # Valid
sub greeting2(Str $name, --> Str) { say "Hello, $name" } # Valid

sub favorite-number1(--> 42) {        } # OUTPUT: 42
sub favorite-number2(--> 42) { return } # OUTPUT: 42

If the type constraint is a constant expression, it is used as the return value of the routine. Any return statement in that routine has to be argumentless.

sub foo(Str $word --> 123) { say $word; return; }
my $value = foo("hello"); # OUTPUT: hello
say $value;               # OUTPUT: 123
# The code below will not compile
sub foo(Str $word --> 123) { say $word; return $word; }
my $value = foo("hello");
say $value;

returns

The keyword returns following a signature declaration has the same function as --> with the caveat that this form does not work with constant values. You cannot use it in a block either. That is why the pointy arrow form is always preferred.

sub greeting(Str $name) returns Str { say "Hello, $name" } # Valid
sub favorite-number returns 42 {        } # This will fail.

of

of is just the real name of the returns keyword.

sub foo() of Int { 42 }; # Valid
sub foo() of 42 {  };    # This will fail.

prefix(C-like) form

This is similar to placing type constraints on variables like my Type $var = 20;, except the $var is a definition for a routine.

my Int sub bar { 1 };     # Valid
my 42 sub bad-answer {};  # This will fail.

Coercion type

To accept one type but coerce it automatically to another, use the accepted type as an argument to the target type. If the accepted type is Any it can be omitted.

sub f(Int(Str) $want-int, Str() $want-str) {
        say $want-int.^name ~ ' ' ~ $want-str.^name
    }
    f '10', 10;
    # OUTPUT: «Int Str␤»
sub foo(Date(Str) $d) { say $d.^name; say $d };
    foo "2016-12-01";
    # OUTPUT: «Date␤2016-12-01␤»

The coercion is performed by calling the method with the name of the type to coerce to, if it exists. In this example, we're calling the builtin method Date on the Str class. The method is assumed to return the correct type—no additional checks on the result are currently performed.

Coercion can also be performed on return types:

sub square-str (Int $x --> Str(Int)) {
    $x²
}

for 2,4, *²  … 256 -> $a {
    say $a, "² is ", square-str( $a ).chars, " figures long";
}

# OUTPUT: «2² is 1 figures long␤
#          4² is 2 figures long␤
#          16² is 3 figures long␤
#          256² is 5 figures long␤»

In this example, coercing the return type to Str allows us to directly apply string methods, such as the number of characters.

Note: The appropriate method must be available on the argument, so be careful when trying to coerce custom types.

class Foo { }

sub bar(Foo(Int) $x) { say $x }

bar(3);
# OUTPUT: «Impossible coercion from 'Int' into 'Foo': no acceptable coercion method found␤»

Slurpy parameters

A function is variadic if it can take a varying number of arguments; that is, its arity is not fixed. Therefore, optional, named, and slurpy parameters make routines that use them variadic, and by extension are called variadic arguments. Here we will focus on slurpy parameters, or simply slurpies.

An array or hash parameter can be marked as slurpy by leading single (*) or double asterisk (**) or a leading plus (+). A slurpy parameter can bind to an arbitrary number of arguments (zero or more), and it will result in a type that is compatible with the sigil.

These are called "slurpy" because they slurp up any remaining arguments to a function, like someone slurping up noodles.

my $sig1 = :($a, @b);  # exactly two arguments, second must be Positional
my $sig2 = :($a, *@b); # at least one argument, @b slurps up any beyond that
my $sig3 = :(*%h);     # no positional arguments, but any number
                       # of named arguments

sub one-arg (@)  { }
sub slurpy  (*@) { }
one-arg (5, 6, 7); # ok, same as one-arg((5, 6, 7))
slurpy  (5, 6, 7); # ok
slurpy   5, 6, 7 ; # ok
# one-arg(5, 6, 7) ; # X::TypeCheck::Argument
# one-arg  5, 6, 7 ; # X::TypeCheck::Argument

sub named-names (*%named-args) { %named-args.keys };
say named-names :foo(42) :bar<baz>; # OUTPUT: «foo bar␤»

Positional and named slurpies can be combined; named arguments (i.e., Pairs) are collected in the specified hash, positional arguments in the array:

sub combined-slurpy (*@a, *%h) { { array => @a, hash => %h } }
# or: sub combined-slurpy (*%h, *@a) { ... }

say combined-slurpy(one => 1, two => 2);
# OUTPUT: «{array => [], hash => {one => 1, two => 2}}␤»
say combined-slurpy(one => 1, two => 2, 3, 4);
# OUTPUT: «{array => [3 4], hash => {one => 1, two => 2}}␤»
say combined-slurpy(one => 1, two => 2, 3, 4, five => 5);
# OUTPUT: «{array => [3 4], hash => {five => 5, one => 1, two => 2}}␤»
say combined-slurpy(one => 1, two => 2, 3, 4, five => 5, 6);
# OUTPUT: «{array => [3 4 6], hash => {five => 5, one => 1, two => 2}}␤»

Note that positional parameters aren't allowed after slurpy (or, in fact, after any type of variadic) parameters:

:(*@args, $last);
# ===SORRY!=== Error while compiling:
# Cannot put required parameter $last after variadic parameters

Normally a slurpy parameter will create an Array (or compatible type), create a new Scalar container for each argument, and assign the value from each argument to those Scalars. If the original argument also had an intermediary Scalar it is bypassed during this process, and is not available inside the called function.

Sigiled parameters will always impose a context on the collected arguments. Sigilless parameters can also be used slurpily, preceded by a + sign, to work with whatever initial type they started with:

sub zipi( +zape ) {
    zape.^name => zape
};
say zipi( "Hey "); # OUTPUT: «List => (Hey )␤»
say zipi( 1...* ); # OUTPUT: «Seq => (...)␤»

Slurpy parameters have special behaviors when combined with some traits and modifiers, as described in the section on slurpy array parameters.

Methods automatically get a *%_ slurpy named parameter added if they don't have another slurpy named parameter declared.

Types of slurpy array parameters

There are three variations to slurpy array parameters.

  • The single asterisk form flattens passed arguments.

  • The double asterisk form does not flatten arguments.

  • The plus form flattens according to the single argument rule.

Each will be described in detail in the next few sections. As the difference between each is a bit nuanced, examples are provided for each to demonstrate how each slurpy convention varies from the others.

Flattened slurpy

Slurpy parameters declared with one asterisk will flatten arguments by dissolving one or more layers of bare Iterables.

my @array = <a b c>;
my $list := <d e f>;
sub a(*@a)  { @a.raku.say };
a(@array);                 # OUTPUT: «["a", "b", "c"]␤»
a(1, $list, [2, 3]);       # OUTPUT: «[1, "d", "e", "f", 2, 3]␤»
a([1, 2]);                 # OUTPUT: «[1, 2]␤»
a(1, [1, 2], ([3, 4], 5)); # OUTPUT: «[1, 1, 2, 3, 4, 5]␤»
a(($_ for 1, 2, 3));       # OUTPUT: «[1, 2, 3]␤»

A single asterisk slurpy flattens all given iterables, effectively hoisting any object created with commas up to the top level.

Unflattened slurpy

Slurpy parameters declared with two stars do not flatten any Iterable arguments within the list, but keep the arguments more or less as-is:

my @array = <a b c>;
my $list := <d e f>;
sub b(**@b) { @b.raku.say };
b(@array);                 # OUTPUT: «[["a", "b", "c"],]␤»
b(1, $list, [2, 3]);       # OUTPUT: «[1, ("d", "e", "f"), [2, 3]]␤»
b([1, 2]);                 # OUTPUT: «[[1, 2],]␤»
b(1, [1, 2], ([3, 4], 5)); # OUTPUT: «[1, [1, 2], ([3, 4], 5)]␤»
b(($_ for 1, 2, 3));       # OUTPUT: «[(1, 2, 3),]␤»

The double asterisk slurpy hides the nested comma objects and leaves them as-is in the slurpy array.

Single argument rule slurpy

A slurpy parameter created using a plus engages the "single argument rule", which decides how to handle the slurpy argument based upon context. Basically, if only a single argument is passed and that argument is Iterable, that argument is used to fill the slurpy parameter array. In any other case, +@ works like **@.

my @array = <a b c>;
my $list := <d e f>;
sub c(+@b) { @b.raku.say };
c(@array);                 # OUTPUT: «["a", "b", "c"]␤»
c(1, $list, [2, 3]);       # OUTPUT: «[1, ("d", "e", "f"), [2, 3]]␤»
c([1, 2]);                 # OUTPUT: «[1, 2]␤»
c(1, [1, 2], ([3, 4], 5)); # OUTPUT: «[1, [1, 2], ([3, 4], 5)]␤»
c(($_ for 1, 2, 3));       # OUTPUT: «[1, 2, 3]␤»

For additional discussion and examples, see Slurpy Conventions for Functions.

Type captures

Type captures allow deferring the specification of a type constraint to the time the function is called. They allow referring to a type both in the signature and the function body.

sub f(::T $p1, T $p2, ::C){
        # $p1 and $p2 are of the same type T, that we don't know yet
        # C will hold a type we derive from a type object or value
        my C $division = $p1 / $p2;
        return sub (T $p1) {
            $division * $p1;
        }
    }
# The first parameter is Int and so must be the 2nd.
    # We derive the 3rd type from calling the operator that is used in &f.
    my &s = f(10, 2, Int.new / Int.new);
    say s(2); # 10 / 2 * 2 == 10

Positional vs. named arguments

An argument can be positional or named. By default, arguments are positional, except slurpy hash and arguments marked with a leading colon :. The latter is called a colon-pair. Check the following signatures and what they denote:

$sig1 = :($a);               # a positional argument
$sig2 = :(:$a);              # a named argument of name 'a'
$sig3 = :(*@a);              # a slurpy positional argument
$sig4 = :(*%h);              # a slurpy named argument

On the caller side, positional arguments are passed in the same order as the arguments are declared.

sub pos($x, $y) { "x=$x y=$y" }
    pos(4, 5);                          # OUTPUT: «x=4 y=5»

In the case of named arguments and parameters, only the name is used for mapping arguments to parameters. If a fat arrow is used to construct a Pair only those with valid identifiers as keys are recognized as named arguments.

sub named(:$x, :$y) { "x=$x y=$y" }
named( y => 5, x => 4);             # OUTPUT: «x=4 y=5»

You can invoke the routine using a variable with the same name as the named argument; in that case : will be used for the invocation so that the name of the variable is understood as the key of the argument.

sub named-shortcut( :$shortcut ) {
        say "Looks like $shortcut"
    }
    named-shortcut( shortcut => "to here"); # OUTPUT: «Looks like to here␤»
    my $shortcut = "Þor is mighty";
    named-shortcut( :$shortcut );           # OUTPUT: «Looks like Þor is mighty␤»

It is possible to have a different name for a named argument than the variable name:

sub named(:official($private)) { "Official business!" if $private }
    named :official;

Argument aliases

The colon-pair syntax can be used to provide aliases for arguments:

sub alias-named(:color(:$colour), :type(:class($kind))) {
        say $colour ~ " " ~ $kind
    }
    alias-named(color => "red", type => "A");    # both names can be used
    alias-named(colour => "green", type => "B"); # more than two names are ok
    alias-named(color => "white", class => "C"); # every alias is independent

The presence of the colon : will decide whether we are creating a new named argument or not. :$colour will not only be the name of the aliased variable, but also a new named argument (used in the second invocation). However, $kind will just be the name of the aliased variable, that does not create a new named argument. More uses of aliases can be found in sub MAIN.

A function with named arguments can be called dynamically, dereferencing a Pair with | to turn it into a named argument.

multi f(:$named) { note &?ROUTINE.signature };
    multi f(:$also-named) { note &?ROUTINE.signature };
    for 'named', 'also-named' -> $n {
        f(|($n => rand))                # OUTPUT: «(:$named)␤(:$also-named)␤»
    }
my $pair = :named(1);
    f |$pair;                           # OUTPUT: «(:$named)␤»

The same can be used to convert a Hash into named arguments.

sub f(:$also-named) { note &?ROUTINE.signature };
    my %pairs = also-named => 4;
    f |%pairs;                              # OUTPUT: «(:$also-named)␤»

A Hash that contains a list may prove problematic when slipped into named arguments. To avoid the extra layer of containers coerce to Map before slipping.

class C { has $.x; has $.y; has @.z };
    my %h = <x y z> Z=> (5, 20, [1,2]);
    say C.new(|%h.Map);
    # OUTPUT: «C.new(x => 5, y => 20, z => [1, 2])␤»

You can create as many aliases to a named argument as you want:

sub alias-named(:color(:$colour),
                :variety(:style(:sort(:type(:class($kind)))))) {
    return $colour ~ " " ~ $kind
}
say alias-named(color => "red", style => "A");
say alias-named(colour => "green", variety => "B");
say alias-named(color => "white", class => "C");

You can create named arguments that do not create any variables by making the argument an alias for an anonymous argument. This can be useful when using named arguments solely as a means of selecting a multi candidate, which is often the case with traits, for instance:

# Timestamps calls to a routine.
multi trait_mod:<is>(Routine:D $r is raw, :timestamped($)!) {
    $r does my role timestamped { has Instant:D @.timestamps };
    $r.wrap: -> | { ENTER $r.timestamps.push: now; callsame };
}

sub foo is timestamped { }
foo;
say +&foo.?timestamps; # OUTPUT: «1␤»

Optional and mandatory arguments

Positional parameters are mandatory by default, and can be made optional with a default value or a trailing question mark:

$sig1 = :(Str $id);         # required parameter
$sig2 = :($base = 10);      # optional parameter, default value 10
$sig3 = :(Int $x?);         # optional parameter, default is the Int type object

Named parameters are optional by default, and can be made mandatory with a trailing exclamation mark:

$sig1 = :(:%config);        # optional parameter
$sig2 = :(:$debug = False); # optional parameter, defaults to False
$sig3 = :(:$name!);         # mandatory 'name' named parameter

Default values can depend on previous parameters, and are (at least notionally) computed anew for each call

$sig1 = :($goal, $accuracy = $goal / 100);
$sig2 = :(:$excludes = ['.', '..']);        # a new Array for every call

Was an argument passed for a parameter?

Table showing checks of whether an argument was passed for a given parameter:

Parameter kind Example Comment Check for no arg passed
Slurpy *@array Don't check using .defined if not @array
Required $foo Can't be omitted (not applicable)
Optional @bar = default Pick a suitable default¹ if @bar =:= default

¹ A suitable default is an Object that has a distinct identity, as may be checked by the WHICH method.

A parameter with a default is always optional, so there is no need to mark it with a ? or the is optional trait.

Then you can use the =:= container identity operator in the body of the routine to check whether this exact default was bound to the parameter.

Example with a positional parameter:

my constant PositionalAt = Positional.new;

sub a (@list = PositionalAt) { say @list =:= PositionalAt }
a;              # OUTPUT: «True␤»
a [1, 2, 3];    # OUTPUT: «False␤»

Example with some scalar parameters:

my constant AnyAt = Any.new;

sub b ($x=AnyAt, :$y=AnyAt) { say $x =:= AnyAt; say $y =:= AnyAt }
b 1;            # OUTPUT: «False␤True␤»
b 1, :2y;       # OUTPUT: «False␤False␤»

If your parameters are typed, then the type smileys can be used with multis like this:

multi c (Int:U $z)  { say 'Undefined' }
multi c (Int:D $z)  { say 'Defined'   }
multi c (Int   $z?) { say 'Omitted'   }
c;              #Omitted
c (Int);        #Undefined
c 42;           #Defined

The examples use names like PositionalAt to reflect that the .WHICH test returns an object of type ObjAt, you are free to make up you own names.

Dynamic variables

Dynamic variables are allowed in signatures although they don't provide special behavior because argument binding does connect two scopes anyway.

Destructuring arguments

Non-scalar parameters can be followed or substituted by a sub-signature in parentheses, which will destructure the argument given. The destructuring of a list is just its elements:

sub first(@array ($first, *@rest)) { $first }

or

sub first([$f, *@]) { $f }

While the destructuring of a hash is its pairs:

sub all-dimensions(% (:length(:$x), :width(:$y), :depth(:$z))) {
        $x andthen $y andthen $z andthen True
    }

Pointy loops can also destructure hashes, allowing assignment to variables:

my %hhgttu = (:40life, :41universe, :42everything);
    for %hhgttu -> (:$key, :$value) {
      say "$key → $value";
    }
    # OUTPUT: «universe → 41␤life → 40␤everything → 42␤»

In general, an object is destructured based on its attributes. A common idiom is to unpack a Pair's key and value in a for loop:

for <Peter Paul Merry>.pairs -> (:key($index), :value($guest)) { }

However, this unpacking of objects as their attributes is only the default behavior. To make an object get destructured differently, change its Capture method.

Sub-signatures

To match against a compound parameter use a sub-signature following the argument name in parentheses.

sub foo(|c(Int, Str)){
       put "called with {c.raku}"
    };
    foo(42, "answer");
    # OUTPUT: «called with \(42, "answer")␤»

Long names

To exclude certain parameters from being considered in multiple dispatch, separate them with a double semicolon.

multi f(Int $i, Str $s;; :$b) { say "$i, $s, {$b.raku}" };
    f(10, 'answer');
    # OUTPUT: «10, answer, Any␤»

Capture parameters

Prefixing a parameter with a vertical bar | makes the parameter a Capture, using up all the remaining positional and named arguments.

This is often used in proto definitions (like proto foo (|) {*}) to indicate that the routine's multi definitions can have any type constraints. See proto for an example.

If bound to a variable, arguments can be forwarded as a whole using the slip operator |.

sub a(Int $i, Str $s) { say $i.^name ~ ' ' ~ $s.^name }
    sub b(|c) { say c.^name; a(|c) }
    b(42, "answer");
    # OUTPUT: «Capture␤Int Str␤»

Parameter traits and modifiers

By default, parameters are bound to their argument and marked as read-only. One can change that with traits on the parameter.

The is copy trait causes the argument to be copied, and allows it to be modified inside the routine

sub count-up($x is copy) {
        $x = ∞ if $x ~~ Whatever;
        .say for 1..$x;
    }

The is rw trait, which stands for is read-write, makes the parameter bind to a variable (or other writable container). Assigning to the parameter changes the value of the variable at the caller side.

sub swap($x is rw, $y is rw) {
        ($x, $y) = ($y, $x);
    }

On slurpy parameters, is rw is reserved for future use by language designers.

The is raw trait is automatically applied to parameters declared with a backslash or a plus sign as a "sigil", and may also be used to make normally sigiled parameters behave like these do. In the special case of slurpies, which normally produce an Array full of Scalars as described above, is raw will instead cause the parameter to produce a List. Each element of that list will be bound directly as raw parameter.

To explicitly ask for a read-only parameter use the is readonly trait. Please note that this applies only to the container. The object inside can very well have mutator methods and Raku will not enforce immutability on the attributes of the object.

Traits can be followed by the where clause:

sub ip-expand-ipv6($ip is copy where m:i/^<[a..f\d\:]>**3..39$/) { }

See Also

Containers

A low-level explanation of Raku containers

Contexts and contextualizers

What are contexts and how to switch into them

Control flow

Statements used to control the flow of execution

Enumeration

An example using the enum type

Exceptions

Using exceptions in Raku

Functions

Functions and functional programming in Raku

Grammars

Parsing and interpreting text

Hashes and maps

Working with associative arrays/dictionaries/hashes

Input/Output the definitive guide

Correctly use Raku IO

Lists, sequences, and arrays

Positional data constructs

Metaobject protocol (MOP)

Introspection and the Raku object system

Native calling interface

Call into dynamic libraries that follow the C calling convention

Raku native types

Using the types the compiler and hardware make available to you

Newline handling in Raku

How the different newline characters are handled, and how to change the behavior

Numerics

Numeric types available in Raku

Object orientation

Object orientation in Raku

Operators

Common Raku infixes, prefixes, postfixes, and more!

Packages

Organizing and referencing namespaced program elements

Performance

Measuring and improving runtime or compile-time performance

Phasers

Program execution phases and corresponding phaser blocks

Pragmas

Special modules that define certain aspects of the behavior of the code

Quoting constructs

Writing strings and word lists, in Raku

Regexes

Pattern matching against strings

Sets, bags, and mixes

Unordered collections of unique and weighted objects in Raku

Statement prefixes

Prefixes that alter the behavior of a statement or a set of them

Data structures

How Raku deals with data structures and what we can expect from them

Subscripts

Accessing data structure elements by index or key

Syntax

General rules of Raku syntax

System interaction

Working with the underlying operating system and running applications

Date and time functions

Processing date and time in Raku

Traits

Compile-time specification of behavior made easy

Unicode versus ASCII symbols

Unicode symbols and their ASCII equivalents

Unicode

Unicode support in Raku

Variables

Variables in Raku

Independent routines

Routines not defined within any class or role.

The Camelia image is copyright 2009 by Larry Wall. "Raku" is trademark of the Yet Another Society. All rights reserved.