package fields;

=head1 NAME

fields - compile-time class fields

=head1 SYNOPSIS

    {
        package Foo;
        use fields qw(foo bar _Foo_private);
	sub new {
	    my Foo $self = shift;
	    unless (ref $self) {
		$self = fields::new($self);
		$self->{_Foo_private} = "this is Foo's secret";
	    }
	    $self->{foo} = 10;
	    $self->{bar} = 20;
	    return $self;
	}
    }

    my Foo $var = Foo::->new;
    $var->{foo} = 42;

    # this will generate a compile-time error
    $var->{zap} = 42;

    # subclassing
    {
        package Bar;
        use base 'Foo';
        use fields qw(baz _Bar_private);	# not shared with Foo
	sub new {
	    my $class = shift;
	    my $self = fields::new($class);
	    $self->SUPER::new();		# init base fields
	    $self->{baz} = 10;			# init own fields
	    $self->{_Bar_private} = "this is Bar's secret";
	    return $self;
	}
    }

=head1 DESCRIPTION

The C<fields> pragma enables compile-time verified class fields.

NOTE: The current implementation keeps the declared fields in the %FIELDS
hash of the calling package, but this may change in future versions.
Do B<not> update the %FIELDS hash directly, because it must be created
at compile-time for it to be fully useful, as is done by this pragma.

If a typed lexical variable holding a reference is used to access a
hash element and a package with the same name as the type has declared
class fields using this pragma, then the operation is turned into an
array access at compile time.

The related C<base> pragma will combine fields from base classes and any
fields declared using the C<fields> pragma.  This enables field
inheritance to work properly.

Field names that start with an underscore character are made private to
the class and are not visible to subclasses.  Inherited fields can be
overridden but will generate a warning if used together with the C<-w>
switch.

The effect of all this is that you can have objects with named fields
which are as compact and as fast arrays to access.  This only works
as long as the objects are accessed through properly typed variables.
If the objects are not typed, access is only checked at run time.

The following functions are supported:

=over 8

=item new

fields::new() creates and blesses a pseudo-hash comprised of the fields
declared using the C<fields> pragma into the specified class.
This makes it possible to write a constructor like this:

    package Critter::Sounds;
    use fields qw(cat dog bird);

    sub new {
	my Critter::Sounds $self = shift;
	$self = fields::new($self) unless ref $self;
	$self->{cat} = 'meow';				# scalar element
	@$self{'dog','bird'} = ('bark','tweet');	# slice
	return $self;
    }

=item phash

fields::phash() can be used to create and initialize a plain (unblessed)
pseudo-hash.  This function should always be used instead of creating
pseudo-hashes directly.

If the first argument is a reference to an array, the pseudo-hash will
be created with keys from that array.  If a second argument is supplied,
it must also be a reference to an array whose elements will be used as
the values.  If the second array contains less elements than the first,
the trailing elements of the pseudo-hash will not be initialized.
This makes it particularly useful for creating a pseudo-hash from
subroutine arguments:

    sub dogtag {
	my $tag = fields::phash([qw(name rank ser_num)], [@_]);
    }

fields::phash() also accepts a list of key-value pairs that will
be used to construct the pseudo hash.  Examples:

    my $tag = fields::phash(name => "Joe",
			    rank => "captain",
			    ser_num => 42);

    my $pseudohash = fields::phash(%args);

=back

=head1 SEE ALSO

L<base>,
L<perlref/Pseudo-hashes: Using an array as a hash>

=cut

use 5.005_64;
use strict;
no strict 'refs';
use warnings::register;
our(%attr, $VERSION);

$VERSION = "1.01";

# some constants
sub _PUBLIC    () { 1 }
sub _PRIVATE   () { 2 }

# The %attr hash holds the attributes of the currently assigned fields
# per class.  The hash is indexed by class names and the hash value is
# an array reference.  The first element in the array is the lowest field
# number not belonging to a base class.  The remaining elements' indices
# are the field numbers.  The values are integer bit masks, or undef
# in the case of base class private fields (which occupy a slot but are
# otherwise irrelevant to the class).

sub import {
    my $class = shift;
    return unless @_;
    my $package = caller(0);
    # avoid possible typo warnings
    %{"$package\::FIELDS"} = () unless %{"$package\::FIELDS"};
    my $fields = \%{"$package\::FIELDS"};
    my $fattr = ($attr{$package} ||= [1]);
    my $next = @$fattr;

    if ($next > $fattr->[0]
	and ($fields->{$_[0]} || 0) >= $fattr->[0])
    {
	# There are already fields not belonging to base classes.
	# Looks like a possible module reload...
	$next = $fattr->[0];
    }
    foreach my $f (@_) {
	my $fno = $fields->{$f};

	# Allow the module to be reloaded so long as field positions
	# have not changed.
	if ($fno and $fno != $next) {
	    require Carp;
            if ($fno < $fattr->[0]) {
                warnings::warn("Hides field '$f' in base class") 
		    if warnings::enabled();
            } else {
                Carp::croak("Field name '$f' already in use");
            }
	}
	$fields->{$f} = $next;
        $fattr->[$next] = ($f =~ /^_/) ? _PRIVATE : _PUBLIC;
	$next += 1;
    }
    if (@$fattr > $next) {
	# Well, we gave them the benefit of the doubt by guessing the
	# module was reloaded, but they appear to be declaring fields
	# in more than one place.  We can't be sure (without some extra
	# bookkeeping) that the rest of the fields will be declared or
	# have the same positions, so punt.
	require Carp;
	Carp::croak ("Reloaded module must declare all fields at once");
    }
}

sub inherit  { # called by base.pm when $base_fields is nonempty
    my($derived, $base) = @_;
    my $base_attr = $attr{$base};
    my $derived_attr = $attr{$derived} ||= [];
    # avoid possible typo warnings
    %{"$base\::FIELDS"} = () unless %{"$base\::FIELDS"};
    %{"$derived\::FIELDS"} = () unless %{"$derived\::FIELDS"};
    my $base_fields    = \%{"$base\::FIELDS"};
    my $derived_fields = \%{"$derived\::FIELDS"};

    $derived_attr->[0] = $base_attr ? scalar(@$base_attr) : 1;
    while (my($k,$v) = each %$base_fields) {
	my($fno);
	if ($fno = $derived_fields->{$k} and $fno != $v) {
	    require Carp;
	    Carp::croak ("Inherited %FIELDS can't override existing %FIELDS");
	}
	if ($base_attr->[$v] & _PRIVATE) {
	    $derived_attr->[$v] = undef;
	} else {
	    $derived_attr->[$v] = $base_attr->[$v];
	    $derived_fields->{$k} = $v;
	}
     }
}

sub _dump  # sometimes useful for debugging
{
    for my $pkg (sort keys %attr) {
	print "\n$pkg";
	if (@{"$pkg\::ISA"}) {
	    print " (", join(", ", @{"$pkg\::ISA"}), ")";
	}
	print "\n";
	my $fields = \%{"$pkg\::FIELDS"};
	for my $f (sort {$fields->{$a} <=> $fields->{$b}} keys %$fields) {
	    my $no = $fields->{$f};
	    print "   $no: $f";
	    my $fattr = $attr{$pkg}[$no];
	    if (defined $fattr) {
		my @a;
		push(@a, "public")    if $fattr & _PUBLIC;
		push(@a, "private")   if $fattr & _PRIVATE;
		push(@a, "inherited") if $no < $attr{$pkg}[0];
		print "\t(", join(", ", @a), ")";
	    }
	    print "\n";
	}
    }
}

sub new {
    my $class = shift;
    $class = ref $class if ref $class;
    return bless [\%{$class . "::FIELDS"}], $class;
}

sub phash {
    my $h;
    my $v;
    if (@_) {
	if (ref $_[0] eq 'ARRAY') {
	    my $a = shift;
	    @$h{@$a} = 1 .. @$a;
	    if (@_) {
		$v = shift;
		unless (! @_ and ref $v eq 'ARRAY') {
		    require Carp;
		    Carp::croak ("Expected at most two array refs\n");
		}
	    }
	}
	else {
	    if (@_ % 2) {
		require Carp;
		Carp::croak ("Odd number of elements initializing pseudo-hash\n");
	    }
	    my $i = 0;
	    @$h{grep ++$i % 2, @_} = 1 .. @_ / 2;
	    $i = 0;
	    $v = [grep $i++ % 2, @_];
	}
    }
    else {
	$h = {};
	$v = [];
    }
    [ $h, @$v ];
}

1;