Signature.pod6

=begin pod

=TITLE class Signature

=SUBTITLE Parameter list pattern

    class Signature { }

A signature is a static description of the L<parameter|Parameter> list
of a code object.  That is, it describes what and how many arguments
you need to pass to the code or function in order to call it.

Passing arguments to a signature I<binds> the arguments, contained in
a L<Capture|/type/Capture>, to the signature.

X<|signature literal>
X<|:()>
=head1 Signature literals

Signatures appear inside parentheses after L<subroutine|/type/Sub> and
L<method|/type/Method> names, on blocks after a C<< -> >> or C<< <-> >> arrow,
as the input to
L<variable declarators|/language/variables#Variable_declarators_and_scope> like
L<C<my>|/syntax/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' } );

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 C<when> clause since C<:($, $)> represents a
C<Signature> with two scalar, anonymous, arguments, which is a more general
version of C<$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␤»

=head2 Parameter separators

A signature consists of zero or more I<L<parameters|Parameter>>, 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.  The invocant is the
object that was used to call the method, which is usually bound to L<C<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!␤»

X<|type constraint>
X<|Constraint>
=head2 Type constraints

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

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

Type constraints can have any compile-time defined value

=begin code
subset Positive-integer of Int where * > 0;
sub divisors(Positive-integer $n) { $_ if $n %% $_ for 1..$n };
divisors 2.5;
# ERROR «Type check failed in binding to parameter '$n';
# expected Positive-integer but got Rat (2.5) $n)»
divisors -3;
# ERROR: «Constraint type check failed in binding to parameter '$n';
# expected Positive-integer but got Int (-3)»
=end code

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 C<Int>. If it fails, a C<Type check> error is
produced. Once that filter is cleared, the constraint that defined the subset is
checked, producing a C<Constraint type check> error if it fails.

X<|anonymous arguments>
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
L<multi|/language/functions#index-entry-declarator_multi-Multi-dispatch> or to
check the signature of a L<Callable|/type/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 L<type captures|/type/Signature#Type_captures>.

X<|where clause>
In addition to those I<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 C<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 C<where> clause may run more than once for a single
typecheck in some implementations.

The C<where> clause doesn't need to be a code block, anything on the right of
the C<where>-clause will be used to L<smartmatch|/language/operators#infix_~~>
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.

B<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
C<($y ~~ ($y.so && $y.name))>; that is "call C<.so>, and if that is C<True>,
call C<.name>; if that is also C<True> use its value for smartmatching…". It's
the B<result> of C<(.so && .name)> it will be smartmatched against, but we
want to check that both C<.so> and C<.name> are truthy values. That is why
an explicit Block or a L<Junction|/type/Junction> is the right version.

All previous arguments that are not part of a sub-signature in a C<Signature>
are accessible in a C<where>-clause that follows an argument. Therefore,
the C<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 C<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'…»

=head3 Constraining optional arguments

L<Optional arguments|#Optional_and_mandatory_arguments> can have constraints,
too. Any C<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 C<where> clause.

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

=head3 Constraining slurpy arguments

L<Slurpy arguments|#Slurpy_(A.K.A._variadic)_parameters> can not have type
constraints. A C<where>-clause in conjunction with a L<Junction|/type/Junction>
can be used to that effect.

=for code
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' ...»

=head3 Constraining named arguments

Constraints against L<Named arguments|#Positional_vs._named_arguments> apply to
the value part of the L<colon-pair|/type/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")␤»

X<|type constraint,:D>
X<|type constraint,:U>
X<|type constraint,:_>

=head3 Constraining argument definiteness

Normally, a type constraint only checks whether the value of the parameter is of
the correct type. Crucially, both I<object instances> and I<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 C<42> and C<Int> satisfy the match.

Sometimes we need to distinguish between these object instances (C<42>)
and type objects (C<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).perl; # "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 C<:D> type constraint.  This constraint checks that the
value passed is an I<object instance>, in a similar fashion to calling its
L<DEFINITE|/language/mop#DEFINITE> (meta)method.

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

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

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

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

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

Now C<42> fails to match C<Int:U> while C<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 C<can-turn-into> with an object instance as its second parameter
will yield a constraint violation as intended:

=for code :preamble< sub can-turn-into(Str $, Any:U $) {...}>
say can-turn-into("a string", 123);
# OUTPUT: «Parameter '$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, C<:_> can be used, but this
is unnecessary. C<:(Num:_ $)> is the same as C<:(Num $)>.

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

    # Checking a type object
    say Int ~~ Any:D;    # OUTPUT: «False␤»
    say Int ~~ Any:U;    # OUTPUT: «True␤»
    say Int ~~ Any:_;    # 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 L<Classes and Objects|/language/classtut#Starting_with_class>
document further elaborates on the concepts of instances and type
objects and discovering them with the C<.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
L<Any|/type/Any> type object for methods, submethods, and subroutines, and a
L<Mu|/type/Mu> type object for blocks. This means that if you use the  C<: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 I<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␤»

C<$a> has 42 as its default value. With no value, C<$a> will be assigned the
default value declared in the Signature. However, in the second case, it
I<does> receive a value, which happens to be C<Nil>. Assigning C<Nil> to
any variable resets it to its default value, which has been declared as
C<'answer'> by use of the I<default> trait. That explains what happens the second time we call C<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 C<=> for parameters, using the C<default>
trait for variables).

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

=for code :solo
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 I<default> is the type object without the smiley
constraint:

=for code :solo
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 C<:D> and C<:U> to constrain the definiteness of arguments.
Occasionally I<definedness> is used as a synonym for I<definiteness>; this may
be confusing, since the terms have subtly different meanings.

As explained above, I<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
L<DEFINITE|/language/mop#DEFINITE> (meta)method.

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

=head3 Constraining signatures of C<Callable>s

The signature of a L<Callable|/type/Callable> parameter can be constrained by
specifying a L<Signature|/type/Signature> literal right after the parameter (no
whitespace allowed):

    sub f(&c:(Int, Str))  { say c(10, 'ten') };
    sub g(Int $i, Str $s) { $s ~ $i };
    f(&g);
    # OUTPUT: «ten10␤»

This shorthand syntax is available only for parameters with the C<&>
sigil. For others, you need to use the long version:

    sub f($c where .signature ~~ :(Int, Str))  { say $c(10, 'ten') }
    sub g(Num $i, Str $s) { $s ~ $i }
    sub h(Int $i, Str $s) { $s ~ $i }
    # f(&g); # Constraint type check failed
    f(&h);   # OUTPUT: «ten10␤»

=head3 Constraining return types

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

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

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

Type captures are not supported.

X«|-->»
X«|Return type arrow»
=head4 Return type arrow: C<-->>

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 C<,> before it.

    =begin code
    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
    =end code

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.

    =begin code
    sub foo(Str $word --> 123) { say $word; return; }
    my $value = foo("hello"); # OUTPUT: hello
    say $value;               # OUTPUT: 123
    =end code

    =begin code :skip-test<compile time error>
    # The code below will not compile
    sub foo(Str $word --> 123) { say $word; return $word; }
    my $value = foo("hello");
    say $value;
    =end code

=head4 C<returns>

The keyword C<returns> following a signature declaration has the same function
as C«-->» 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.

=for code
sub greeting(Str $name) returns Str { say "Hello, $name" } # Valid

=for code :skip-test<compile time error>
sub favorite-number returns 42 {        } # This will fail.

=head4 C<of>

C<of> is just the real name of the C<returns> keyword.

=for code
sub foo() of Int { 42 }; # Valid

=for code :skip-test<compile time error>
sub foo() of 42 {  };    # This will fail.

=head4 prefix(C-like) form

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

=for code
my Int sub bar { 1 };     # Valid
=for code :skip-test<compile time error>
my 42 sub bad-answer {};  # This will fail.

=head3 X<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
C<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␤»

    use MONKEY;
    augment class Str { method Date() { Date.new(self) } };
    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 (e.g. C<Foo(Bar)> coercer, would call
method C<Foo>). 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:

=begin code
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␤»
=end code

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

X<|parameter,*@>
X<|parameter,*%>
X<|slurpy argument>
=head2 Slurpy (A.K.A. variadic) parameters

A function is X<variadic> if it can take a varying number of arguments; that is,
its arity is not fixed. Therefore, optional, named, and slurpy parameters are
variadic. An array or hash parameter can be marked as I<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.

=begin code
$ = :($a, @b);  # exactly two arguments, where the second one must be Positional
$ = :($a, *@b); # at least one argument, @b slurps up any beyond that
$ = :(*%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␤»
=end code

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

=begin code
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}}␤»
=end code

Note that positional parameters aren't allowed after slurpy parameters:

=begin code :skip-test<compile time error>
:(*@args, $last);
# ===SORRY!=== Error while compiling:
# Cannot put required parameter $last after variadic parameters
=end code

Normally a slurpy parameter will create an L<Array|/type/Array> (or compatible
type), create a new L<Scalar|/type/Scalar> container for each argument, and
assign the value from each argument to those C<Scalar>s.  If the original
argument also had an intermediary C<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:

=for code
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
L<traits and modifiers|#Parameter_Traits_and_Modifiers>,
as described in
L<the section on slurpy array parameters|/type/Signature#Types_of_slurpy_array_parameters>.

=head2 Types of slurpy array parameters

There are three variations to slurpy array parameters.

=item The single asterisk form flattens passed arguments.

=item The double asterisk form does not flatten arguments.

=item 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.

=head3 Flattened slurpy

Slurpy parameters declared with one asterisk will flatten arguments by
dissolving one or more layers of bare L<Iterable|/type/Iterable>s.

=begin code
my @array = <a b c>;
my $list := <d e f>;
sub a(*@a)  { @a.perl.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]»
=end code

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

=head3 X<Unflattened slurpy|parameter, **@>

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

=begin code
my @array = <a b c>;
my $list := <d e f>;
sub b(**@b) { @b.perl.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),]␤»
=end code

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

X<|+ (Single argument rule slurpy)>
=head3 Single argument rule slurpy


A slurpy parameter created using a plus engages the I<"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
L<Iterable|/type/Iterable>, that argument is used to fill the slurpy parameter
array. In any other case, C<+@> works like C<**@>.

=begin code
my @array = <a b c>;
my $list := <d e f>;
sub c(+@b) { @b.perl.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]␤»
=end code

For additional discussion and examples, see L<Slurpy Conventions for Functions|/language/functions#Slurpy_conventions>.

X<|Type capture>
=head2 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

X<|positional argument>
X<|named argument>
=head2 Positional vs. named arguments

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

    $ = :($a);               # a positional argument
    $ = :(:$a);              # a named argument of name 'a'
    $ = :(*@a);              # a slurpy positional argument
    $ = :(*%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
L<Pair|/type/Pair> only those with valid identifiers as keys are recognized as
named arguments.

=for code :allow<L>
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 C<:> 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;

X<|argument aliases>
=head2 Argument aliases

The L<colon-pair|/type/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 C<:> will decide whether we are creating a new named
argument or not. C<:$colour> will not only be the name of the aliased variable,
but also a new named argument (used in the second invocation). However,
C<$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
L<sub MAIN|/language/create-cli#sub_MAIN>.

A function with named arguments can be called dynamically, dereferencing a
L<Pair|/type/Pair> with C<|> 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 C<Hash> into named arguments.

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

A C<Hash> that contains a list may prove problematic when slipped into named
arguments. To avoid the extra layer of containers coerce to L<Map|/type/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:

=for code
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");

X<|optional argument>
=head2 Optional and mandatory arguments

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

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

X<|mandatory named argument>
Named parameters are optional by default, and can be made mandatory with a
trailing exclamation mark:

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

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

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

=head2 Dynamic variables

L<Dynamic variables|/language/variables#The_*_twigil> are allowed in signatures
although they don't provide special behavior because argument binding does
connect two scopes anyway.

X<|destructuring arguments>
=head2 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 L<C<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
L<C<Capture>|/routine/Capture> method.

X<|sub-signature>
=head2 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.perl}"
    };
    foo(42, "answer");
    # OUTPUT: «called with \(42, "answer")␤»

=head2 X<Long names>

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

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

=head2 X<Capture parameters|parameter,|>

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

This is often used in C<proto> definitions (like C<proto foo (|) {*}>) to
indicate that the routine's L<C<multi> definitions|multi> can have any L<type
constraints|#Type_constraints>. See L<proto|/language/functions#proto> for an
example.

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

    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␤»

=head2 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.

X<|trait,is copy>
The C<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;
    }

X<|trait,is rw>
The C<is rw> trait, which stands for I<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, C<is rw> is reserved for future use by language
designers.

X<|trait, is raw>
The L<C<is raw> trait|/type/Parameter#method_raw> is automatically applied to
parameters declared with a L<backslash|/language/variables#Sigilless_variables>
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
C<Array> full of C<Scalar>s as described above, C<is raw> will instead cause
the parameter to produce a C<List>.  Each element of that list will be bound
directly as raw parameter.

X<|trait,is readonly>
To explicitly ask for a read-only parameter use the C<is readonly> trait.
Please note that this applies only to the container. The object inside can very
well have mutator methods and Perl 6 will not enforce immutability on the
attributes of the object.

Traits can be followed by the where clause:

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

=head1 Methods

=head2 method params

    method params(Signature:D: --> Positional)

Returns the list of L<C<Parameter>> objects that make up the signature.

=head2 method arity

    method arity(Signature:D: --> Int:D)

Returns the I<minimal> number of positional arguments required to satisfy
the signature.

=head2 method count

    method count(Signature:D: --> Real:D)

Returns the I<maximal> number of positional arguments which can be bound
to the signature. Returns C<Inf> if there is a slurpy positional parameter.

=head2 method returns

Whatever the Signature's return constraint is:

    :($a, $b --> Int).returns # OUTPUT: «(Int)»

=head2 method ACCEPTS

    multi method ACCEPTS(Signature:D: Signature $topic)
    multi method ACCEPTS(Signature:D: Capture $topic)
    multi method ACCEPTS(Signature:D: Mu \topic)

If C<$topic> is a L<Signature|/type/Signature> returns C<True> if anything
accepted by C<$topic> would also be accepted by the invocant, otherwise returns
C<False>:

    :($a, $b) ~~ :($foo, $bar, $baz?);   # OUTPUT: «True»
    :(Int $n) ~~ :(Str);                 # OUTPUT: «False»

The C<$topic> is a L<Capture|/type/Capture>, returns C<True> if it can be bound
to the invocant, i.e., if a function with invocant's C<Signature> would be able
to be called with the C<$topic>:

    \(1, 2, :foo) ~~ :($a, $b, :foo($bar)); # OUTPUT: «True»
    \(1, :bar)    ~~ :($a);                 # OUTPUT: «False»

Lastly, the candidate with C<Mu \topic> converts C<topic> to
L<Capture|/type/Capture> and follows the same semantics as
L<Capture|/type/Capture> C<$topic>:

    <a b c d>  ~~ :(Int $a);      # OUTPUT: «False»
    42         ~~ :(Int);         # OUTPUT: «False» (Int.Capture throws)
    set(<a b>) ~~ :(:$a, :$b);    # OUTPUT: «True»

Since L«C<where> clauses|/type/Signature#index-entry-where_clause_(Signature)»
are not introspectable, the method cannot determine whether two signatures
L<ACCEPTS|/type/Signature#method_ACCEPTS> the same sort of C<where>-constrained
parameters. Such comparisons will return C<False>. This includes signatures with
literals, which are just sugar for the C<where>-constraints:

    say :(42) ~~ :($ where 42)    # OUTPUT: «False␤»

=head2 method Capture

Defined as:

    method Capture()

Throws C<X::Cannot::Capture>.

=head1 Runtime creation of Signature objects (6.d, 2019.03 and later)

=for code :preamble<role Type {}>
Signature.new(params => (...), returns => Type, arity => 1, count => 1)

In some situations, specifically when working with the MetaObject Protocol,
it makes sense to create C<Signature> objects programmatically.  For this
purpose, you can call the C<new> method with the following named parameters:

=item params

A list of L<Parameter|/type/Parameter> objects for this signature.

=item returns

Any constraint the return value should match.  Defaults to C<Mu>, which
effectively implies no return value constraint check.

=item arity

The I<minimal> number of positional arguments required to satisfy the
signature.  Defaults to the number of C<Parameter> objects given with
the C<params> parameter.

=item count

The I<maximal> number of positional arguments which can be bound to the
signature. Defaults to the C<arity> if not specified.  Specify C<Inf> if
there is a slurpy positional parameter.

=end pod

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