Integrate with Sarathy. perl.h and util.c required manual resolving.

[perl5.git] / pod / perlsub.pod

=head1 NAME

perlsub - Perl subroutines

=head1 SYNOPSIS

To declare subroutines:

    sub NAME;             # A "forward" declaration.
    sub NAME(PROTO);      #  ditto, but with prototypes

    sub NAME BLOCK        # A declaration and a definition.
    sub NAME(PROTO) BLOCK #  ditto, but with prototypes

To define an anonymous subroutine at runtime:

    $subref = sub BLOCK;            # no proto
    $subref = sub (PROTO) BLOCK;    # with proto

To import subroutines:

    use MODULE qw(NAME1 NAME2 NAME3);

To call subroutines:

    NAME(LIST);    # & is optional with parentheses.
    NAME LIST;     # Parentheses optional if predeclared/imported.
    &NAME(LIST);   # Circumvent prototypes.
    &NAME;         # Makes current @_ visible to called subroutine.

=head1 DESCRIPTION

Like many languages, Perl provides for user-defined subroutines.
These may be located anywhere in the main program, loaded in from
other files via the C<do>, C<require>, or C<use> keywords, or
generated on the fly using C<eval> or anonymous subroutines (closures).
You can even call a function indirectly using a variable containing
its name or a CODE reference.

The Perl model for function call and return values is simple: all
functions are passed as parameters one single flat list of scalars, and
all functions likewise return to their caller one single flat list of
scalars.  Any arrays or hashes in these call and return lists will
collapse, losing their identities--but you may always use
pass-by-reference instead to avoid this.  Both call and return lists may
contain as many or as few scalar elements as you'd like.  (Often a
function without an explicit return statement is called a subroutine, but
there's really no difference from Perl's perspective.)

Any arguments passed in show up in the array C<@_>.  Therefore, if
you called a function with two arguments, those would be stored in
C<$_[0]> and C<$_[1]>.  The array C<@_> is a local array, but its
elements are aliases for the actual scalar parameters.  In particular,
if an element C<$_[0]> is updated, the corresponding argument is
updated (or an error occurs if it is not updatable).  If an argument
is an array or hash element which did not exist when the function
was called, that element is created only when (and if) it is modified
or a reference to it is taken.  (Some earlier versions of Perl
created the element whether or not the element was assigned to.)
Assigning to the whole array C<@_> removes that aliasing, and does
not update any arguments.

The return value of a subroutine is the value of the last expression
evaluated.  More explicitly, a C<return> statement may be used to exit the
subroutine, optionally specifying the returned value, which will be
evaluated in the appropriate context (list, scalar, or void) depending
on the context of the subroutine call.  If you specify no return value,
the subroutine returns an empty list in list context, the undefined
value in scalar context, or nothing in void context.  If you return
one or more aggregates (arrays and hashes), these will be flattened
together into one large indistinguishable list.

Perl does not have named formal parameters.  In practice all you
do is assign to a C<my()> list of these.  Variables that aren't
declared to be private are global variables.  For gory details
on creating private variables, see L<"Private Variables via my()">
and L<"Temporary Values via local()">.  To create protected
environments for a set of functions in a separate package (and
probably a separate file), see L<perlmod/"Packages">.

Example:

    sub max {
        my $max = shift(@_);
        foreach $foo (@_) {
            $max = $foo if $max < $foo;
        }
        return $max;
    }
    $bestday = max($mon,$tue,$wed,$thu,$fri);

Example:

    # get a line, combining continuation lines
    #  that start with whitespace

    sub get_line {
        $thisline = $lookahead;  # global variables!
        LINE: while (defined($lookahead = <STDIN>)) {
            if ($lookahead =~ /^[ \t]/) {
                $thisline .= $lookahead;
            }
            else {
                last LINE;
            }
        }
        return $thisline;
    }

    $lookahead = <STDIN>;       # get first line
    while (defined($line = get_line())) {
        ...
    }

Asisng to a list of private variables to name your arguments:

    sub maybeset {
        my($key, $value) = @_;
        $Foo{$key} = $value unless $Foo{$key};
    }

Because the assignment copies the values, this also has the effect
of turning call-by-reference into call-by-value.  Otherwise a
function is free to do in-place modifications of C<@_> and change
its caller's values.

    upcase_in($v1, $v2);  # this changes $v1 and $v2
    sub upcase_in {
        for (@_) { tr/a-z/A-Z/ }
    }

You aren't allowed to modify constants in this way, of course.  If an
argument were actually literal and you tried to change it, you'd take a
(presumably fatal) exception.   For example, this won't work:

    upcase_in("frederick");

It would be much safer if the C<upcase_in()> function
were written to return a copy of its parameters instead
of changing them in place:

    ($v3, $v4) = upcase($v1, $v2);  # this doesn't change $v1 and $v2
    sub upcase {
        return unless defined wantarray;  # void context, do nothing
        my @parms = @_;
        for (@parms) { tr/a-z/A-Z/ }
        return wantarray ? @parms : $parms[0];
    }

Notice how this (unprototyped) function doesn't care whether it was
passed real scalars or arrays.  Perl sees all arugments as one big,
long, flat parameter list in C<@_>.  This is one area where
Perl's simple argument-passing style shines.  The C<upcase()>
function would work perfectly well without changing the C<upcase()>
definition even if we fed it things like this:

    @newlist   = upcase(@list1, @list2);
    @newlist   = upcase( split /:/, $var );

Do not, however, be tempted to do this:

    (@a, @b)   = upcase(@list1, @list2);

Like the flattened incoming parameter list, the return list is also
flattened on return.  So all you have managed to do here is stored
everything in C<@a> and made C<@b> an empty list.  See L<Pass by
Reference> for alternatives.

A subroutine may be called using an explicit C<&> prefix.  The
C<&> is optional in modern Perl, as are parentheses if the
subroutine has been predeclared.  The C<&> is I<not> optional
when just naming the subroutine, such as when it's used as
an argument to defined() or undef().  Nor is it optional when you
want to do an indirect subroutine call with a subroutine name or
reference using the C<&$subref()> or C<&{$subref}()> constructs,
although the C<$subref-E<gt>()> notation solves that problem.
See L<perlref> for more about all that.

Subroutines may be called recursively.  If a subroutine is called
using the C<&> form, the argument list is optional, and if omitted,
no C<@_> array is set up for the subroutine: the C<@_> array at the
time of the call is visible to subroutine instead.  This is an
efficiency mechanism that new users may wish to avoid.

    &foo(1,2,3);        # pass three arguments
    foo(1,2,3);         # the same

    foo();              # pass a null list
    &foo();             # the same

    &foo;               # foo() get current args, like foo(@_) !!
    foo;                # like foo() IFF sub foo predeclared, else "foo"

Not only does the C<&> form make the argument list optional, it also
disables any prototype checking on arguments you do provide.  This
is partly for historical reasons, and partly for having a convenient way
to cheat if you know what you're doing.  See L<Prototypes> below.

Function whose names are in all upper case are reserved to the Perl
core, as are modules whose names are in all lower case.  A
function in all capitals is a loosely-held convention meaning it
will be called indirectly by the run-time system itself, usually
due to a triggered event.  Functions that do special, pre-defined
things include C<BEGIN>, C<END>, C<AUTOLOAD>, and C<DESTROY>--plus
all functions mentioned in L<perltie>.  The 5.005 release adds
C<INIT> to this list.

=head2 Private Variables via my()

Synopsis:

    my $foo;            # declare $foo lexically local
    my (@wid, %get);    # declare list of variables local
    my $foo = "flurp";  # declare $foo lexical, and init it
    my @oof = @bar;     # declare @oof lexical, and init it

The C<my> operator declares the listed variables to be lexically
confined to the enclosing block, conditional (C<if/unless/elsif/else>),
loop (C<for/foreach/while/until/continue>), subroutine, C<eval>,
or C<do/require/use>'d file.  If more than one value is listed, the
list must be placed in parentheses.  All listed elements must be
legal lvalues.  Only alphanumeric identifiers may be lexically
scoped--magical built-in like C<$/> must currently be C<local>ize
with C<local> instead.

Unlike dynamic variables created by the C<local> operator, lexical
variables declared with C<my> are totally hidden from the outside
world, including any called subroutines.  This is true if it's the
same subroutine called from itself or elsewhere--every call gets
its own copy.

This doesn't mean that a C<my> variable declared in a statically
enclosing lexical scope would be invisible.  Only dynamic scopes
are cut off.   For example, the C<bumpx()> function below has access
to the lexical $x variable because both the C<my> and the C<sub>
occurred at the same scope, presumably file scope.

    my $x = 10;
    sub bumpx { $x++ } 

An C<eval()>, however, can see lexical variables of the scope it is
being evaluated in, so long as the names aren't hidden by declarations within
the C<eval()> itself.  See L<perlref>.

The parameter list to my() may be assigned to if desired, which allows you
to initialize your variables.  (If no initializer is given for a
particular variable, it is created with the undefined value.)  Commonly
this is used to name input parameters to a subroutine.  Examples:

    $arg = "fred";        # "global" variable
    $n = cube_root(27);
    print "$arg thinks the root is $n\n";
 fred thinks the root is 3

    sub cube_root {
        my $arg = shift;  # name doesn't matter
        $arg **= 1/3;
        return $arg;
    }

The C<my> is simply a modifier on something you might assign to.  So when
you do assign to variables in its argument list, C<my> doesn't
change whether those variables are viewed as a scalar or an array.  So

    my ($foo) = <STDIN>;                # WRONG?
    my @FOO = <STDIN>;

both supply a list context to the right-hand side, while

    my $foo = <STDIN>;

supplies a scalar context.  But the following declares only one variable:

    my $foo, $bar = 1;                  # WRONG

That has the same effect as

    my $foo;
    $bar = 1;

The declared variable is not introduced (is not visible) until after
the current statement.  Thus,

    my $x = $x;

can be used to initialize a new $x with the value of the old $x, and
the expression

    my $x = 123 and $x == 123

is false unless the old $x happened to have the value C<123>.

Lexical scopes of control structures are not bounded precisely by the
braces that delimit their controlled blocks; control expressions are
part of that scope, too.  Thus in the loop

    while (my $line = <>) {
        $line = lc $line;
    } continue {
        print $line;
    }

the scope of $line extends from its declaration throughout the rest of
the loop construct (including the C<continue> clause), but not beyond
it.  Similarly, in the conditional

    if ((my $answer = <STDIN>) =~ /^yes$/i) {
        user_agrees();
    } elsif ($answer =~ /^no$/i) {
        user_disagrees();
    } else {
        chomp $answer;
        die "'$answer' is neither 'yes' nor 'no'";
    }

the scope of $answer extends from its declaration through the rest
of that conditional, including any C<elsif> and C<else> clauses, 
but not beyond it.

None of the foregoing text applies to C<if/unless> or C<while/until>
modifiers appended to simple statements.  Such modifiers are not
control structures and have no effect on scoping.

The C<foreach> loop defaults to scoping its index variable dynamically
in the manner of C<local>.  However, if the index variable is
prefixed with the keyword C<my>, or if there is already a lexical
by that name in scope, then a new lexical is created instead.  Thus
in the loop

    for my $i (1, 2, 3) {
        some_function();
    }

the scope of $i extends to the end of the loop, but not beyond it,
rendering the value of $i inaccessible within C<some_function()>.

Some users may wish to encourage the use of lexically scoped variables.
As an aid to catching implicit uses to package variables,
which are always global, if you say

    use strict 'vars';

then any variable mentioned from there to the end of the enclosing
block must either refer to a lexical variable, be predeclared via
C<use vars>, or else must be fully qualified with the package name.
A compilation error results otherwise.  An inner block may countermand
this with C<no strict 'vars'>.

A C<my> has both a compile-time and a run-time effect.  At compile
time, the compiler takes notice of it.  The principle usefulness
of this is to quiet C<use strict 'vars'>, but it is also essential
for generation of closures as detailed in L<perlref>.  Actual
initialization is delayed until run time, though, so it gets executed
at the appropriate time, such as each time through a loop, for
example.

Variables declared with C<my> are not part of any package and are therefore
never fully qualified with the package name.  In particular, you're not
allowed to try to make a package variable (or other global) lexical:

    my $pack::var;      # ERROR!  Illegal syntax
    my $_;              # also illegal (currently)

In fact, a dynamic variable (also known as package or global variables)
are still accessible using the fully qualified C<::> notation even while a
lexical of the same name is also visible:

    package main;
    local $x = 10;
    my    $x = 20;
    print "$x and $::x\n";

That will print out C<20> and C<10>.

You may declare C<my> variables at the outermost scope of a file
to hide any such identifiers from the world outside that file.  This
is similar in spirit to C's static variables when they are used at
the file level.  To do this with a subroutine requires the use of
a closure (an anonymous function that accesses enclosing lexicals).
If you want to create a private subroutine that cannot be called
from outside that block, it can declare a lexical variable containing
an anonymous sub reference:

    my $secret_version = '1.001-beta';
    my $secret_sub = sub { print $secret_version };
    &$secret_sub();

As long as the reference is never returned by any function within the
module, no outside module can see the subroutine, because its name is not in
any package's symbol table.  Remember that it's not I<REALLY> called
C<$some_pack::secret_version> or anything; it's just $secret_version,
unqualified and unqualifiable.

This does not work with object methods, however; all object methods
have to be in the symbol table of some package to be found.  See
L<perlref/"Function Templates"> for something of a work-around to
this.

=head2 Persistent Private Variables

Just because a lexical variable is lexically (also called statically)
scoped to its enclosing block, C<eval>, or C<do> FILE, this doesn't mean that
within a function it works like a C static.  It normally works more
like a C auto, but with implicit garbage collection.  

Unlike local variables in C or C++, Perl's lexical variables don't
necessarily get recycled just because their scope has exited.
If something more permanent is still aware of the lexical, it will
stick around.  So long as something else references a lexical, that
lexical won't be freed--which is as it should be.  You wouldn't want
memory being free until you were done using it, or kept around once you
were done.  Automatic garbage collection takes care of this for you.

This means that you can pass back or save away references to lexical
variables, whereas to return a pointer to a C auto is a grave error.
It also gives us a way to simulate C's function statics.  Here's a
mechanism for giving a function private variables with both lexical
scoping and a static lifetime.  If you do want to create something like
C's static variables, just enclose the whole function in an extra block,
and put the static variable outside the function but in the block.

    {
        my $secret_val = 0;
        sub gimme_another {
            return ++$secret_val;
        }
    }
    # $secret_val now becomes unreachable by the outside
    # world, but retains its value between calls to gimme_another

If this function is being sourced in from a separate file
via C<require> or C<use>, then this is probably just fine.  If it's
all in the main program, you'll need to arrange for the C<my>
to be executed early, either by putting the whole block above
your main program, or more likely, placing merely a C<BEGIN>
sub around it to make sure it gets executed before your program
starts to run:

    sub BEGIN {
        my $secret_val = 0;
        sub gimme_another {
            return ++$secret_val;
        }
    }

See L<perlmod/"Package Constructors and Destructors"> about the
special triggered functions, C<BEGIN> and C<INIT>.

If declared at the outermost scope (the file scope), then lexicals
work somewhat like C's file statics.  They are available to all
functions in that same file declared below them, but are inaccessible
from outside that file.  This strategy is sometimes used in modules
to create private variables that the whole module can see.

=head2 Temporary Values via local()

B<WARNING>: In general, you should be using C<my> instead of C<local>, because
it's faster and safer.  Exceptions to this include the global punctuation
variables, filehandles and formats, and direct manipulation of the Perl
symbol table itself.  Format variables often use C<local> though, as do
other variables whose current value must be visible to called
subroutines.

Synopsis:

    local $foo;                 # declare $foo dynamically local
    local (@wid, %get);         # declare list of variables local
    local $foo = "flurp";       # declare $foo dynamic, and init it
    local @oof = @bar;          # declare @oof dynamic, and init it

    local *FH;                  # localize $FH, @FH, %FH, &FH  ...
    local *merlyn = *randal;    # now $merlyn is really $randal, plus
                                #     @merlyn is really @randal, etc
    local *merlyn = 'randal';   # SAME THING: promote 'randal' to *randal
    local *merlyn = \$randal;   # just alias $merlyn, not @merlyn etc

A C<local> modifies its listed variables to be "local" to the
enclosing block, C<eval>, or C<do FILE>--and to I<any subroutine
called from within that block>.  A C<local> just gives temporary
values to global (meaning package) variables.  It does I<not> create
a local variable.  This is known as dynamic scoping.  Lexical scoping
is done with C<my>, which works more like C's auto declarations.

If more than one variable is given to C<local>, they must be placed in
parentheses.  All listed elements must be legal lvalues.  This operator works
by saving the current values of those variables in its argument list on a
hidden stack and restoring them upon exiting the block, subroutine, or
eval.  This means that called subroutines can also reference the local
variable, but not the global one.  The argument list may be assigned to if
desired, which allows you to initialize your local variables.  (If no
initializer is given for a particular variable, it is created with an
undefined value.)  Commonly this is used to name the parameters to a
subroutine.  Examples:

    for $i ( 0 .. 9 ) {
        $digits{$i} = $i;
    }
    # assume this function uses global %digits hash
    parse_num();

    # now temporarily add to %digits hash
    if ($base12) {
        # (NOTE: not claiming this is efficient!)
        local %digits  = (%digits, 't' => 10, 'e' => 11);
        parse_num();  # parse_num gets this new %digits!
    }
    # old %digits restored here

Because C<local> is a run-time operator, it gets executed each time
through a loop.  In releases of Perl previous to 5.0, this used more stack
storage each time until the loop was exited.  Perl now reclaims the space
each time through, but it's still more efficient to declare your variables
outside the loop.

A C<local> is simply a modifier on an lvalue expression.  When you assign to
a C<local>ized variable, the C<local> doesn't change whether its list is viewed
as a scalar or an array.  So

    local($foo) = <STDIN>;
    local @FOO = <STDIN>;

both supply a list context to the right-hand side, while

    local $foo = <STDIN>;

supplies a scalar context.

A note about C<local()> and composite types is in order.  Something
like C<local(%foo)> works by temporarily placing a brand new hash in
the symbol table.  The old hash is left alone, but is hidden "behind"
the new one.

This means the old variable is completely invisible via the symbol
table (i.e. the hash entry in the C<*foo> typeglob) for the duration
of the dynamic scope within which the C<local()> was seen.  This
has the effect of allowing one to temporarily occlude any magic on
composite types.  For instance, this will briefly alter a tied
hash to some other implementation:

    tie %ahash, 'APackage';
    [...]
    {
       local %ahash;
       tie %ahash, 'BPackage';
       [..called code will see %ahash tied to 'BPackage'..]
       {
          local %ahash;
          [..%ahash is a normal (untied) hash here..]
       }
    }
    [..%ahash back to its initial tied self again..]

As another example, a custom implementation of C<%ENV> might look
like this:

    {
        local %ENV;
        tie %ENV, 'MyOwnEnv';
        [..do your own fancy %ENV manipulation here..]
    }
    [..normal %ENV behavior here..]

It's also worth taking a moment to explain what happens when you
C<local>ize a member of a composite type (i.e. an array or hash element).
In this case, the element is C<local>ized I<by name>. This means that
when the scope of the C<local()> ends, the saved value will be
restored to the hash element whose key was named in the C<local()>, or
the array element whose index was named in the C<local()>.  If that
element was deleted while the C<local()> was in effect (e.g. by a
C<delete()> from a hash or a C<shift()> of an array), it will spring
back into existence, possibly extending an array and filling in the
skipped elements with C<undef>.  For instance, if you say

    %hash = ( 'This' => 'is', 'a' => 'test' );
    @ary  = ( 0..5 );
    {
         local($ary[5]) = 6;
         local($hash{'a'}) = 'drill';
         while (my $e = pop(@ary)) {
             print "$e . . .\n";
             last unless $e > 3;
         }
         if (@ary) {
             $hash{'only a'} = 'test';
             delete $hash{'a'};
         }
    }
    print join(' ', map { "$_ $hash{$_}" } sort keys %hash),".\n";
    print "The array has ",scalar(@ary)," elements: ",
          join(', ', map { defined $_ ? $_ : 'undef' } @ary),"\n";

Perl will print

    6 . . .
    4 . . .
    3 . . .
    This is a test only a test.
    The array has 6 elements: 0, 1, 2, undef, undef, 5

The behavior of local() on non-existent members of composite
types is subject to change in future.

=head2 Passing Symbol Table Entries (typeglobs)

B<WARNING>: The mechanism described in this section was originally
the only way to simulate pass-by-reference in older versions of
Perl.  While it still works fine in modern versions, the new reference
mechanism is generally easier to work with.  See below.

Sometimes you don't want to pass the value of an array to a subroutine
but rather the name of it, so that the subroutine can modify the global
copy of it rather than working with a local copy.  In perl you can
refer to all objects of a particular name by prefixing the name
with a star: C<*foo>.  This is often known as a "typeglob", because the
star on the front can be thought of as a wildcard match for all the
funny prefix characters on variables and subroutines and such.

When evaluated, the typeglob produces a scalar value that represents
all the objects of that name, including any filehandle, format, or
subroutine.  When assigned to, it causes the name mentioned to refer to
whatever C<*> value was assigned to it.  Example:

    sub doubleary {
        local(*someary) = @_;
        foreach $elem (@someary) {
            $elem *= 2;
        }
    }
    doubleary(*foo);
    doubleary(*bar);

Scalars are already passed by reference, so you can modify
scalar arguments without using this mechanism by referring explicitly
to C<$_[0]> etc.  You can modify all the elements of an array by passing
all the elements as scalars, but you have to use the C<*> mechanism (or
the equivalent reference mechanism) to C<push>, C<pop>, or change the size of
an array.  It will certainly be faster to pass the typeglob (or reference).

Even if you don't want to modify an array, this mechanism is useful for
passing multiple arrays in a single LIST, because normally the LIST
mechanism will merge all the array values so that you can't extract out
the individual arrays.  For more on typeglobs, see
L<perldata/"Typeglobs and Filehandles">.

=head2 When to Still Use local()

Despite the existence of C<my>, there are still three places where the
C<local> operator still shines.  In fact, in these three places, you
I<must> use C<local> instead of C<my>.

=over

=item 1. You need to give a global variable a temporary value, especially $_.

The global variables, like C<@ARGV> or the punctuation variables, must be 
C<local>ized with C<local()>.  This block reads in F</etc/motd>, and splits
it up into chunks separated by lines of equal signs, which are placed
in C<@Fields>.

    {
        local @ARGV = ("/etc/motd");
        local $/ = undef;
        local $_ = <>;  
        @Fields = split /^\s*=+\s*$/;
    } 

It particular, it's important to C<local>ize $_ in any routine that assigns
to it.  Look out for implicit assignments in C<while> conditionals.

=item 2. You need to create a local file or directory handle or a local function.

A function that needs a filehandle of its own must use C<local()> uses
C<local()> on complete typeglob.   This can be used to create new symbol
table entries:

    sub ioqueue {
        local  (*READER, *WRITER);    # not my!
        pipe    (READER,  WRITER);    or die "pipe: $!";
        return (*READER, *WRITER);
    }
    ($head, $tail) = ioqueue();

See the Symbol module for a way to create anonymous symbol table
entries.

Because assignment of a reference to a typeglob creates an alias, this
can be used to create what is effectively a local function, or at least,
a local alias.

    {
        local *grow = \&shrink; # only until this block exists
        grow();                 # really calls shrink()
        move();                 # if move() grow()s, it shrink()s too
    }
    grow();                     # get the real grow() again

See L<perlref/"Function Templates"> for more about manipulating
functions by name in this way.

=item 3. You want to temporarily change just one element of an array or hash.

You can C<local>ize just one element of an aggregate.  Usually this
is done on dynamics:

    {
        local $SIG{INT} = 'IGNORE';
        funct();                            # uninterruptible
    } 
    # interruptibility automatically restored here

But it also works on lexically declared aggregates.  Prior to 5.005,
this operation could on occasion misbehave.

=back

=head2 Pass by Reference

If you want to pass more than one array or hash into a function--or
return them from it--and have them maintain their integrity, then
you're going to have to use an explicit pass-by-reference.  Before you
do that, you need to understand references as detailed in L<perlref>.
This section may not make much sense to you otherwise.

Here are a few simple examples.  First, let's pass in several arrays
to a function and have it C<pop> all of then, returning a new list
of all their former last elements:

    @tailings = popmany ( \@a, \@b, \@c, \@d );

    sub popmany {
        my $aref;
        my @retlist = ();
        foreach $aref ( @_ ) {
            push @retlist, pop @$aref;
        }
        return @retlist;
    }

Here's how you might write a function that returns a
list of keys occurring in all the hashes passed to it:

    @common = inter( \%foo, \%bar, \%joe );
    sub inter {
        my ($k, $href, %seen); # locals
        foreach $href (@_) {
            while ( $k = each %$href ) {
                $seen{$k}++;
            }
        }
        return grep { $seen{$_} == @_ } keys %seen;
    }

So far, we're using just the normal list return mechanism.
What happens if you want to pass or return a hash?  Well,
if you're using only one of them, or you don't mind them
concatenating, then the normal calling convention is ok, although
a little expensive.

Where people get into trouble is here:

    (@a, @b) = func(@c, @d);
or
    (%a, %b) = func(%c, %d);

That syntax simply won't work.  It sets just C<@a> or C<%a> and
clears the C<@b> or C<%b>.  Plus the function didn't get passed
into two separate arrays or hashes: it got one long list in C<@_>,
as always.

If you can arrange for everyone to deal with this through references, it's
cleaner code, although not so nice to look at.  Here's a function that
takes two array references as arguments, returning the two array elements
in order of how many elements they have in them:

    ($aref, $bref) = func(\@c, \@d);
    print "@$aref has more than @$bref\n";
    sub func {
        my ($cref, $dref) = @_;
        if (@$cref > @$dref) {
            return ($cref, $dref);
        } else {
            return ($dref, $cref);
        }
    }

It turns out that you can actually do this also:

    (*a, *b) = func(\@c, \@d);
    print "@a has more than @b\n";
    sub func {
        local (*c, *d) = @_;
        if (@c > @d) {
            return (\@c, \@d);
        } else {
            return (\@d, \@c);
        }
    }

Here we're using the typeglobs to do symbol table aliasing.  It's
a tad subtle, though, and also won't work if you're using C<my>
variables, because only globals (even in disguised as C<local>s)
are in the symbol table.

If you're passing around filehandles, you could usually just use the bare
typeglob, like C<*STDOUT>, but typeglobs references work, too.
For example:

    splutter(\*STDOUT);
    sub splutter {
        my $fh = shift;
        print $fh "her um well a hmmm\n";
    }

    $rec = get_rec(\*STDIN);
    sub get_rec {
        my $fh = shift;
        return scalar <$fh>;
    }

If you're planning on generating new filehandles, you could do this.
Notice to pass back just the bare *FH, not its reference.

    sub openit {
        my $path = shift;
        local *FH;
        return open (FH, $path) ? *FH : undef;
    }

=head2 Prototypes

Perl supports a very limited kind of compile-time argument checking
using function prototyping.  If you declare

    sub mypush (\@@)

then C<mypush()> takes arguments exactly like C<push()> does.  The
function declaration must be visible at compile time.  The prototype
affects only interpretation of new-style calls to the function,
where new-style is defined as not using the C<&> character.  In
other words, if you call it like a built-in function, then it behaves
like a built-in function.  If you call it like an old-fashioned
subroutine, then it behaves like an old-fashioned subroutine.  It
naturally falls out from this rule that prototypes have no influence
on subroutine references like C<\&foo> or on indirect subroutine
calls like C<&{$subref}> or C<$subref-E<gt>()>.

Method calls are not influenced by prototypes either, because the
function to be called is indeterminate at compile time, since
the exact code called depends on inheritance.

Because the intent of this feature is primarily to let you define
subroutines that work like built-in functions, here are prototypes
for some other functions that parse almost exactly like the
corresponding built-in.

    Declared as                 Called as

    sub mylink ($$)          mylink $old, $new
    sub myvec ($$$)          myvec $var, $offset, 1
    sub myindex ($$;$)       myindex &getstring, "substr"
    sub mysyswrite ($$$;$)   mysyswrite $buf, 0, length($buf) - $off, $off
    sub myreverse (@)        myreverse $a, $b, $c
    sub myjoin ($@)          myjoin ":", $a, $b, $c
    sub mypop (\@)           mypop @array
    sub mysplice (\@$$@)     mysplice @array, @array, 0, @pushme
    sub mykeys (\%)          mykeys %{$hashref}
    sub myopen (*;$)         myopen HANDLE, $name
    sub mypipe (**)          mypipe READHANDLE, WRITEHANDLE
    sub mygrep (&@)          mygrep { /foo/ } $a, $b, $c
    sub myrand ($)           myrand 42
    sub mytime ()            mytime

Any backslashed prototype character represents an actual argument
that absolutely must start with that character.  The value passed
as part of C<@_> will be a reference to the actual argument given
in the subroutine call, obtained by applying C<\> to that argument.

Unbackslashed prototype characters have special meanings.  Any
unbackslashed C<@> or C<%> eats all remaining arguments, and forces
list context.  An argument represented by C<$> forces scalar context.  An
C<&> requires an anonymous subroutine, which, if passed as the first
argument, does not require the C<sub> keyword or a subsequent comma.  A
C<*> allows the subroutine to accept a bareword, constant, scalar expression,
typeglob, or a reference to a typeglob in that slot.  The value will be
available to the subroutine either as a simple scalar, or (in the latter
two cases) as a reference to the typeglob.

A semicolon separates mandatory arguments from optional arguments.
It is redundant before C<@> or C<%>, which gobble up everything else.

Note how the last three examples in the table above are treated
specially by the parser.  C<mygrep()> is parsed as a true list
operator, C<myrand()> is parsed as a true unary operator with unary
precedence the same as C<rand()>, and C<mytime()> is truly without
arguments, just like C<time()>.  That is, if you say

    mytime +2;

you'll get C<mytime() + 2>, not C<mytime(2)>, which is how it would be parsed
without a prototype.

The interesting thing about C<&> is that you can generate new syntax with it,
provided it's in the initial position:

    sub try (&@) {
        my($try,$catch) = @_;
        eval { &$try };
        if ($@) {
            local $_ = $@;
            &$catch;
        }
    }
    sub catch (&) { $_[0] }

    try {
        die "phooey";
    } catch {
        /phooey/ and print "unphooey\n";
    };

That prints C<"unphooey">.  (Yes, there are still unresolved
issues having to do with visibility of C<@_>.  I'm ignoring that
question for the moment.  (But note that if we make C<@_> lexically
scoped, those anonymous subroutines can act like closures... (Gee,
is this sounding a little Lispish?  (Never mind.))))

And here's a reimplementation of the Perl C<grep> operator:

    sub mygrep (&@) {
        my $code = shift;
        my @result;
        foreach $_ (@_) {
            push(@result, $_) if &$code;
        }
        @result;
    }

Some folks would prefer full alphanumeric prototypes.  Alphanumerics have
been intentionally left out of prototypes for the express purpose of
someday in the future adding named, formal parameters.  The current
mechanism's main goal is to let module writers provide better diagnostics
for module users.  Larry feels the notation quite understandable to Perl
programmers, and that it will not intrude greatly upon the meat of the
module, nor make it harder to read.  The line noise is visually
encapsulated into a small pill that's easy to swallow.

It's probably best to prototype new functions, not retrofit prototyping
into older ones.  That's because you must be especially careful about
silent impositions of differing list versus scalar contexts.  For example,
if you decide that a function should take just one parameter, like this:

    sub func ($) {
        my $n = shift;
        print "you gave me $n\n";
    }

and someone has been calling it with an array or expression
returning a list:

    func(@foo);
    func( split /:/ );

Then you've just supplied an automatic C<scalar> in front of their
argument, which can be more than a bit surprising.  The old C<@foo>
which used to hold one thing doesn't get passed in.  Instead,
C<func()> now gets passed in a C<1>; that is, the number of elements
in C<@foo>.  And the C<split> gets called in scalar context so it
starts scribbling on your C<@_> parameter list.  Ouch!

This is all very powerful, of course, and should be used only in moderation
to make the world a better place.

=head2 Constant Functions

Functions with a prototype of C<()> are potential candidates for
inlining.  If the result after optimization and constant folding
is either a constant or a lexically-scoped scalar which has no other
references, then it will be used in place of function calls made
without C<&>.  Calls made using C<&> are never inlined.  (See
F<constant.pm> for an easy way to declare most constants.)

The following functions would all be inlined:

    sub pi ()           { 3.14159 }             # Not exact, but close.
    sub PI ()           { 4 * atan2 1, 1 }      # As good as it gets,
                                                # and it's inlined, too!
    sub ST_DEV ()       { 0 }
    sub ST_INO ()       { 1 }

    sub FLAG_FOO ()     { 1 << 8 }
    sub FLAG_BAR ()     { 1 << 9 }
    sub FLAG_MASK ()    { FLAG_FOO | FLAG_BAR }

    sub OPT_BAZ ()      { not (0x1B58 & FLAG_MASK) }
    sub BAZ_VAL () {
        if (OPT_BAZ) {
            return 23;
        }
        else {
            return 42;
        }
    }

    sub N () { int(BAZ_VAL) / 3 }
    BEGIN {
        my $prod = 1;
        for (1..N) { $prod *= $_ }
        sub N_FACTORIAL () { $prod }
    }

If you redefine a subroutine that was eligible for inlining, you'll get
a mandatory warning.  (You can use this warning to tell whether or not a
particular subroutine is considered constant.)  The warning is
considered severe enough not to be optional because previously compiled
invocations of the function will still be using the old value of the
function.  If you need to be able to redefine the subroutine, you need to
ensure that it isn't inlined, either by dropping the C<()> prototype
(which changes calling semantics, so beware) or by thwarting the
inlining mechanism in some other way, such as

    sub not_inlined () {
        23 if $];
    }

=head2 Overriding Built-in Functions

Many built-in functions may be overridden, though this should be tried
only occasionally and for good reason.  Typically this might be
done by a package attempting to emulate missing built-in functionality
on a non-Unix system.

Overriding may be done only by importing the name from a
module--ordinary predeclaration isn't good enough.  However, the
C<use subs> pragma lets you, in effect, predeclare subs
via the import syntax, and these names may then override built-in ones:

    use subs 'chdir', 'chroot', 'chmod', 'chown';
    chdir $somewhere;
    sub chdir { ... }

To unambiguously refer to the built-in form, precede the
built-in name with the special package qualifier C<CORE::>.  For example,
saying C<CORE::open()> always refers to the built-in C<open()>, even
if the current package has imported some other subroutine called
C<&open()> from elsewhere.  Even though it looks like a regular
function calls, it isn't: you can't take a reference to it, such as
the incorrect C<\&CORE::open> might appear to produce.

Library modules should not in general export built-in names like C<open>
or C<chdir> as part of their default C<@EXPORT> list, because these may
sneak into someone else's namespace and change the semantics unexpectedly.
Instead, if the module adds that name to C<@EXPORT_OK>, then it's
possible for a user to import the name explicitly, but not implicitly.
That is, they could say

    use Module 'open';

and it would import the C<open> override.  But if they said

    use Module;

they would get the default imports without overrides.

The foregoing mechanism for overriding built-in is restricted, quite
deliberately, to the package that requests the import.  There is a second
method that is sometimes applicable when you wish to override a built-in
everywhere, without regard to namespace boundaries.  This is achieved by
importing a sub into the special namespace C<CORE::GLOBAL::>.  Here is an
example that quite brazenly replaces the C<glob> operator with something
that understands regular expressions.

    package REGlob;
    require Exporter;
    @ISA = 'Exporter';
    @EXPORT_OK = 'glob';

    sub import {
        my $pkg = shift;
        return unless @_;
        my $sym = shift;
        my $where = ($sym =~ s/^GLOBAL_// ? 'CORE::GLOBAL' : caller(0));
        $pkg->export($where, $sym, @_);
    }

    sub glob {
        my $pat = shift;
        my @got;
        local *D;
        if (opendir D, '.') { 
            @got = grep /$pat/, readdir D; 
            closedir D;   
        }
        return @got;
    }
    1;

And here's how it could be (ab)used:

    #use REGlob 'GLOBAL_glob';      # override glob() in ALL namespaces
    package Foo;
    use REGlob 'glob';              # override glob() in Foo:: only
    print for <^[a-z_]+\.pm\$>;     # show all pragmatic modules

The initial comment shows a contrived, even dangerous example.
By overriding C<glob> globally, you would be forcing the new (and
subversive) behavior for the C<glob> operator for I<every> namespace,
without the complete cognizance or cooperation of the modules that own
those namespaces.  Naturally, this should be done with extreme caution--if
it must be done at all.

The C<REGlob> example above does not implement all the support needed to
cleanly override perl's C<glob> operator.  The built-in C<glob> has
different behaviors depending on whether it appears in a scalar or list
context, but our C<REGlob> doesn't.  Indeed, many perl built-in have such
context sensitive behaviors, and these must be adequately supported by
a properly written override.  For a fully functional example of overriding
C<glob>, study the implementation of C<File::DosGlob> in the standard
library.

=head2 Autoloading

If you call a subroutine that is undefined, you would ordinarily
get an immediate, fatal error complaining that the subroutine doesn't
exist.  (Likewise for subroutines being used as methods, when the
method doesn't exist in any base class of the class's package.)
However, if an C<AUTOLOAD> subroutine is defined in the package or
packages used to locate the original subroutine, then that
C<AUTOLOAD> subroutine is called with the arguments that would have
been passed to the original subroutine.  The fully qualified name
of the original subroutine magically appears in the global $AUTOLOAD
variable of the same package as the C<AUTOLOAD> routine.  The name
is not passed as an ordinary argument because, er, well, just
because, that's why...

Many C<AUTOLOAD> routines load in a definition for the requested
subroutine using eval(), then execute that subroutine using a special
form of goto() that erases the stack frame of the C<AUTOLOAD> routine
without a trace.  (See the source to the standard module documented
in L<AutoLoader>, for example.)  But an C<AUTOLOAD> routine can
also just emulate the routine and never define it.   For example,
let's pretend that a function that wasn't defined should just invoke
C<system> with those arguments.  All you'd do is:

    sub AUTOLOAD {
        my $program = $AUTOLOAD;
        $program =~ s/.*:://;
        system($program, @_);
    }
    date();
    who('am', 'i');
    ls('-l');

In fact, if you predeclare functions you want to call that way, you don't
even need parentheses:

    use subs qw(date who ls);
    date;
    who "am", "i";
    ls -l;

A more complete example of this is the standard Shell module, which
can treat undefined subroutine calls as calls to external programs.

Mechanisms are available to help modules writers split their modules
into autoloadable files.  See the standard AutoLoader module
described in L<AutoLoader> and in L<AutoSplit>, the standard
SelfLoader modules in L<SelfLoader>, and the document on adding C
functions to Perl code in L<perlxs>.

=head1 SEE ALSO

See L<perlref/"Function Templates"> for more about references and closures.
See L<perlxs> if you'd like to learn about calling C subroutines from Perl.  
See L<perlembed> if you'd like to learn about calling PErl subroutines from C.  
See L<perlmod> to learn about bundling up your functions in separate files.
See L<perlmodlib> to learn what library modules come standard on your system.
See L<perltoot> to learn how to make object method calls.