functions.pod6

=begin pod :tag<perl6>

=TITLE Functions

=SUBTITLE Functions and functional programming in Perl 6

Routines are one of the means Perl 6 has to reuse code. They come in
several forms, most notably L<methods|/type/Method>, which belong in
classes and roles and are associated with an object; and functions (also
called I<subroutines> or L<sub|/type/Sub>s, for short), which can be
called independently of objects.

Subroutines default to lexical (C<my>) scoping, and calls to them are
generally resolved at compile time.

Subroutines can have a L<signature|/type/Signature>, also called
I<parameter list>, which specifies which, if any, arguments the
signature expects. It can specify (or leave open) both the number and
types of arguments, and the return value.

Introspection on subroutines is provided via L<C<Routine>|/type/Routine>.

=head1 Defining/Creating/Using functions

=head2 X<Subroutines|declarator,sub>

The basic way to create a subroutine is to use the C<sub> declarator followed by
an optional L<identifier|/language/syntax#Identifiers>:

    sub my-func { say "Look ma, no args!" }
    my-func;

The sub declarator returns a value of type L<Sub|/type/Sub> that can be stored
in any container:

    my &c = sub { say "Look ma, no name!" }
    c;     # OUTPUT: «Look ma, no name!␤»

    my Any:D $f = sub { say 'Still nameless...' }
    $f();  # OUTPUT: «Still nameless...␤»

    my Code \a = sub { say ‚raw containers don't implement postcircumfix:<( )>‘ };
    a.();  # OUTPUT: «raw containers don't implement postcircumfix:<( )>␤»

The declarator C<sub> will declare a new name in the current scope at compile
time. As such any indirection has to be resolved at compile time:

    constant aname = 'foo';
    sub ::(aname) { say 'oi‽' };
    foo;

This will become more useful once macros are added to Perl 6.

To have the subroutine take arguments, a L<signature|Signature> goes
between the subroutine's name and its body, in parentheses:

=for code :allow<B L>
sub exclaim B<($phrase)> {
    say $phrase L<~> "!!!!"
}
exclaim "Howdy, World";

By default, subroutines are L<lexically scoped|/syntax/my>. That is,
C<sub foo {...}> is the same as C<my sub foo {...}> and is only
defined within the current scope.

=begin code
sub escape($str) {
    # Puts a slash before non-alphanumeric characters
    S:g[<-alpha -digit>] = "\\$/" given $str
}

say escape 'foo#bar?'; # OUTPUT: «foo\#bar\?␤»

{
    sub escape($str) {
        # Writes each non-alphanumeric character in its hexadecimal escape
        S:g[<-alpha -digit>] = "\\x[{ $/.ord.base(16) }]" given $str
    }

    say escape 'foo#bar?' # OUTPUT: «foo\x[23]bar\x[3F]␤»
}

# Back to original escape function
say escape 'foo#bar?'; # OUTPUT: «foo\#bar\?␤»
=end code

Subroutines don't have to be named. If unnamed, they're called I<anonymous>
subroutines.

    say sub ($a, $b) { $a ** 2 + $b ** 2 }(3, 4) # OUTPUT: «25␤»

But in this case, it's often desirable to use the more succinct L<block|Block>
syntax. Subroutines and blocks can be called in place, as in the example above.

    say -> $a, $b { $a ** 2 + $b ** 2 }(3, 4)    # OUTPUT: «25␤»

Or even

    say { $^a ** 2 + $^b ** 2 }(3, 4)            # OUTPUT: «25␤»


=head2 X«Blocks and lambdas|syntax,->»

Whenever you see something like C«{ $_ + 42 }»,
C«-> $a, $b { $a ** $b }», or C«{ $^text.indent($:spaces) }», that's
L<Block|/type/Block> syntax. It's used after every C<if>, C<for>, C<while>, etc.

    for 1, 2, 3, 4 -> $a, $b {
        say $a ~ $b;
    }
    # OUTPUT: «12␤34␤»

They can also be used on their own as anonymous blocks of code.

    say { $^a ** 2 + $^b ** 2}(3, 4) # OUTPUT: «25␤»

For block syntax details, see the documentation for the L<Block|/type/Block> type.

=head2 Signatures

The parameters that a function accepts are described in its I<signature>.

=for code :allow<B>
sub formatB<(Str $s)> { ... }
-> B<$a, $b> { ... }

Details about the syntax and use of signatures can be found in the
L<documentation on the C<Signature> class|Signature>.

=head3 Automatic signatures

X<|@_>X<|%_>
If no signature is provided but either of the two automatic variables C<@_> or
C<%_> are used in the function body, a signature with C<*@_> or C<*%_> will be
generated. Both automatic variables can be used at the same time.

    sub s { say @_, %_ };
    say &s.signature # OUTPUT: «(*@_, *%_)␤»

=head2 Arguments
X<|Argument>

Arguments are supplied as a comma separated list. To disambiguate nested calls,
use parentheses:

    sub f(&c){ c() * 2 }; # call the function reference c with empty parameter list
    sub g($p){ $p - 2 };
    say(g(42), 45);       # pass only 42 to g()

When calling a function, positional arguments should be supplied
in the same order as the function's signature.  Named arguments
may be supplied in any order, but it's considered good form to
place named arguments after positional arguments.  Inside the
argument list of a function call, some special syntax is supported:

    sub f(|c){};
    f :named(35);     # A named argument (in "adverb" form)
    f named => 35;    # Also a named argument
    f :35named;       # A named argument using abbreviated adverb form
    f 'named' => 35;  # Not a named argument, a Pair in a positional argument
    my \c = <a b c>.Capture;
    f |c;             # Merge the contents of Capture $c as if they were supplied

Arguments passed to a function are conceptually first collected in a
C<Capture> container.  Details about the syntax and use of these
containers can be found in the L<documentation on the C<Capture> class|Capture>.

When using named arguments, note that normal List "pair-chaining" allows
one to skip commas between named arguments.

    sub f(|c){};
    f :dest</tmp/foo> :src</tmp/bar> :lines(512);
    f :32x :50y :110z;   # This flavor of "adverb" works, too
    f :a:b:c;            # The spaces are also optional.

=head2 Return values

Any C<Block> or C<Routine> will provide the value of its last expression as a return value
to the caller. If either L<return|/language/control#return> or
L<return-rw|/language/control#return-rw> is called, then its parameter, if any,
will become the return value. The default return value is L<Nil|/type/Nil>.

    sub a { 42 };
    sub b { say a };
    sub c { };
    b;     # OUTPUT: «42␤»
    say c; # OUTPUT: «Nil␤»

Multiple return values are returned as a list or by creating a
L<Capture|/type/Capture>. Destructuring can be used to untangle multiple return
values.

    sub a { 42, 'answer' };
    put a.perl;
    # OUTPUT: «(42, "answer")␤»

    my ($n, $s) = a;
    put [$s, $n];
    # OUTPUT: «answer 42␤»

    sub b { <a b c>.Capture };
    put b.perl;
    # OUTPUT: «\("a", "b", "c")␤»

=head2 Return type constraints

Perl 6 has many ways to specify a function's return type:

=for code
sub foo(--> Int)      {}; say &foo.returns; # OUTPUT: «(Int)␤»
=for code
sub foo() returns Int {}; say &foo.returns; # OUTPUT: «(Int)␤»
=for code
sub foo() of Int      {}; say &foo.returns; # OUTPUT: «(Int)␤»
=for code
my Int sub foo()      {}; say &foo.returns; # OUTPUT: «(Int)␤»

Attempting to return values of another type will cause a compilation error.

=for code
sub foo() returns Int { "a"; }; foo; # Type check fails

C<returns> and C<of> are equivalent, and both take only a Type since they are declaring a trait of the L<Callable|/type/Callable>. The last declaration is, in fact, a type declaration, which obviously can take only a type. C«-->», however, can take either undefined or definite values.

Note that C<Nil> and C<Failure> are exempt from return type constraints and
can be returned from any routine, regardless of its constraint:

=for code
sub foo() returns Int { fail   }; foo; # Failure returned
sub bar() returns Int { return }; bar; # Nil returned

=head2 X<Multi-dispatch|declarator,multi>

Perl 6 allows for writing several routines with the same name but different
signatures. When the routine is called by name, the runtime environment
determines the proper I<candidate> and invokes it.

Each candidate is declared with the C<multi> keyword. Dispatch happens depending
on the number (L<arity|/type/Routine#(Code)_method_arity>), type and name of arguments.
Consider the following example:

=begin code
# version 1
multi happy-birthday( $name ) {
    say "Happy Birthday $name !";
}

# version 2
multi happy-birthday( $name, $age ) {
    say "Happy {$age}th Birthday $name !";
}

# version 3
multi happy-birthday( :$name, :$age, :$title  = 'Mr' ) {
    say "Happy {$age}th Birthday $title $name !";
}


# calls version 1 (arity)
happy-birthday 'Larry';                        # OUTPUT: «Happy Birthday Larry !␤»
# calls version 2 (arity)
happy-birthday 'Luca', 40;                     # OUTPUT: «Happy 40th Birthday Luca !␤»
# calls version 3
# (named arguments win against arity)
happy-birthday( age => '50', name => 'John' ); # OUTPUT: «Happy 50th Birthday Mr John !␤»
# calls version 2 (arity)
happy-birthday( 'Jack', 25 );                  # OUTPUT: «Happy 25th Birthday Jack !␤»

=end code

The first two versions of the C<happy-birthday> sub differs only in the arity
(number of arguments), while the third version uses named arguments and is
chosen only when named arguments are used, even if the arity is the same of
another C<multi> candidate.

When two sub have the same arity, the type of the arguments drive the dispatch;
when there are named arguments they drive the dispatch even when their type is
the same as another candidate:

=begin code
multi happy-birthday( Str $name, Int $age ) {
    say "Happy {$age}th Birthday $name !";
}

multi happy-birthday( Str $name, Str $title ) {
    say "Happy Birthday $title $name !";
}

multi happy-birthday( Str :$name, Int :$age ) {
    say "Happy Birthday $name, you turned $age !";
}

happy-birthday 'Luca', 40;                 # OUTPUT: «Happy 40th Birthday Luca !␤»
happy-birthday 'Luca', 'Mr';               # OUTPUT: «Happy Birthday Mr Luca !␤»
happy-birthday age => 40, name => 'Luca';  # OUTPUT: «Happy Birthday Luca, you turned 40 !␤»

=end code

Named parameters participate in the dispatch even if they are not provided in
the call. Therefore a multi candidate with named parameters will be given
precedence.

For more information about type constraints see the documentation
for the L<Signature|/type/Signature#Type_constraints> class.

    multi as-json(Bool $d) { $d ?? 'true' !! 'false'; }
    multi as-json(Real $d) { ~$d }
    multi as-json(@d)      { sprintf '[%s]', @d.map(&as-json).join(', ') }

    say as-json( True );                        # OUTPUT: «true␤»
    say as-json( 10.3 );                        # OUTPUT: «10.3␤»
    say as-json( [ True, 10.3, False, 24 ] );   # OUTPUT: «[true, 10.3, false, 24]␤»



C<multi> without any specific routine type always defaults to a C<sub>, but you
can use it on methods as well. The candidates are all the multi methods of the
object:

    class Congrats {
        multi method congratulate($reason, $name) {
            say "Hooray for your $reason, $name";
        }
    }

    role BirthdayCongrats {
        multi method congratulate('birthday', $name) {
            say "Happy birthday, $name";
        }
        multi method congratulate('birthday', $name, $age) {
            say "Happy {$age}th birthday, $name";
        }
    }

    my $congrats = Congrats.new does BirthdayCongrats;

    $congrats.congratulate('promotion','Cindy'); # OUTPUT: «Hooray for your promotion, Cindy␤»
    $congrats.congratulate('birthday','Bob');    # OUTPUT: «Happy birthday, Bob␤»

Unlike C<sub>, if you use named parameters with multi methods, the parameters
must be required parameters to behave as expected.

Please note that a non-multi sub or operator will hide multi candidates of the
same name in any parent scope or child scope. The same is true for imported
non-multi candidates.

=head3 X<proto|declarator>

C<proto> is a way to formally declare commonalities between C<multi>
candidates. It acts as a wrapper that can validate but not modify
arguments. Consider this basic example:

    proto congratulate(Str $reason, Str $name, |) {*}
    multi congratulate($reason, $name) {
       say "Hooray for your $reason, $name";
    }
    multi congratulate($reason, $name, Int $rank) {
       say "Hooray for your $reason, $name -- got rank $rank!";
    }

    congratulate('being a cool number', 'Fred');     # OK
    congratulate('being a cool number', 'Fred', 42); # OK

=for code :skip-test<Proto match error>
congratulate('being a cool number', 42);         # Proto match error

The proto insists that all C<multi congratulate> subs conform to the basic
signature of two strings, optionally followed by further parameters. The C<|> is
an un-named C<Capture> parameter, and allows a C<multi> to take additional
arguments. The first two calls succeed, but the third fails (at compile time)
because C<42> doesn't match C<Str>.

=for code :preamble<sub congratulate {}>
say &congratulate.signature # OUTPUT: «(Str $reason, Str $name, | is raw)␤»

You can give the C<proto> a function body, and place the C<{*}> where
you want the dispatch to be done.

    # attempts to notify someone -- False if unsuccessful
    proto notify(Str $user,Str $msg) {
       my \hour = DateTime.now.hour;
       if hour > 8 or hour < 22 {
          return {*};
       } else {
          # we can't notify someone when they might be sleeping
          return False;
       }
    }

C<{*}> always dispatches to candidates with the parameters it's called
with. Parameter defaults and type coercions will work but are not passed on.

=for code
proto mistake-proto(Str() $str, Int $number = 42) {*}
multi mistake-proto($str, $number) { say $str.^name }
mistake-proto(7, 42);  # OUTPUT: «Int␤» -- not passed on
=for code :skip-test<compilation error>
mistake-proto('test'); # fails -- not passed on

=head2 X<only|declarator>

The C<only> keyword preceding C<sub> or C<method> indicates that it will be the
only function with that name that inhabits a given namespace.

    only sub you () {"Can make all the world seem right"};

This will make other declarations in the same namespace, such as

    sub you ( $can ) { "Make the darkness bright" }

fail with an exception of type C<X::Redeclaration>. C<only> is the default value
for all subs; in the case above, not declaring the first subroutine as C<only>
will yield exactly the same error; however, nothing prevents future developers
from declaring a proto and preceding the names with C<multi>. Using C<only>
before a routine is a L<defensive programming|https://en.wikipedia.org/wiki/Defensive_programming> feature that
declares the intention of not having routines with the same name declared in the
same namespace in the future.

=begin code :lang<text>
(exit code 1)
===SORRY!=== Error while compiling /tmp/hDM1N2OAOo
Redeclaration of routine 'you' (did you mean to declare a multi-sub?)
at /tmp/hDM1N2OAOo:1
------> ( $can ) { "Make the darkness bright" }⏏<EOL>
=end code

Anonymous sub cannot be declared C<only>. C<only sub {}'> will throw an error of
type, surprisingly, C<X::Anon::Multi>.

=head1 Conventions and idioms

While the dispatch system described above provides a lot of flexibility,
there are some conventions that most internal functions, and those in
many modules, will follow.

=head2 Slurpy conventions

Perhaps the most important one of these conventions is the way slurpy list
arguments are handled.  Most of the time, functions will not automatically
flatten slurpy lists. The rare exceptions are those functions that don't have a
reasonable behavior on lists of lists (e.g., L<chrs|/routine/chrs>) or where
there is a conflict with an established idiom (e.g., L<pop|/routine/pop> being
the inverse of L<push|/routine/push>).

If you wish to match this look and feel, any L<Iterable|/type/Iterable> argument must
be broken out element-by-element using a C<**@> slurpy, with two nuances:

=item An L<Iterable|/type/Iterable> inside a L<Scalar container|/language/containers#Scalar_containers> doesn't count.

=item L<List|/type/List>s created with a L<C<,>|/routine/,> at the top level only count as one L<Iterable|/type/Iterable>.

This can be achieved by using a slurpy with a C<+> or C<+@> instead of C<**>:

    sub grab(+@a) { "grab $_".say for @a }

which is shorthand for something very close to:

    multi sub grab(**@a) { "grab $_".say for @a }
    multi sub grab(\a) {
        a ~~ Iterable and a.VAR !~~ Scalar ?? nextwith(|a) !! nextwith(a,)
    }

This results in the following behavior, which is known as the
I<"single argument rule"> and is important to understand when invoking slurpy functions:

=for code :preamble<sub grab(+@a) {};>
grab(1, 2);      # OUTPUT: «grab 1␤grab 2␤»
grab((1, 2));    # OUTPUT: «grab 1␤grab 2␤»
grab($(1, 2));   # OUTPUT: «grab 1 2␤»
grab((1, 2), 3); # OUTPUT: «grab 1 2␤grab 3␤»

This also makes user-requested flattening feel consistent whether there is
one sublist, or many:

=for code :preamble<sub grab(+@a) {};>
grab(flat (1, 2), (3, 4));   # OUTPUT: «grab 1␤grab 2␤grab 3␤grab 4␤»
grab(flat $(1, 2), $(3, 4)); # OUTPUT: «grab 1 2␤grab 3 4␤»
grab(flat (1, 2));           # OUTPUT: «grab 1␤grab 2␤»
grab(flat $(1, 2));          # OUTPUT: «grab 1␤grab 2␤»

It's worth noting that mixing binding and sigilless variables
in these cases requires a bit of finesse, because there is no L<Scalar|/type/Scalar>
intermediary used during binding.

=for code :preamble<sub grab(+@a) {};>
my $a = (1, 2);  # Normal assignment, equivalent to $(1, 2)
grab($a);        # OUTPUT: «grab 1 2␤»
my $b := (1, 2); # Binding, $b links directly to a bare (1, 2)
grab($b);        # OUTPUT: «grab 1␤grab 2␤»
my \c = (1, 2);  # Sigilless variables always bind, even with '='
grab(c);         # OUTPUT: «grab 1␤grab 2␤»

=head1 Functions are first-class objects

Functions and other code objects can be passed around as values, just like any
other object.

There are several ways to get hold of a code object. You can assign it to a
variable at the point of declaration:

    my $square = sub (Numeric $x) { $x * $x }
    # and then use it:
    say $square(6);    # OUTPUT: «36␤»

X<|prefix &>
Or you can reference an existing named function by using the C<&>-sigil in
front of it.

    sub square($x) { $x * $x };

    # get hold of a reference to the function:
    my $func = &square

This is very useful for I<higher order functions>, that is, functions that
take other functions as input. A simple one is L<map|/type/List#routine_map>,
which applies a function to each input element:

    sub square($x) { $x * $x };
    my @squared = map &square,  1..5;
    say join ', ', @squared;        # OUTPUT: «1, 4, 9, 16, 25␤»

=head2 Z<>Infix form

To call a subroutine with 2 arguments like an infix operator, use a subroutine
reference surrounded by C<[> and C<]>.

    sub plus { $^a + $^b };
    say 21 [&plus] 21;
    # OUTPUT: «42␤»

X<|closures>
=head2 Closures

All code objects in Perl 6 are I<closures>, which means they can reference
lexical variables from an outer scope.

    sub generate-sub($x) {
        my $y = 2 * $x;
        return sub { say $y };
        #      ^^^^^^^^^^^^^^  inner sub, uses $y
    }
    my $generated = generate-sub(21);
    $generated(); # OUTPUT: «42␤»

Here, C<$y> is a lexical variable inside C<generate-sub>, and the inner
subroutine that is returned uses it. By the time that inner sub is called,
C<generate-sub> has already exited. Yet the inner sub can still use C<$y>,
because it I<closed> over the variable.

Another closure example is the use of L<map|/type/List#routine_map> to multiply
a list of numbers:

    my $multiply-by = 5;
    say join ', ', map { $_ * $multiply-by }, 1..5;     # OUTPUT: «5, 10, 15, 20, 25␤»

Here, the block passed to C<map> references the variable C<$multiply-by> from
the outer scope, making the block a closure.

Languages without closures cannot easily provide higher-order functions that
are as easy to use and powerful as C<map>.

=head2 Routines

Routines are code objects that conform to L<type C<Routine>|/type/Routine>, most
notably L<C<Sub>|/type/Sub>, L<C<Method>|/type/Method>, L<C<Regex>|/type/Regex>
and L<C<Submethod>|/type/Submethod>.

They carry extra functionality in addition to what a L<C<Block>|/type/Block>
supplies: they can come as L<multis|#Multi-dispatch>, you can
L<wrap|/type/Routine#method_wrap> them, and exit early with C<return>:

    my $keywords = set <if for unless while>;

    sub has-keyword(*@words) {
        for @words -> $word {
            return True if $word (elem) $keywords;
        }
        False;
    }

    say has-keyword 'not', 'one', 'here';       # OUTPUT: «False␤»
    say has-keyword 'but', 'here', 'for';       # OUTPUT: «True␤»

Here, C<return> doesn't just leave the block inside which it was called, but
the whole routine. In general, blocks are transparent to C<return>, they
attach to the outermost routine.

X<|use soft (pragma)>
Routines can be inlined and as such provide an obstacle for wrapping. Use the
pragma C<use soft;> to prevent inlining to allow wrapping at runtime.

    sub testee(Int $i, Str $s){
        rand.Rat * $i ~ $s;
    }

    sub wrap-to-debug(&c){
        say "wrapping {&c.name} with arguments {&c.signature.perl}";
        &c.wrap: sub (|args){
            note "calling {&c.name} with {args.gist}";
            my \ret-val := callwith(|args);
            note "returned from {&c.name} with return value {ret-val.perl}";
            ret-val
        }
    }

    my $testee-handler = wrap-to-debug(&testee);
    # OUTPUT: «wrapping testee with arguments :(Int $i, Str $s)»

    say testee(10, "ten");
    # OUTPUT: «calling testee with \(10, "ten")␤returned from testee with return value "6.151190ten"␤6.151190ten»
    &testee.unwrap($testee-handler);
    say testee(10, "ten");
    # OUTPUT: «6.151190ten␤»


=comment Important ones: candidates, wrap, unwrap, assuming, arity, count

=head1 Defining operators

Operators are just subroutines with funny names. The funny names are composed
of the category name (C<infix>, C<prefix>, C<postfix>, C<circumfix>,
C<postcircumfix>), followed by a colon, and a list of the operator name or
names (two components in the case of circumfix and postcircumfix).

This works both for adding multi candidates to existing operators and for
defining new ones. In the latter case, the definition of the new subroutine
automatically installs the new operator into the grammar, but only in the
current lexical scope. Importing an operator via C<use> or C<import> also
makes it available.

=begin code
# adding a multi candidate to an existing operator:
multi infix:<+>(Int $x, "same") { 2 * $x };
say 21 + "same";            # OUTPUT: «42␤»

# defining a new operator
sub postfix:<!>(Int $x where { $x >= 0 }) { [*] 1..$x };
say 6!;                     # OUTPUT: «720␤»
=end code

The operator declaration becomes available as soon as possible, so you can
recurse into a just-defined operator:

=begin code
sub postfix:<!>(Int $x where { $x >= 0 }) {
    $x == 0 ?? 1 !! $x * ($x - 1)!
}
say 6!;                     # OUTPUT: «720␤»
=end code

Circumfix and postcircumfix operators are made of two delimiters, one opening
and one closing.

=begin code
sub circumfix:<START END>(*@elems) {
    "start", @elems, "end"
}

say START 'a', 'b', 'c' END;        # OUTPUT: «(start [a b c] end)␤»
=end code

Postcircumfixes also receive the term after which they are parsed as
an argument:

=begin code
sub postcircumfix:<!! !!>($left, $inside) {
    "$left -> ( $inside )"
}
say 42!! 1 !!;      # OUTPUT: «42 -> ( 1 )␤»
=end code

Blocks can be assigned directly to operator names. Use a variable declarator and
prefix the operator name with a C<&>-sigil.

    my &infix:<ieq> = -> |l { [eq] l>>.fc };
    say "abc" ieq "Abc";
    # OUTPUT: «True␤»

X«|is tighter»X«|is equiv»X«|is looser»
=head2 Precedence

Operator precedence in Perl 6 is specified relatively to existing operators. The
traits C<is tighter>, C<is equiv> and C<is looser> can be provided with an
operator to indicate how the precedence of the new operators is related to
other, existing ones. More than one trait can be applied.

For example, C«infix:<*>» has a tighter precedence than C«infix:<+>»,
and squeezing one in between works like this:

=begin code
sub infix:<!!>($a, $b) is tighter(&infix:<+>) {
    2 * ($a + $b)
}

say 1 + 2 * 3 !! 4;     # OUTPUT: «21␤»
=end code

Here, the C<1 + 2 * 3 !! 4> is parsed as C<1 + ((2 * 3) !! 4)>, because the
precedence of the new C<!!> operator is between that of C<+> and C<*>.

The same effect could have been achieved with:

    sub infix:<!!>($a, $b) is looser(&infix:<*>) { ... }

To put a new operator on the same precedence level as an existing operator,
use C<is equiv(&other-operator)> instead.


=head2 Associativity

When the same operator appears several times in a row, there are multiple
possible interpretations. For example:

    1 + 2 + 3

could be parsed as

    (1 + 2) + 3         # left associative

or as

    1 + (2 + 3)         # right associative

For addition of real numbers, the distinction is somewhat moot, because C<+> is
L<mathematically associative|https://en.wikipedia.org/wiki/Associative_property>.

But for other operators it matters a great deal. For example, for the
exponentiation/power operator, C<< infix:<**> >>:

    say 2 ** (2 ** 3);      # OUTPUT: «256␤»
    say (2 ** 2) ** 3;      # OUTPUT: «64␤»

Perl 6 has the following possible associativity configurations:

=begin table

    A   Assoc     Meaning of $a ! $b ! $c
    =   =====     =======================
    L   left      ($a ! $b) ! $c
    R   right     $a ! ($b ! $c)
    N   non       ILLEGAL
    C   chain     ($a ! $b) and ($b ! $c)
    X   list      infix:<!>($a; $b; $c)

=end table

X<|is assoc (trait)>
You can specify the associativity of an operator with the C<is assoc> trait,
where C<left> is the default associativity.

=begin code
sub infix:<§>(*@a) is assoc<list> {
    '(' ~ @a.join('|') ~ ')';
}

say 1 § 2 § 3;      # OUTPUT: «(1|2|3)␤»
=end code

=head1 Traits

I<Traits> are subroutines that run at compile time and modify the behavior of a
type, variable, routine, attribute, or other language object.

Examples of traits are:

=for code :skip-test
class ChildClass is ParentClass { ... }
#                ^^ trait, with argument ParentClass
has $.attrib is rw;
#            ^^^^^  trait with name 'rw'
class SomeClass does AnotherRole { ... }
#               ^^^^ trait
has $!another-attribute handles <close>;
#                       ^^^^^^^ trait

... and also C<is tighter>, C<is looser>, C<is equiv> and C<is assoc> from the
previous section.

Traits are subs declared in the form C<< trait_mod<VERB> >>, where C<VERB>
stands for the name like C<is>, C<does> or C<handles>. It receives the modified
thing as argument, and the name as a named argument. See L<Sub|/type/Sub#Traits>
for details.

=begin code
multi sub trait_mod:<is>(Routine $r, :$doubles!) {
    $r.wrap({
        2 * callsame;
    });
}

sub square($x) is doubles {
    $x * $x;
}

say square 3;       # OUTPUT: «18␤»
=end code

See L<type Routine|/type/Routine> for the documentation of built-in routine
traits.

=head1 Re-dispatching

There are cases in which a routine might want to call the next method
from a chain. This chain could be a list of parent classes in a class
hierarchy, or it could be less specific multi candidates from a multi
dispatch, or it could be the inner routine from a C<wrap>.

Fortunately, we have a series of re-dispatching tools that help us to make
it easy.

=head2 sub callsame

X<|dispatch,callsame>

C<callsame> calls the next matching candidate with the same arguments that were
used for the current candidate and returns that candidate's return value.

=begin code
proto a(|) {*}

multi a(Any $x) {
    say "Any $x";
    return 5;
}

multi a(Int $x) {
    say "Int $x";
    my $res = callsame;
    say "Back in Int with $res";
}

a 1;        # OUTPUT: «Int 1␤Any 1␤Back in Int with 5␤»
=end code

=head2 sub callwith

X<|dispatch,callwith>

C<callwith> calls the next candidate matching the original signature, that is,
the next function that could possibly be used with the arguments provided by
users and returns that candidate's return value.

=begin code
proto a(|) {*}

multi a(Any $x) {
    say "Any $x";
    return 5;
}
multi a(Int $x) {
    say "Int $x";
    my $res = callwith($x + 1);
    say "Back in Int with $res";
}

a 1;        # OUTPUT: «Int 1␤Any 2␤Back in Int with 5␤»
=end code

Here, C<a 1> calls the most specific C<Int> candidate first, and C<callwith>
re-dispatches to the less specific C<Any> candidate. Note that although our
parameter C<$x + 1> is an C<Int>, still we call the next candidate in the chain.

In this case, for example:

=begin code
proto how-many(|) {*}

multi how-many( Associative $a ) {
    say "Associative $a ";
    my $calling = callwith( 1 => $a );
    return $calling;
}

multi how-many( Pair $a ) {
    say "Pair $a ";
    return "There is $a "

}

multi how-many( Hash $a ) {
    say "Hash $a";
    return "Hashing $a";
}

my $little-piggie = little => 'piggie';
say $little-piggie.^name;        # OUTPUT: «Pair␤»
say &how-many.cando( \( $little-piggie ));
# OUTPUT: «(sub how-many (Pair $a) { #`(Sub|68970512) ... } sub how-many (Associative $a) { #`(Sub|68970664) ... })␤»
say how-many( $little-piggie  ); # OUTPUT: «Pair little     piggie␤There is little piggie␤»
=end code

the only candidates that take the C<Pair> argument supplied by the user are the
two functions defined first. Although a C<Pair> can be easily coerced to a
C<Hash>, here is how signatures match:

=for code
say :( Pair ) ~~ :( Associative ); # OUTPUT: «True␤»
say :( Pair ) ~~ :( Hash );        # OUTPUT: «False␤»

The arguments provided by us are a C<Pair>. It does not match a C<Hash>, so the
corresponding function is thus not included in the list of candidates, as can be
seen by the output of C<&how-many.cando( \( $little-piggie ));>.

=head2 sub nextsame

X<|dispatch,nextsame>

C<nextsame> calls the next matching candidate with the same arguments that were
used for the current candidate and B<never> returns.

=begin code
proto a(|) {*}

multi a(Any $x) {
    say "Any $x";
    return 5;
}
multi a(Int $x) {
    say "Int $x";
    nextsame;
    say "never executed because nextsame doesn't return";
}

a 1;        # OUTPUT: «Int 1␤Any 1␤»
=end code

=head2 sub nextwith

X<|dispatch,nextwith>

C<nextwith> calls the next matching candidate with arguments provided by users
and B<never> returns.

=begin code
proto a(|) {*}

multi a(Any $x) {
    say "Any $x";
    return 5;
}
multi a(Int $x) {
    say "Int $x";
    nextwith($x + 1);
    say "never executed because nextsame doesn't return";
}

a 1;        # OUTPUT: «Int 1␤Any 2␤»
=end code

=head2 sub samewith

X<|dispatch,samewith>

C<samewith> calls current candidate again with arguments provided by users
and returns return value of the new instance of current candidate.

=begin code
proto a(|) {*}

multi a(Int $x) {
  return 1 unless $x > 1;
  return $x * samewith($x-1);
}

say (a 10); # OUTPUT: «36288002␤»
=end code

X<|dispatch,nextcallee>
=head2 sub nextcallee

Redispatch may be required to call a block that is not the current scope what
provides C<nextsame> and friends with the problem to referring to the wrong
scope. Use C<nextcallee> to capture the right candidate and call it at the
desired time.

=begin code
proto pick-winner(|) {*}

multi pick-winner (Int \s) {
    my &nextone = nextcallee;
    Promise.in(π²).then: { nextone s }
}
multi pick-winner { say "Woot! $^w won" }

with pick-winner ^5 .pick -> \result {
    say "And the winner is...";
    await result;
}

# OUTPUT:
# And the winner is...
# Woot! 3 won
=end code

The Int candidate takes the C<nextcallee> and then fires up a Promise to be
executed in parallel, after some timeout, and then returns. We can't use
C<nextsame> here, because it'd be trying to C<nextsame> the Promise's block
instead of our original routine.

X<|dispatch,wrapped routines>
=head2 Wrapped routines


Besides those are mentioned above, re-dispatch is helpful in more situations.
One is for dispatching to wrapped routines:

=begin code
# enable wrapping:
use soft;

# function to be wrapped:
sub square-root($x) { $x.sqrt }

&square-root.wrap(sub ($num) {
   nextsame if $num >= 0;
   1i * callwith(abs($num));
});

say square-root(4);     # OUTPUT: «2␤»
say square-root(-4);    # OUTPUT: «0+2i␤»
=end code

=head2 Routines of parent class

Another use case is to re-dispatch to methods from parent classes.

=begin code
class LoggedVersion is Version {
    method new(|c) {
        note "New version object created with arguments " ~ c.perl;
        nextsame;
    }
}

say LoggedVersion.new('1.0.2');
=end code

=head1 Coercion types

Coercion types force a specific type for routine arguments while allowing
the routine itself to accept a wider input. When invoked, the arguments are
narrowed automatically to the stricter type, and therefore within the routine
the arguments have always the desired type.

In the case the arguments cannot be converted to the stricter type, a I<Type Check> error
is thrown.

=begin code
sub double(Int(Cool) $x) {
    2 * $x
}

say double '21';    # OUTPUT: «42␤»
say double  21;     # OUTPUT: «42␤»
say double Any;     # Type check failed in binding $x; expected 'Cool' but got 'Any'
=end code

In the above example, the L<Int|/type/Int> is the target type to which the argument C<$x> will be coerced, and
L<Cool|/type/Cool> is the type that the routine accepts as wider input.

If the accepted wider input type is L<Any|/type/Any>, it is possible to abbreviate the coercion C<Int(Any)>
omitting the C<Any> type, thus resulting in  C<Int()>.

The coercion works by looking for a method with the same name
as the target type: if such method is found on the argument, it is invoked to
convert the latter to the expected narrow type.
From the above, it is clear that it is possible to provide coercion among user types
just providing the required methods:

=begin code
class Bar {
   has $.msg;
}

class Foo {
   has $.msg = "I'm a foo!";

   # allows coercion from Foo to Bar
   method Bar {
       Bar.new(:msg($.msg ~ ' But I am now Bar.'));
   }
}

# wants a Bar, but accepts Any
sub print-bar(Bar() $bar) {
   say $bar.^name; # OUTPUT: «Bar␤»
   say $bar.msg;   # OUTPUT: «I'm a foo! But I am now Bar.␤»
}

print-bar Foo.new;
=end code

In the above code, once a C<Foo> instance is passed as argument to C<print-bar>,
the C<Foo.Bar> method is called and the result is placed into C<$bar>.

Coercion types are supposed to work wherever types work, but Rakudo currently
(2018.05) only implements them in signatures, for both parameters and return
types.

=comment Non-parameter coercion types will theoretically be working in 6.2. Updated the reference above to latest version.

Coercion also works with return types:

=begin code
sub are-equal (Int $x, Int $y --> Bool(Int) ) { $x - $y };

for (2,4) X (1,2) -> ($a,$b) {
    say "Are $a and $b equal? ", are-equal($a, $b);
} #  OUTPUT: «Are 2 and 1 equal? True␤Are 2 and 2 equal? False␤Are 4 and 1 equal? True␤Are 4 and 2 equal? True␤»
=end code

In this case, we are coercing an C<Int> to a C<Bool>, which is then printed (put
into a string context) in the C<for> loop that calls the function.

=head1 sub MAIN

Declaring a C<sub MAIN> is not compulsory in Perl 6 scripts, but you can
provide one to create a
L<command line interface|/language/create-cli>
for your script.

=end pod

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