[scpubgit/stemmatology.git] / lib / Text / Tradition / Analysis.pm

package Text::Tradition::Analysis;

use strict;
use warnings;
use Benchmark;
use Exporter 'import';
use Text::Tradition;
use Text::Tradition::Stemma;

use vars qw/ @EXPORT_OK /;
@EXPORT_OK = qw/ run_analysis group_variants wit_stringify /;

sub new {
	my( $class, $args ) = @_;
	my $self = {};
	bless( $self, $class );	
	$self->{'data'} = [];
	foreach my $t ( @{$args->{'traditions'}} ) {
	    $self->run_analysis( $t->{'file'}, $t->{'stemmadot'} );
	}
	return $self;
}

sub run_analysis {
	my( $self, $file, $stemmadot ) = @_;
	# What we will return
	my $svg;
	my $variants = [];
	my $data = {};
	
	# Read in the file and stemma	
	my $tradition = Text::Tradition->new( 
		'input'  => 'Self',
		'file'   => $file,
		'linear' => 1,
		);
	$data->{'title'} = $tradition->name;
	
	my $stemma = Text::Tradition::Stemma->new(
		'collation' => $tradition->collation,
		'dot' => $stemmadot,
		);
	# We will return the stemma picture
	$svg = $stemma->as_svg( { size => "8,7.5" } );;
	$data->{'svg'} = $svg;
	
	# We have the collation, so get the alignment table with witnesses in rows.
	# Also return the reading objects in the table, rather than just the words.
	my $wits = {};
	map { $wits->{$_} = 1 } $stemma->witnesses;
	# For each column in the alignment table, we want to see if the existing
	# groupings of witnesses match our stemma hypothesis. We also need to keep
	# track of the maximum number of variants at any one location.
	my $max_variants = 0;
	my ( $total, $genealogical, $conflicts ) = ( 0, 0, 0 );
	
    my $t0 = Benchmark->new();
    my $variant_groups = group_variants( $tradition->collation, $wits );
    foreach my $rank ( 0 .. $#{$variant_groups} ) {
        my $groups = $variant_groups->[$rank]->{'groups'};
        my $readings = $variant_groups->[$rank]->{'readings'};
        my $lacunose = $variant_groups->[$rank]->{'lacunose'};
		
		$max_variants = scalar @$groups if scalar @$groups > $max_variants;
		
		# We can already look up witnesses for a reading; we also want to look
		# up readings for a given witness.
		my $group_readings = {};
		foreach my $x ( 0 .. $#$groups ) {
			$group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x];
		}
		
		# For all the groups with more than one member, collect the list of all
		# contiguous vertices needed to connect them.
		my $variant_loc = analyze_variant_location( $group_readings, $groups, 
		    $stemma->graph, $lacunose );
		$variant_loc->{'id'} = $rank;
		$genealogical++ if $variant_loc->{'genealogical'};
		$conflicts += grep { $_->{'conflict'} } @{$variant_loc->{'readings'}};

		# Now run the same analysis given the calculated distance tree(s).
# 		my @trees = @{$stemma->distance_trees};
# 		if( @trees ) {
#             foreach my $tree ( 0 .. $#trees ) {
#                 my $dc = analyze_variant_location( $group_readings, $groups, $tree, $lacunose, 'undirected' );
#                 foreach my $rdg ( keys %$dc ) {
#                     my $var = $dc->{$rdg};
#                     # TODO Do something with this
#                 }
#             }
# 	    }

		# Record that we used this variant in an analysis
		push( @$variants, $variant_loc );
	}
    my $t1 = Benchmark->new();
    print STDERR "Analysis of graph for " . $tradition->name . " took " 
        . timestr( timediff( $t1, $t0 ) ) . "seconds\n";
	
	# Go through our variant locations, after we have seen all of them once,
	# and add the number of empty columns needed by each.
	foreach my $row ( @$variants ) {
		my $empty = $max_variants - scalar @{$row->{'readings'}};
		$row->{'empty'} = $empty;
	}
	
	# Populate self with our analysis data.
	$data->{'variants'} = $variants;
	$data->{'variant_count'} = $total;
	$data->{'conflict_count'} = $conflicts;
	$data->{'genealogical_count'} = $genealogical;
	push( @{$self->{'data'}}, $data );
}

sub group_variants {
	my( $c, $wits ) = @_;
	my $variant_groups = [];
	
	# We have the collation, so get the alignment table with witnesses in rows.
	# Also return the reading objects in the table, rather than just the words.
	my $all_wits_table = $c->make_alignment_table( 'refs', $wits );
	# Strip the list of sigla and save it for correlation to the readings.
	my $col_wits = shift @$all_wits_table;
	# Any witness in the stemma that has no row should be noted.
    foreach ( @$col_wits ) {
        $wits->{$_}++; # Witnesses present in table and stemma now have value 2.
    }
    my @not_collated = grep { $wits->{$_} == 1 } keys %$wits;
	foreach my $i ( 0 .. $#$all_wits_table ) {
		# For each column in the table, group the readings by witness.
		my $rdg_wits = {};
		my $col_rdgs = shift @$all_wits_table;
		my $lacunose = [ @not_collated ];
		foreach my $j ( 0 .. $#{$col_rdgs} ) {
			my $rdg = $col_rdgs->[$j];
			my $rdg_text = '(omitted)';  # Initialize in case of empty reading
			if( $rdg ) {
			    if( $rdg->is_lacuna ) {
			        $rdg_text = undef;   # Don't count lacunae
			        push( @$lacunose, $col_wits->[$j] );
			    } else {
    				$rdg_text = $rdg->text; 
				}
			}
			if( defined $rdg_text ) {
				# Initialize the witness array if we haven't got one yet
				$rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text};
				# Add the relevant witness, subject to a.c. logic
				add_variant_wit( $rdg_wits->{$rdg_text}, $col_wits->[$j],
					$c->ac_label );
			}
		}
		
		# See if this column has any potentially genealogical variants.
		# If not, skip to the next.
		my( $groups, $readings ) = useful_variant( $rdg_wits );
		next unless $groups && $readings;  

		push( @$variant_groups, 
		    { 'groups' => $groups, 'readings' => $readings, 'lacunose' => $lacunose } );
	}
	return $variant_groups;
}


# variant_row -> genealogical
#             -> readings [ { text, group, conflict, missing } ]

sub analyze_variant_location {
    my( $group_readings, $groups, $graph, $lacunose, $undirected ) = @_;
    my $contig = {};
    my $subgraph = {};
    my $is_conflicted;
    my $conflict = {};
    my $missing = {};
    map { $missing->{$_} = 1 } @$lacunose;
    my $variant_row = { 'readings' => [] };
    # Mark each ms as in its own group, first.
    foreach my $g ( @$groups ) {
        my $gst = wit_stringify( $g );
        map { $contig->{$_} = $gst } @$g;
    }
    # Now for each unmarked node in the graph, initialize an array
    # for possible group memberships.  We will use this later to
    # resolve potential conflicts.
    map { $contig->{$_} = [] unless $contig->{$_} } $graph->vertices;
    foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) {
        my $gst = wit_stringify( $g );  # This is the group name
        my $reachable = { $g->[0] => 1 };
        # Copy the graph, and delete all non-members from the new graph.
        my $part = $graph->copy;
        my $group_root;
        $part->delete_vertices( 
            grep { !ref( $contig->{$_} ) && $contig->{$_} ne $gst } $graph->vertices );
                
        # Now look to see if our group is connected.
        if( $undirected ) { # For use with distance trees etc.
            # Find all vertices reachable from the first (arbitrary) group
            # member.  If we are genealogical this should include them all. 
            map { $reachable->{$_} = 1 } $part->all_reachable( $g->[0] );
            # TODO This is a terrible way to do distance trees, since all
            # non-leaf nodes are included in every graph part now. We may
            # have to go back to SPDP.
        } else {
            if( @$g > 1 ) {
                # Dispense with the trivial case of one reading.
                # We have to take directionality into account.
                # How many root nodes do we have?
                my @roots = grep { ref( $contig->{$_} ) || $contig->{$_} eq $gst } 
                    $part->source_vertices;
                # Assuming that @$g > 1, find the first root node that has at
                # least one successor belonging to our group. If this reading
                # is genealogical, there should be only one, but we will check
                # that implicitly later.
                my $nodes_in_subtree = 0;
                foreach my $root ( @roots ) {
                    # Prune the tree to get rid of extraneous hypotheticals.
                    $root = prune_subtree( $part, $root, $contig );
                    # Get all the successor nodes of our root.
                    my $tmp_reach = { $root => 1 };
                    map { $tmp_reach->{$_} = 1 } $part->all_successors( $root );
                    # Skip this root if none of our successors are in our group
                    # (e.g. isolated 'hypothetical' witnesses with no group)
                    next unless grep { $contig->{$_} } keys %$tmp_reach;
                    if( keys %$tmp_reach > $nodes_in_subtree ) {
                        $nodes_in_subtree = keys %$tmp_reach;
                        $reachable = $tmp_reach;
                        $group_root = $root;
                    }
                }
            } # else it is a single-node group, nothing to calculate.
        }
        
        # None of the 'reachable' nodes should be marked as being in another 
        # group.  Paint the 'hypotheticals' with our group while we are at it,
        # unless there is a conflict present.
        foreach ( keys %$reachable ) {
            if( ref $contig->{$_} ) {
                push( @{$contig->{$_}}, $gst );
            } elsif( $contig->{$_} ne $gst ) {
                $conflict->{$group_readings->{$gst}} = $group_readings->{$contig->{$_}};
            } # else it is an 'extant' node marked with our group already.
        }
        # None of the unreachable nodes should be in our group either.
        foreach ( $part->vertices ) {
            next if $reachable->{$_};
            if( $contig->{$_} eq $gst ) {
                $conflict->{$group_readings->{$gst}} = $group_readings->{$gst};
                last;
            }
        }
        
        # Now, if we have a conflict, we can write the reading in full.  If not, 
        # we have to save the subgraph so that we can resolve possible conflicts 
        # on hypothetical nodes.
        $is_conflicted = 1 if exists $conflict->{$group_readings->{$gst}};
        
        # Write the reading.
        my $reading = { 'text' => $group_readings->{$gst},
                        'missing' => wit_stringify( $lacunose ),
                        'group' => $gst };  # This will change if we find no conflict
        if( $is_conflicted ) {
            $reading->{'conflict'} = $conflict->{$group_readings->{$gst}}
        } else {
            # Save the relevant subgraph.
            $subgraph->{$gst} = { 'graph' => $part,
                                'root' => $group_root,
                                'reachable' => $reachable };
        }
        push( @{$variant_row->{'readings'}}, $reading );
    }
    
    # Now that we have gone through all the rows, check the hypothetical
    # readings for conflict if we haven't found one yet.
    if( keys %$subgraph && !keys %$conflict ) {
        my @resolve;
        foreach ( keys %$contig ) {
            next unless ref $contig->{$_};
            if( scalar @{$contig->{$_}} > 1 ) {
                push( @resolve, $_ );
            } else {
                $contig->{$_} = scalar @{$contig->{$_}} ? $contig->{$_}->[0] : '';
            }
        }
        # Do we still have a possible conflict?
        my $still_contig = {};
        foreach my $h ( @resolve ) {
            # For each of the hypothetical readings with more than one possibility,
            # try deleting it from each of its member subgraphs in turn, and see
            # if that breaks the contiguous grouping.
            # TODO This can still break in a corner case where group A can use 
            # either vertex 1 or 2, and group B can use either vertex 2 or 1.
            # Revisit this if necessary; it could get brute-force nasty.
            foreach my $gst ( @{$contig->{$h}} ) {
                my $gpart = $subgraph->{$gst}->{'graph'}->copy;
                my $reachable = $subgraph->{$gst}->{'reachable'};
                $gpart->delete_vertex( $h );
                # Is everything else still reachable from the root?
                # TODO If $h was the root, see if we still have a single root.
                my %still_reachable = ( $subgraph->{$gst}->{'root'} => 1 );
                map { $still_reachable{$_} = 1 }
                    $gpart->all_successors( $subgraph->{$gst}->{'root'} );
                foreach my $v ( keys %$reachable ) {
                    next if $v eq $h;
                    if( !$still_reachable{$v}
                        && ( $contig->{$v} eq $gst 
                             || ( exists $still_contig->{$v} 
                                  && $still_contig->{$v} eq $gst ) ) ) {
                        # We need $h.
                        if( exists $still_contig->{$h} ) {
                            # Conflict!
                            $conflict->{$group_readings->{$gst}} = 
                                $group_readings->{$still_contig->{$h}};
                        } else {
                            $still_contig->{$h} = $gst;
                        }
                        last;
                    } # else we don't need $h in this group.
                }
            }
        }
        
        # Now, assuming no conflict, we have some hypothetical vertices in
        # $still_contig that are the "real" group memberships.  Replace these
        # in $contig.
        unless ( keys %$conflict ) {
            foreach my $v ( keys %$contig ) {
                next unless ref $contig->{$v};
                $contig->{$v} = $still_contig->{$v};
            }
        }
    }
            
    # Now write the group and conflict information into the respective rows.
    foreach my $rdg ( @{$variant_row->{'readings'}} ) {
        $rdg->{'conflict'} = $conflict->{$rdg->{'text'}};
        next if $rdg->{'conflict'};
        my @members = grep { $contig->{$_} eq $rdg->{'group'} && !$missing->{$_} } 
            keys %$contig;
        $rdg->{'group'} = wit_stringify( \@members );
    }
    
    $variant_row->{'genealogical'} = !( keys %$conflict );
    return $variant_row;
}

sub prune_subtree {
    my( $tree, $root, $contighash ) = @_;
    # First, delete hypothetical leaves / orphans until there are none left.
    my @orphan_hypotheticals = grep { ref( $contighash->{$_} ) } 
        $tree->successorless_vertices;
    while( @orphan_hypotheticals ) {
        $tree->delete_vertices( @orphan_hypotheticals );
        @orphan_hypotheticals = grep { ref( $contighash->{$_} ) } 
            $tree->successorless_vertices;
    }
    # Then delete a hypothetical root with only one successor, moving the
    # root to the child.
    while( $tree->successors( $root ) == 1 && ref $contighash->{$root} ) {
        my @nextroot = $tree->successors( $root );
        $tree->delete_vertex( $root );
        $root = $nextroot[0];
    }
    # The tree has been modified in place, but we need to know the new root.
    return $root;
}
# Add the variant, subject to a.c. representation logic.
# This assumes that we will see the 'main' version before the a.c. version.
sub add_variant_wit {
    my( $arr, $wit, $acstr ) = @_;
    my $skip;
    if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
        my $real = $1;
        $skip = grep { $_ =~ /^\Q$real\E$/ } @$arr;
    } 
    push( @$arr, $wit ) unless $skip;
}

# Return an answer if the variant is useful, i.e. if there are at least 2 variants
# with at least 2 witnesses each.
sub useful_variant {
    my( $readings ) = @_;
    my $total = keys %$readings;
    foreach my $var ( keys %$readings ) {
        $total-- if @{$readings->{$var}} == 1;
    }
    return( undef, undef ) if $total <= 1;
    my( $groups, $text );
    foreach my $var ( keys %$readings ) {
        push( @$groups, $readings->{$var} );
        push( @$text, $var );
    }
    return( $groups, $text );
}

# Take an array of witness groupings and produce a string like
# ['A','B'] / ['C','D','E'] / ['F']

sub wit_stringify {
    my $groups = shift;
    my @gst;
    # If we were passed an array of witnesses instead of an array of 
    # groupings, then "group" the witnesses first.
    unless( ref( $groups->[0] ) ) {
        my $mkgrp = [ $groups ];
        $groups = $mkgrp;
    }
    foreach my $g ( @$groups ) {
        push( @gst, '[' . join( ',', map { "'$_'" } @$g ) . ']' );
    }
    return join( ' / ', @gst );
}
    
1;
Commit	Line	Data
d71100ed	1	package Text::Tradition::Analysis;
	2
	3	use strict;
	4	use warnings;
e4386ba9	5	use Benchmark;
d1348d38	6	use Exporter 'import';
d71100ed	7	use Text::Tradition;
	8	use Text::Tradition::Stemma;
	9
d1348d38	10	use vars qw/ @EXPORT_OK /;
	11	@EXPORT_OK = qw/ run_analysis group_variants wit_stringify /;
	12
d71100ed	13	sub new {
	14	my( $class, $args ) = @_;
	15	my $self = {};
d71100ed	16	bless( $self, $class );
e367f5c0	17	$self->{'data'} = [];
	18	foreach my $t ( @{$args->{'traditions'}} ) {
	19	$self->run_analysis( $t->{'file'}, $t->{'stemmadot'} );
	20	}
d71100ed	21	return $self;
	22	}
	23
	24	sub run_analysis {
732152b1	25	my( $self, $file, $stemmadot ) = @_;
d71100ed	26	# What we will return
	27	my $svg;
	28	my $variants = [];
e367f5c0	29	my $data = {};
d71100ed	30
3d79e248	31	# Read in the file and stemma
d71100ed	32	my $tradition = Text::Tradition->new(
3d79e248	33	'input' => 'Self',
3d79e248	34	'file' => $file,
d71100ed	35	'linear' => 1,
d71100ed	36	);
e367f5c0	37	$data->{'title'} = $tradition->name;
732152b1	38
d71100ed	39	my $stemma = Text::Tradition::Stemma->new(
	40	'collation' => $tradition->collation,
	41	'dot' => $stemmadot,
	42	);
	43	# We will return the stemma picture
e367f5c0	44	$svg = $stemma->as_svg( { size => "8,7.5" } );;
e367f5c0	45	$data->{'svg'} = $svg;
d71100ed	46
	47	# We have the collation, so get the alignment table with witnesses in rows.
	48	# Also return the reading objects in the table, rather than just the words.
08e0fb85	49	my $wits = {};
08e0fb85	50	map { $wits->{$_} = 1 } $stemma->witnesses;
d71100ed	51	# For each column in the alignment table, we want to see if the existing
f6c8ea08	52	# groupings of witnesses match our stemma hypothesis. We also need to keep
	53	# track of the maximum number of variants at any one location.
	54	my $max_variants = 0;
732152b1	55	my ( $total, $genealogical, $conflicts ) = ( 0, 0, 0 );
d71100ed	56
e4386ba9	57	my $t0 = Benchmark->new();
f6c8ea08	58	my $variant_groups = group_variants( $tradition->collation, $wits );
	59	foreach my $rank ( 0 .. $#{$variant_groups} ) {
	60	my $groups = $variant_groups->[$rank]->{'groups'};
	61	my $readings = $variant_groups->[$rank]->{'readings'};
	62	my $lacunose = $variant_groups->[$rank]->{'lacunose'};
d71100ed	63
f6c8ea08	64	$max_variants = scalar @$groups if scalar @$groups > $max_variants;
d71100ed	65
	66	# We can already look up witnesses for a reading; we also want to look
	67	# up readings for a given witness.
	68	my $group_readings = {};
	69	foreach my $x ( 0 .. $#$groups ) {
	70	$group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x];
	71	}
	72
	73	# For all the groups with more than one member, collect the list of all
	74	# contiguous vertices needed to connect them.
f6c8ea08	75	my $variant_loc = analyze_variant_location( $group_readings, $groups,
08e0fb85	76	$stemma->graph, $lacunose );
f6c8ea08	77	$variant_loc->{'id'} = $rank;
	78	$genealogical++ if $variant_loc->{'genealogical'};
	79	$conflicts += grep { $_->{'conflict'} } @{$variant_loc->{'readings'}};
732152b1	80
d71100ed	81	# Now run the same analysis given the calculated distance tree(s).
732152b1	82	# my @trees = @{$stemma->distance_trees};
	83	# if( @trees ) {
	84	# foreach my $tree ( 0 .. $#trees ) {
c4a4fb1b	85	# my $dc = analyze_variant_location( $group_readings, $groups, $tree, $lacunose, 'undirected' );
732152b1	86	# foreach my $rdg ( keys %$dc ) {
	87	# my $var = $dc->{$rdg};
	88	# # TODO Do something with this
	89	# }
	90	# }
	91	# }
	92
d71100ed	93	# Record that we used this variant in an analysis
f6c8ea08	94	push( @$variants, $variant_loc );
d71100ed	95	}
e4386ba9	96	my $t1 = Benchmark->new();
	97	print STDERR "Analysis of graph for " . $tradition->name . " took "
	98	. timestr( timediff( $t1, $t0 ) ) . "seconds\n";
d71100ed	99
f6c8ea08	100	# Go through our variant locations, after we have seen all of them once,
732152b1	101	# and add the number of empty columns needed by each.
d71100ed	102	foreach my $row ( @$variants ) {
f6c8ea08	103	my $empty = $max_variants - scalar @{$row->{'readings'}};
d71100ed	104	$row->{'empty'} = $empty;
	105	}
	106
732152b1	107	# Populate self with our analysis data.
e367f5c0	108	$data->{'variants'} = $variants;
	109	$data->{'variant_count'} = $total;
	110	$data->{'conflict_count'} = $conflicts;
	111	$data->{'genealogical_count'} = $genealogical;
	112	push( @{$self->{'data'}}, $data );
d71100ed	113	}
d71100ed	114
d1348d38	115	sub group_variants {
	116	my( $c, $wits ) = @_;
	117	my $variant_groups = [];
	118
f6c8ea08	119	# We have the collation, so get the alignment table with witnesses in rows.
f6c8ea08	120	# Also return the reading objects in the table, rather than just the words.
d1348d38	121	my $all_wits_table = $c->make_alignment_table( 'refs', $wits );
	122	# Strip the list of sigla and save it for correlation to the readings.
	123	my $col_wits = shift @$all_wits_table;
	124	# Any witness in the stemma that has no row should be noted.
	125	foreach ( @$col_wits ) {
	126	$wits->{$_}++; # Witnesses present in table and stemma now have value 2.
	127	}
f6c8ea08	128	my @not_collated = grep { $wits->{$_} == 1 } keys %$wits;
d1348d38	129	foreach my $i ( 0 .. $#$all_wits_table ) {
	130	# For each column in the table, group the readings by witness.
	131	my $rdg_wits = {};
	132	my $col_rdgs = shift @$all_wits_table;
d1348d38	133	my $lacunose = [ @not_collated ];
	134	foreach my $j ( 0 .. $#{$col_rdgs} ) {
	135	my $rdg = $col_rdgs->[$j];
	136	my $rdg_text = '(omitted)'; # Initialize in case of empty reading
	137	if( $rdg ) {
	138	if( $rdg->is_lacuna ) {
	139	$rdg_text = undef; # Don't count lacunae
	140	push( @$lacunose, $col_wits->[$j] );
	141	} else {
	142	$rdg_text = $rdg->text;
d1348d38	143	}
	144	}
	145	if( defined $rdg_text ) {
	146	# Initialize the witness array if we haven't got one yet
	147	$rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text};
	148	# Add the relevant witness, subject to a.c. logic
	149	add_variant_wit( $rdg_wits->{$rdg_text}, $col_wits->[$j],
	150	$c->ac_label );
	151	}
	152	}
	153
	154	# See if this column has any potentially genealogical variants.
	155	# If not, skip to the next.
	156	my( $groups, $readings ) = useful_variant( $rdg_wits );
	157	next unless $groups && $readings;
	158
f6c8ea08	159	push( @$variant_groups,
f6c8ea08	160	{ 'groups' => $groups, 'readings' => $readings, 'lacunose' => $lacunose } );
d1348d38	161	}
	162	return $variant_groups;
	163	}
	164
f6c8ea08	165
f6c8ea08	166
732152b1	167	# variant_row -> genealogical
	168	# -> readings [ { text, group, conflict, missing } ]
	169
d71100ed	170	sub analyze_variant_location {
c4a4fb1b	171	my( $group_readings, $groups, $graph, $lacunose, $undirected ) = @_;
231d71fc	172	my $contig = {};
231d71fc	173	my $subgraph = {};
c4a4fb1b	174	my $is_conflicted;
d71100ed	175	my $conflict = {};
231d71fc	176	my $missing = {};
231d71fc	177	map { $missing->{$_} = 1 } @$lacunose;
732152b1	178	my $variant_row = { 'readings' => [] };
94a077d6	179	# Mark each ms as in its own group, first.
	180	foreach my $g ( @$groups ) {
	181	my $gst = wit_stringify( $g );
231d71fc	182	map { $contig->{$_} = $gst } @$g;
94a077d6	183	}
c4a4fb1b	184	# Now for each unmarked node in the graph, initialize an array
	185	# for possible group memberships. We will use this later to
	186	# resolve potential conflicts.
231d71fc	187	map { $contig->{$_} = [] unless $contig->{$_} } $graph->vertices;
d71100ed	188	foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) {
c4a4fb1b	189	my $gst = wit_stringify( $g ); # This is the group name
c4a4fb1b	190	my $reachable = { $g->[0] => 1 };
08e0fb85	191	# Copy the graph, and delete all non-members from the new graph.
c4a4fb1b	192	my $part = $graph->copy;
	193	my $group_root;
	194	$part->delete_vertices(
231d71fc	195	grep { !ref( $contig->{$_} ) && $contig->{$_} ne $gst } $graph->vertices );
c4a4fb1b	196
	197	# Now look to see if our group is connected.
	198	if( $undirected ) { # For use with distance trees etc.
	199	# Find all vertices reachable from the first (arbitrary) group
	200	# member. If we are genealogical this should include them all.
	201	map { $reachable->{$_} = 1 } $part->all_reachable( $g->[0] );
	202	# TODO This is a terrible way to do distance trees, since all
	203	# non-leaf nodes are included in every graph part now. We may
	204	# have to go back to SPDP.
	205	} else {
	206	if( @$g > 1 ) {
	207	# Dispense with the trivial case of one reading.
	208	# We have to take directionality into account.
	209	# How many root nodes do we have?
231d71fc	210	my @roots = grep { ref( $contig->{$_} ) \|\| $contig->{$_} eq $gst }
c4a4fb1b	211	$part->source_vertices;
	212	# Assuming that @$g > 1, find the first root node that has at
	213	# least one successor belonging to our group. If this reading
	214	# is genealogical, there should be only one, but we will check
	215	# that implicitly later.
	216	my $nodes_in_subtree = 0;
	217	foreach my $root ( @roots ) {
231d71fc	218	# Prune the tree to get rid of extraneous hypotheticals.
231d71fc	219	$root = prune_subtree( $part, $root, $contig );
c4a4fb1b	220	# Get all the successor nodes of our root.
	221	my $tmp_reach = { $root => 1 };
	222	map { $tmp_reach->{$_} = 1 } $part->all_successors( $root );
	223	# Skip this root if none of our successors are in our group
	224	# (e.g. isolated 'hypothetical' witnesses with no group)
231d71fc	225	next unless grep { $contig->{$_} } keys %$tmp_reach;
c4a4fb1b	226	if( keys %$tmp_reach > $nodes_in_subtree ) {
	227	$nodes_in_subtree = keys %$tmp_reach;
	228	$reachable = $tmp_reach;
	229	$group_root = $root;
	230	}
	231	}
	232	} # else it is a single-node group, nothing to calculate.
	233	}
	234
	235	# None of the 'reachable' nodes should be marked as being in another
	236	# group. Paint the 'hypotheticals' with our group while we are at it,
	237	# unless there is a conflict present.
	238	foreach ( keys %$reachable ) {
231d71fc	239	if( ref $contig->{$_} ) {
	240	push( @{$contig->{$_}}, $gst );
	241	} elsif( $contig->{$_} ne $gst ) {
	242	$conflict->{$group_readings->{$gst}} = $group_readings->{$contig->{$_}};
c4a4fb1b	243	} # else it is an 'extant' node marked with our group already.
d71100ed	244	}
08e0fb85	245	# None of the unreachable nodes should be in our group either.
08e0fb85	246	foreach ( $part->vertices ) {
c4a4fb1b	247	next if $reachable->{$_};
231d71fc	248	if( $contig->{$_} eq $gst ) {
c4a4fb1b	249	$conflict->{$group_readings->{$gst}} = $group_readings->{$gst};
	250	last;
	251	}
08e0fb85	252	}
08e0fb85	253
c4a4fb1b	254	# Now, if we have a conflict, we can write the reading in full. If not,
	255	# we have to save the subgraph so that we can resolve possible conflicts
	256	# on hypothetical nodes.
	257	$is_conflicted = 1 if exists $conflict->{$group_readings->{$gst}};
	258
732152b1	259	# Write the reading.
	260	my $reading = { 'text' => $group_readings->{$gst},
	261	'missing' => wit_stringify( $lacunose ),
c4a4fb1b	262	'group' => $gst }; # This will change if we find no conflict
	263	if( $is_conflicted ) {
	264	$reading->{'conflict'} = $conflict->{$group_readings->{$gst}}
732152b1	265	} else {
c4a4fb1b	266	# Save the relevant subgraph.
231d71fc	267	$subgraph->{$gst} = { 'graph' => $part,
c4a4fb1b	268	'root' => $group_root,
c4a4fb1b	269	'reachable' => $reachable };
732152b1	270	}
732152b1	271	push( @{$variant_row->{'readings'}}, $reading );
d71100ed	272	}
c4a4fb1b	273
	274	# Now that we have gone through all the rows, check the hypothetical
	275	# readings for conflict if we haven't found one yet.
231d71fc	276	if( keys %$subgraph && !keys %$conflict ) {
c4a4fb1b	277	my @resolve;
231d71fc	278	foreach ( keys %$contig ) {
	279	next unless ref $contig->{$_};
	280	if( scalar @{$contig->{$_}} > 1 ) {
c4a4fb1b	281	push( @resolve, $_ );
c4a4fb1b	282	} else {
231d71fc	283	$contig->{$_} = scalar @{$contig->{$_}} ? $contig->{$_}->[0] : '';
c4a4fb1b	284	}
	285	}
	286	# Do we still have a possible conflict?
231d71fc	287	my $still_contig = {};
c4a4fb1b	288	foreach my $h ( @resolve ) {
	289	# For each of the hypothetical readings with more than one possibility,
	290	# try deleting it from each of its member subgraphs in turn, and see
	291	# if that breaks the contiguous grouping.
	292	# TODO This can still break in a corner case where group A can use
	293	# either vertex 1 or 2, and group B can use either vertex 2 or 1.
	294	# Revisit this if necessary; it could get brute-force nasty.
231d71fc	295	foreach my $gst ( @{$contig->{$h}} ) {
	296	my $gpart = $subgraph->{$gst}->{'graph'}->copy;
	297	my $reachable = $subgraph->{$gst}->{'reachable'};
c4a4fb1b	298	$gpart->delete_vertex( $h );
	299	# Is everything else still reachable from the root?
	300	# TODO If $h was the root, see if we still have a single root.
231d71fc	301	my %still_reachable = ( $subgraph->{$gst}->{'root'} => 1 );
c4a4fb1b	302	map { $still_reachable{$_} = 1 }
231d71fc	303	$gpart->all_successors( $subgraph->{$gst}->{'root'} );
c4a4fb1b	304	foreach my $v ( keys %$reachable ) {
	305	next if $v eq $h;
	306	if( !$still_reachable{$v}
231d71fc	307	&& ( $contig->{$v} eq $gst
	308	\|\| ( exists $still_contig->{$v}
	309	&& $still_contig->{$v} eq $gst ) ) ) {
c4a4fb1b	310	# We need $h.
231d71fc	311	if( exists $still_contig->{$h} ) {
c4a4fb1b	312	# Conflict!
c4a4fb1b	313	$conflict->{$group_readings->{$gst}} =
231d71fc	314	$group_readings->{$still_contig->{$h}};
c4a4fb1b	315	} else {
231d71fc	316	$still_contig->{$h} = $gst;
c4a4fb1b	317	}
	318	last;
	319	} # else we don't need $h in this group.
	320	}
	321	}
	322	}
	323
	324	# Now, assuming no conflict, we have some hypothetical vertices in
	325	# $still_contig that are the "real" group memberships. Replace these
	326	# in $contig.
	327	unless ( keys %$conflict ) {
231d71fc	328	foreach my $v ( keys %$contig ) {
	329	next unless ref $contig->{$v};
	330	$contig->{$v} = $still_contig->{$v};
c4a4fb1b	331	}
	332	}
	333	}
	334
	335	# Now write the group and conflict information into the respective rows.
	336	foreach my $rdg ( @{$variant_row->{'readings'}} ) {
	337	$rdg->{'conflict'} = $conflict->{$rdg->{'text'}};
	338	next if $rdg->{'conflict'};
231d71fc	339	my @members = grep { $contig->{$_} eq $rdg->{'group'} && !$missing->{$_} }
231d71fc	340	keys %$contig;
c4a4fb1b	341	$rdg->{'group'} = wit_stringify( \@members );
	342	}
	343
08e0fb85	344	$variant_row->{'genealogical'} = !( keys %$conflict );
732152b1	345	return $variant_row;
d71100ed	346	}
d71100ed	347
231d71fc	348	sub prune_subtree {
	349	my( $tree, $root, $contighash ) = @_;
	350	# First, delete hypothetical leaves / orphans until there are none left.
	351	my @orphan_hypotheticals = grep { ref( $contighash->{$_} ) }
	352	$tree->successorless_vertices;
	353	while( @orphan_hypotheticals ) {
	354	$tree->delete_vertices( @orphan_hypotheticals );
	355	@orphan_hypotheticals = grep { ref( $contighash->{$_} ) }
	356	$tree->successorless_vertices;
	357	}
	358	# Then delete a hypothetical root with only one successor, moving the
	359	# root to the child.
	360	while( $tree->successors( $root ) == 1 && ref $contighash->{$root} ) {
	361	my @nextroot = $tree->successors( $root );
	362	$tree->delete_vertex( $root );
	363	$root = $nextroot[0];
	364	}
	365	# The tree has been modified in place, but we need to know the new root.
	366	return $root;
	367	}
d71100ed	368	# Add the variant, subject to a.c. representation logic.
	369	# This assumes that we will see the 'main' version before the a.c. version.
	370	sub add_variant_wit {
	371	my( $arr, $wit, $acstr ) = @_;
	372	my $skip;
	373	if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
	374	my $real = $1;
	375	$skip = grep { $_ =~ /^\Q$real\E$/ } @$arr;
	376	}
	377	push( @$arr, $wit ) unless $skip;
	378	}
	379
	380	# Return an answer if the variant is useful, i.e. if there are at least 2 variants
	381	# with at least 2 witnesses each.
	382	sub useful_variant {
	383	my( $readings ) = @_;
	384	my $total = keys %$readings;
	385	foreach my $var ( keys %$readings ) {
	386	$total-- if @{$readings->{$var}} == 1;
	387	}
	388	return( undef, undef ) if $total <= 1;
	389	my( $groups, $text );
	390	foreach my $var ( keys %$readings ) {
	391	push( @$groups, $readings->{$var} );
	392	push( @$text, $var );
	393	}
	394	return( $groups, $text );
	395	}
	396
	397	# Take an array of witness groupings and produce a string like
	398	# ['A','B'] / ['C','D','E'] / ['F']
	399
	400	sub wit_stringify {
	401	my $groups = shift;
	402	my @gst;
	403	# If we were passed an array of witnesses instead of an array of
	404	# groupings, then "group" the witnesses first.
	405	unless( ref( $groups->[0] ) ) {
	406	my $mkgrp = [ $groups ];
	407	$groups = $mkgrp;
	408	}
	409	foreach my $g ( @$groups ) {
	410	push( @gst, '[' . join( ',', map { "'$_'" } @$g ) . ']' );
	411	}
	412	return join( ' / ', @gst );
	413	}
	414
	415	1;