package Data::DPath::Context;
our $AUTHORITY = 'cpan:SCHWIGON';
# ABSTRACT: Abstraction for a current context that enables incremental searches
$Data::DPath::Context::VERSION = '0.58';
use strict;
use warnings;

use Data::Dumper;
use aliased 'Data::DPath::Point';
use aliased 'Data::DPath::Attrs';
use Scalar::Util 'reftype';
use Data::DPath::Filters;
use Iterator::Util;
use List::Util 1.45 'min', 'uniq';
#use Sys::CPU;
use POSIX ();
use Safe;

# run filter expressions in own Safe.pm compartment
our $COMPARTMENT;
our $THREADCOUNT;

BEGIN {
        #$THREADCOUNT = $Data::DPath::PARALLELIZE ? Sys::CPU::cpu_count : 1;
        #print "THREADCOUNT: $THREADCOUNT\n";
        package
          Data::DPath::Filters;

        $COMPARTMENT = Safe->new;
        $COMPARTMENT->permit(qw":base_core");
        $COMPARTMENT->reval( 'no warnings;' ); # just so warnings is loaded
        if ($] >= 5.010) {
            $COMPARTMENT->deny(qw":load");
        } else {
            $COMPARTMENT->deny(qw"require dofile caller");
        }
        # map DPath filter functions into new namespace
        $COMPARTMENT->share(qw(affe
                               idx
                               size
                               key
                               value
                               isa
                               reftype
                               is_reftype
                             ));
}

# print "use $]\n" if $] >= 5.010; # allow new-school Perl inside filter expressions
# eval "use $]" if $] >= 5.010; # allow new-school Perl inside filter expressions

use Class::XSAccessor::Array
    chained     => 1,
    constructor => 'new',
    accessors   => {
                    current_points  => 0,
                    give_references => 1,
                   };

use constant { HASH             => 'HASH',
               ARRAY            => 'ARRAY',
               SCALAR           => 'SCALAR',
               ROOT             => 'ROOT',
               ANYWHERE         => 'ANYWHERE',
               KEY              => 'KEY',
               ANYSTEP          => 'ANYSTEP',
               NOSTEP           => 'NOSTEP',
               PARENT           => 'PARENT',
               ANCESTOR         => 'ANCESTOR',
               ANCESTOR_OR_SELF => 'ANCESTOR_OR_SELF',
           };

sub _splice_threads {
    my ($cargo) = @_;

    my $nr_cargo    = @$cargo;

    return [[]] unless $nr_cargo;

    my $threadcount = $THREADCOUNT || 1;
    my $blocksize   = POSIX::ceil ($nr_cargo / $threadcount);

    my @result = map {
        my $first =  $_ * $blocksize;
        my $last  = min(($_+1) * $blocksize - 1, $nr_cargo-1);
        ($first <= $last) ? [ @$cargo[$first .. $last]] : ();
    } 0 .. $threadcount-1;

    return \@result;
}

# only finds "inner" values; if you need the outer start value
# then just wrap it into one more level of array brackets.
sub _any
{
        my ($out, $in, $lookahead_key) = @_;

        no warnings 'uninitialized';

        $in = defined $in ? $in : [];
        return $out unless @$in;

        my @newin;
        my $tmp_ref;
        my $tmp_deref;
        my $tmp_reftype;

        foreach my $point (@$in) {
                my @values;
                next unless defined $point;
                my $ref = $point->ref;

                # speed optimization: first try faster ref, then reftype
                if (ref($$ref) eq HASH or reftype($$ref) eq HASH) {
                        push @$out,
                            map {
                                 my $newpoint = Point->new
                                                     ->ref(\($$ref->{$_}))
                                                     ->parent($point)
                                                     ->attrs(Attrs->new(key => $_));
                                 push @newin, $newpoint; # remember added points
                                 $newpoint;
                            }
                            grep {
                                    # speed optimization: only consider a key if lookahead looks promising
                                    not defined $lookahead_key
                                    or $_ eq $lookahead_key
                                    or ($tmp_ref = ref($tmp_deref =$$ref->{$_}))         eq HASH
                                    or $tmp_ref                                   eq ARRAY
                                    or ($tmp_reftype = reftype($tmp_deref)) eq HASH
                                    or $tmp_reftype                               eq ARRAY
                                    # or HASH_or_ARRAY(\($$ref->{$_}))
                            }
                                keys %{$$ref};
                }
                elsif (ref($$ref) eq ARRAY or reftype($$ref) eq ARRAY) {
                        my $idx = 0;
                        push @$out,
                            map {
                                 my $newpoint = Point->new
                                                     ->ref($_->{val_ref})
                                                     ->parent($point)
                                                     ->attrs(Attrs->new(idx => $idx++));
                                 push @newin, $newpoint; # remember added points
                                 $newpoint;
                            }
                            map { { val_ref => \$_ } } @{$$ref}
                }
                else {
                        next
                }
        }
        return _any ($out, \@newin, $lookahead_key);
}

sub _all {
        my ($self) = @_;

        return @{$self->_all_ref};
}

sub _all_ref {
        my ($self) = @_;

        no strict 'refs';
        no warnings 'uninitialized';

        return [
                map { $self->give_references ? $_ : $$_ }
                uniq
                map { defined $_ ? $_->ref : () }
                @{$self->current_points}
            ];
}

# filter current results by array index
sub _filter_points_index {
        my ($self, $index, $points) = @_;

        return $points ? [$points->[$index]] : [];
}

# filter current results by condition
sub _filter_points_eval
{
        my ($self, $filter, $points) = @_;

        return [] unless @$points;
        return $points unless defined $filter;

        my $new_points;
        my $res;
        {
                package Data::DPath::Filters;

                local our $idx = 0;
                $new_points = [
                               grep {
                                       local our $p = $_;
                                       local $_;
                                       my $pref = $p->ref;
                                       if ( defined $pref ) {
                                               $_ = $$pref;
                                               if ($Data::DPath::USE_SAFE) {
                                                       # 'uninitialized' values are the norm
                                                       # but "no warnings 'uninitialized'" does
                                                       # not work in this restrictive Safe.pm config, so
                                                       # we deactivate warnings completely by localizing $^W.
                                                       # on later Perls, ^W doesn't do the whole trick, so explicitly turn
                                                       # all warnings off.  need to do this in a BEGIN, as some warnings
                                                       # are compile time only.
                                                       $res = $COMPARTMENT->reval('BEGIN{ warnings->unimport}; local $^W;'.$filter);
                                               } else {
                                                       # 'uninitialized' values are the norm
                                                       no warnings 'uninitialized';
                                                       no if $] >= 5.018, warnings => 'experimental::smartmatch';
                                                       $res = eval($filter);
                                               }
                                               print STDERR ($@, "\n") if $@;
                                       } else {
                                               $res = 0;
                                       }
                                       $idx++;
                                       $res;
                               } @$points
                              ];
        }
        return $new_points;
}

sub _filter_points {
        my ($self, $step, $points) = @_;

        no strict 'refs';
        no warnings 'uninitialized';

        return [] unless @$points;

        my $filter = $step->filter;
        return $points unless defined $filter;

        $filter =~ s/^\[\s*(.*?)\s*\]$/$1/; # strip brackets and whitespace

        if ($filter =~ /^-?\d+$/)
        {
                return $self->_filter_points_index($filter, $points); # simple array index
        }
        elsif ($filter =~ /\S/)
        {
                return $self->_filter_points_eval($filter, $points); # full condition
        }
        else
        {
                return $points;
        }
}

# the root node
# (only makes sense at first step, but currently not asserted)
sub _select_root {
        my ($self, $step, $current_points, $new_points) = @_;

        my $step_points = $self->_filter_points($step, $current_points);
        push @$new_points, @$step_points;
}


# //
# anywhere in the tree
sub _select_anywhere {
        my ($self, $step, $current_points, $lookahead, $new_points) = @_;

        # speed optimization: only useful points added
        my $lookahead_key;
        if (defined $lookahead and $lookahead->kind eq KEY) {
                $lookahead_key = $lookahead->part;
        }

        # '//'
        # all hash/array nodes of a data structure
        foreach my $point (@$current_points) {
                my @step_points = (@{_any([], [ $point ], $lookahead_key)}, $point);
                push @$new_points, @{$self->_filter_points($step, \@step_points)};
        }
}

# /key
# the value of a key
sub _select_key {
        my ($self, $step, $current_points, $new_points) = @_;

        foreach my $point (@$current_points) {
                no warnings 'uninitialized';
                next unless defined $point;
                my $pref = $point->ref;
                next unless (
                             # speed optimization:
                             # first try faster ref, then reftype
                             ref($$pref)     eq HASH or
                             reftype($$pref) eq HASH
                             );
                                # take point as hash, skip undefs
                my $attrs = Attrs->new(key => $step->part);
                my $step_points = [];
                if (exists $$pref->{$step->part}) {
                        $step_points = [ Point->new->ref(\($$pref->{$step->part}))->parent($point)->attrs($attrs) ];
                }
                push @$new_points, @{$self->_filter_points($step, $step_points)};
        }
}

# '*'
# all leaves of a data tree
sub _select_anystep {
        my ($self, $step, $current_points, $new_points) = @_;

        no warnings 'uninitialized';
        foreach my $point (@$current_points) {
                # take point as array
                my $pref = $point->ref;
                my $ref = $$pref;
                my $step_points = [];
                # speed optimization: first try faster ref, then reftype
                if (ref($ref) eq HASH or reftype($ref) eq HASH) {
                        $step_points = [ map {
                                my $v_ref = \($ref->{$_});
                                my $attrs = Attrs->new(key => $_);
                                Point->new->ref($v_ref)->parent($point)->attrs($attrs)
                        } keys %$ref ];
                } elsif (ref($ref) eq ARRAY or reftype($ref) eq ARRAY) {
                        my $idx = 0;
                        $step_points = [ map {
                                my $attrs = Attrs->new(idx => $idx++);
                                Point->new->ref(\$_)->parent($point)->attrs($attrs)
                        } @$ref ];
                } else {
                        if (ref($pref) eq SCALAR or reftype($pref) eq SCALAR) {
                                # TODO: without map, it's just one value
                                $step_points = [ #map {
                                        Point->new->ref($pref)->parent($point) # XXX? why $_? What happens to $pref?
                                ]; # } $ref ];
                        }
                }
                push @$new_points, @{ $self->_filter_points($step, $step_points) };
        }
}

# '.'
# no step (neither up nor down), just allow filtering
sub _select_nostep {
        my ($self, $step, $current_points, $new_points) = @_;

        foreach my $point (@{$current_points}) {
                my $step_points = [$point];
                push @$new_points, @{ $self->_filter_points($step, $step_points) };
        }
}

# '..'
# the parent
sub _select_parent {
        my ($self, $step, $current_points, $new_points) = @_;

        foreach my $point (@{$current_points}) {
                next unless defined $point;
                my $step_points = [$point->parent];
                push @$new_points, @{ $self->_filter_points($step, $step_points) };
        }
}

# '::ancestor'
# all ancestors (parent, grandparent, etc.) of the current node
sub _select_ancestor {
        my ($self, $step, $current_points, $new_points) = @_;

        foreach my $point (@{$current_points}) {
                my $step_points = [];
                my $parent = $point;
                while ($parent = $parent->parent) {
                        push @$step_points, $parent; # order matters
                }
                push @$new_points, @{ $self->_filter_points($step, $step_points) };
        }
}

# '::ancestor-or-self'
# all ancestors (parent, grandparent, etc.) of the current node and the current node itself
sub _select_ancestor_or_self {
        my ($self, $step, $current_points, $new_points) = @_;

        foreach my $point (@{$current_points}) {
                my $step_points = [$point];
                my $parent = $point;
                while ($parent = $parent->parent) {
                        push @$step_points, $parent; # order matters
                }
                push @$new_points, @{ $self->_filter_points($step, $step_points) };
        }
}

sub ref {
        my ($self) = @_;
        $self->first_point->{ref};
}

sub deref {
        my ($self) = @_;
        ${$self->ref};
}

sub first_point {
        my ($self) = @_;
        $self->current_points->[0];
}

sub all_points {
        my ($self) = @_;
        iarray $self->current_points;
}

sub _iter {
        my ($self) = @_;

        my $iter = iarray $self->current_points;
        return imap { __PACKAGE__->new->current_points([ $_ ]) } $iter;
}

sub isearch
{
        my ($self, $path_str) = @_;
        $self->_search(Data::DPath::Path->new(path => $path_str))->_iter;
}

sub _search
{
        my ($self, $dpath) = @_;

        no strict 'refs';
        no warnings 'uninitialized';

        my $current_points = $self->current_points;
        my $steps = $dpath->_steps;
        for (my $i = 0; $i < @$steps; $i++) {
                my $step = $steps->[$i];
                my $lookahead = $steps->[$i+1];
                my $new_points = [];

                if ($step->kind eq ROOT)
                {
                        $self->_select_root($step, $current_points, $new_points);
                }
                elsif ($step->kind eq ANYWHERE)
                {
                        $self->_select_anywhere($step, $current_points, $lookahead, $new_points);
                }
                elsif ($step->kind eq KEY)
                {
                        $self->_select_key($step, $current_points, $new_points);
                }
                elsif ($step->kind eq ANYSTEP)
                {
                        $self->_select_anystep($step, $current_points, $new_points);
                }
                elsif ($step->kind eq NOSTEP)
                {
                        $self->_select_nostep($step, $current_points, $new_points);
                }
                elsif ($step->kind eq PARENT)
                {
                        $self->_select_parent($step, $current_points, $new_points);
                }
                elsif ($step->kind eq ANCESTOR)
                {
                        $self->_select_ancestor($step, $current_points, $new_points);
                }
                elsif ($step->kind eq ANCESTOR_OR_SELF)
                {
                        $self->_select_ancestor_or_self($step, $current_points, $new_points);
                }
                $current_points = $new_points;
        }
        $self->current_points( $current_points );
        return $self;
}

sub match {
        my ($self, $dpath) = @_;

        $self->_search($dpath)->_all;
}

sub matchr {
        my ($self, $dpath) = @_;

        $self->_search($dpath)->_all_ref;
}

1;

__END__

=pod

=encoding UTF-8

=head1 NAME

Data::DPath::Context - Abstraction for a current context that enables incremental searches

=head1 API METHODS

=head2 new ( %args )

Constructor; creates instance.

Args:

=over 4

=item give_references

Default 0. If set to true value then results are references to the
matched points in the data structure.

=back

=head2 match( $dpath )

Return array of all data that match the given DPath.

=head2 matchr( $dpath )

Return array ref of all data that match the given DPath.

=head2 isearch( $path_str )

Searches a path relative to current context and returns an iterator.
See L<Iterator style|Data::DPath/"Iterator style"> for usage.

=head2 ref()

It returns the reference to the actual data from the current context's
first element. This mostly makes sense on contexts returned by
iterators as there is only one point there.

(Having the reference theoretically allows you to even change the data
on this point. It's not yet clear what impact this has to currently
active iterators, which B<should> still return the original data but
that's not yet tested. So don't rely on that behaviour.)

=head2 deref()

This is one dereference step on top of F<ref()>. It gives you the
actual data found. Most of the time you want this.

=head2 first_point

On a current context consisting on a set of points it returns the
first point. This makes most sense with Iterator style API when the
current iterator contains exactly one point.

=head2 all_points

On a current context consisting on a set of points it returns all
those. This method is a functional complement to F<first_point>.

=head1 UTILITY SUBS/METHODS

=head2 _all

Returns all values covered by current context.

If C<give_references> is set to true value then results are references
to the matched points in the data structure.

=head2 _search( $dpath )

Return new context for a DPath relative to current context.

=head2 _filter_points

Evaluates the filter condition in brackets. It differenciates between
simple integers, which are taken as array index, and all other
conditions, which are taken as evaled perl expression in a grep like
expression onto the set of points found by current step.

=head2 current_points

Attribute / accessor.

=head2 give_references

Attribute / accessor.

=head1 aliased classes

That's just to make Pod::Coverage happy which does not handle aliased
modules.

=head2 Context

=head2 Point

=head2 Step

=head1 AUTHOR

Steffen Schwigon <ss5@renormalist.net>

=head1 COPYRIGHT AND LICENSE

This software is copyright (c) 2019 by Steffen Schwigon.

This is free software; you can redistribute it and/or modify it under
the same terms as the Perl 5 programming language system itself.

=cut