fix graphml subgraph generation; fix tab parsing to only set 'collated' where needed...
[scpubgit/stemmatology.git] / lib / Text / Tradition / Collation / RelationshipStore.pm
index 6f46c14..314d003 100644 (file)
@@ -4,6 +4,7 @@ use strict;
 use warnings;
 use Text::Tradition::Error;
 use Text::Tradition::Collation::Relationship;
+use TryCatch;
 
 use Moose;
 
@@ -22,9 +23,34 @@ general) between readings.
 =begin testing
 
 use Text::Tradition;
+use TryCatch;
 
 use_ok( 'Text::Tradition::Collation::RelationshipStore' );
 
+# Add some relationships, and delete them
+
+my $cxfile = 't/data/Collatex-16.xml';
+my $t = Text::Tradition->new( 
+    'name'  => 'inline', 
+    'input' => 'CollateX',
+    'file'  => $cxfile,
+    );
+my $c = $t->collation;
+
+my @v1 = $c->add_relationship( 'n21', 'n22', { 'type' => 'lexical' } );
+is( scalar @v1, 1, "Added a single relationship" );
+is( $v1[0]->[0], 'n21', "Got correct node 1" );
+is( $v1[0]->[1], 'n22', "Got correct node 2" );
+my @v2 = $c->add_relationship( 'n24', 'n23', 
+       { 'type' => 'spelling', 'scope' => 'global' } );
+is( scalar @v2, 2, "Added a global relationship with two instances" );
+@v1 = $c->del_relationship( 'n22', 'n21' );
+is( scalar @v1, 1, "Deleted first relationship" );
+@v2 = $c->del_relationship( 'n12', 'n13' );
+is( scalar @v2, 2, "Deleted second global relationship" );
+my @v3 = $c->del_relationship( 'n1', 'n2' );
+is( scalar @v3, 0, "Nothing deleted on non-existent relationship" );
+
 =end testing
 
 =head1 METHODS
@@ -66,11 +92,19 @@ Return the relationship object, if any, that exists between two readings.
 =cut
 
 sub get_relationship {
-       my( $self, @vector ) = @_;
+       my $self = shift;
+       my @vector;
+       if( @_ == 1 && ref( $_[0] ) eq 'ARRAY' ) {
+               # Dereference the edge arrayref that was passed.
+               my $edge = shift;
+               @vector = @$edge;
+       } else {
+               @vector = @_;
+       }
        my $relationship;
        if( $self->graph->has_edge_attribute( @vector, 'object' ) ) {
                $relationship = $self->graph->get_edge_attribute( @vector, 'object' );
-       }
+       } 
        return $relationship;
 }
 
@@ -79,7 +113,7 @@ sub _set_relationship {
        $self->graph->add_edge( @vector );
        $self->graph->set_edge_attribute( @vector, 'object', $relationship );
 }
-       
+
 =head2 create
 
 Create a new relationship with the given options and return it.
@@ -94,7 +128,11 @@ sub create {
        my $target = delete $options->{'orig_b'};
        my $rel = $self->get_relationship( $source, $target );
        if( $rel ) {
-               if( $rel->type ne $options->{'type'} ) {
+               if( $rel->type eq 'collated' ) {
+                       # Always replace a 'collated' relationship with a more descriptive
+                       # one, if asked.
+                       $self->del_relationship( $source, $target );
+               } elsif( $rel->type ne $options->{'type'} ) {
                        throw( "Another relationship of type " . $rel->type 
                                . " already exists between $source and $target" );
                } else {
@@ -125,13 +163,16 @@ non-locally.  Key on whichever reading occurs first alphabetically.
 
 sub add_scoped_relationship {
        my( $self, $rel ) = @_;
-       my $r = $self->scoped_relationship( $rel->reading_a, $rel->reading_b );
+       my $rdga = $rel->type eq 'orthographic' ? $rel->reading_a : lc( $rel->reading_a );
+       my $rdgb = $rel->type eq 'orthographic' ? $rel->reading_b : lc( $rel->reading_b );      
+       my $r = $self->scoped_relationship( $rdga, $rdgb );
        if( $r ) {
                warn sprintf( "Scoped relationship of type %s already exists between %s and %s",
-                       $r->type, $rel->reading_a, $rel->reading_b );
+                       $r->type, $rdga, $rdgb );
                return;
        }
-       $self->scopedrels->{$rel->reading_a}->{$rel->reading_b} = $rel;
+       my( $first, $second ) = sort ( $rdga, $rdgb );
+       $self->scopedrels->{$first}->{$second} = $rel;
 }
 
 =head2 scoped_relationship( $reading_a, $reading_b )
@@ -161,69 +202,274 @@ scoped non-locally.
 Returns a status boolean and a list of all reading pairs connected by the call to
 add_relationship.
 
+=begin testing
+
+use Text::Tradition;
+use TryCatch;
+
+my $t1 = Text::Tradition->new( 'input' => 'Self', 'file' => 't/data/legendfrag.xml' );
+# Test 1: try to equate nodes that are prevented with an intermediate collation
+ok( $t1, "Parsed test fragment file" );
+my $c1 = $t1->collation;
+my $trel = $c1->get_relationship( '9,2', '9,3' );
+is( ref( $trel ), 'Text::Tradition::Collation::Relationship',
+       "Troublesome relationship exists" );
+is( $trel->type, 'collated', "Troublesome relationship is a collation" );
+
+# Try to make the link we want
+try {
+       $c1->add_relationship( '8,6', '10,3', { 'type' => 'orthographic' } );
+       ok( 1, "Added cross-collation relationship as expected" );
+} catch {
+       ok( 0, "Existing collation blocked equivalence relationship" );
+}
+
+try {
+       $c1->calculate_ranks();
+       ok( 1, "Successfully calculated ranks" );
+} catch {
+       ok( 0, "Collation now has a cycle" );
+}
+
+# Test 2: try to equate nodes that are prevented with a real intermediate
+# equivalence
+my $t2 = Text::Tradition->new( 'input' => 'Self', 'file' => 't/data/legendfrag.xml' );
+my $c2 = $t2->collation;
+$c2->add_relationship( '9,2', '9,3', { 'type' => 'lexical' } );
+my $trel2 = $c2->get_relationship( '9,2', '9,3' );
+is( ref( $trel2 ), 'Text::Tradition::Collation::Relationship',
+       "Created blocking relationship" );
+is( $trel2->type, 'lexical', "Blocking relationship is not a collation" );
+# This time the link ought to fail
+try {
+       $c2->add_relationship( '8,6', '10,3', { 'type' => 'orthographic' } );
+       ok( 0, "Added cross-equivalent bad relationship" );
+} catch {
+       ok( 1, "Existing equivalence blocked crossing relationship" );
+}
+
+try {
+       $c2->calculate_ranks();
+       ok( 1, "Successfully calculated ranks" );
+} catch {
+       ok( 0, "Collation now has a cycle" );
+}
+
+# Test 3: make a straightforward pair of transpositions.
+my $t3 = Text::Tradition->new( 'input' => 'Self', 'file' => 't/data/lf2.xml' );
+# Test 1: try to equate nodes that are prevented with an intermediate collation
+my $c3 = $t3->collation;
+try {
+       $c3->add_relationship( '36,4', '38,3', { 'type' => 'transposition' } );
+       ok( 1, "Added straightforward transposition" );
+} catch {
+       ok( 0, "Failed to add normal transposition" );
+}
+try {
+       $c3->add_relationship( '36,3', '38,2', { 'type' => 'transposition' } );
+       ok( 1, "Added straightforward transposition complement" );
+} catch {
+       ok( 0, "Failed to add normal transposition complement" );
+}
+
+# Test 4: try to make a transposition that could be a parallel.
+try {
+       $c3->add_relationship( '28,2', '29,2', { 'type' => 'transposition' } );
+       ok( 0, "Added bad colocated transposition" );
+} catch {
+       ok( 1, "Prevented bad colocated transposition" );
+}
+
+# Test 5: make the parallel, and then make the transposition again.
+try {
+       $c3->add_relationship( '28,3', '29,3', { 'type' => 'orthographic' } );
+       ok( 1, "Equated identical readings for transposition" );
+} catch {
+       ok( 0, "Failed to equate identical readings" );
+}
+try {
+       $c3->add_relationship( '28,2', '29,2', { 'type' => 'transposition' } );
+       ok( 1, "Added straightforward transposition complement" );
+} catch {
+       ok( 0, "Failed to add normal transposition complement" );
+}
+
+=end testing
+
 =cut
 
 sub add_relationship {
-       my( $self, $source, $source_rdg, $target, $target_rdg, $options ) = @_;
+       my( $self, $source, $target, $options ) = @_;
+    my $c = $self->collation;
 
-       # Check the options
-       $options->{'scope'} = 'local' unless $options->{'scope'};
-       
-       my( $is_valid, $reason ) = 
-               $self->relationship_valid( $source, $target, $options->{'type'} );
-    unless( $is_valid ) {
-        throw( "Invalid relationship: $reason" );
+       my $relationship;
+       my $thispaironly;
+       my $droppedcolls = [];
+       if( ref( $options ) eq 'Text::Tradition::Collation::Relationship' ) {
+               $relationship = $options;
+               $thispaironly = 1;  # If existing rel, set only where asked.
+       } else {
+               # Check the options
+               $options->{'scope'} = 'local' unless $options->{'scope'};
+               $options->{'scope'} = 'local' if $options->{'type'} eq 'collated';
+               $options->{'scope'} = 'local' if $options->{'type'} eq 'transposition';
+               
+               my( $is_valid, $reason ) = $self->relationship_valid( $source, $target, 
+                       $options->{'type'}, $droppedcolls );
+               unless( $is_valid ) {
+                       throw( "Invalid relationship: $reason" );
+               }
+               
+               # Try to create the relationship object.
+               $options->{'reading_a'} = $c->reading( $source )->text;
+               $options->{'reading_b'} = $c->reading( $target )->text;
+               $options->{'orig_a'} = $source;
+               $options->{'orig_b'} = $target;
+       if( $options->{'scope'} ne 'local' ) {
+                       # Is there a relationship with this a & b already?
+                       # Case-insensitive for non-orthographics.
+                       my $rdga = $options->{'type'} eq 'orthographic' 
+                               ? $options->{'reading_a'} : lc( $options->{'reading_a'} );
+                       my $rdgb = $options->{'type'} eq 'orthographic' 
+                               ? $options->{'reading_b'} : lc( $options->{'reading_b'} );
+                       my $otherrel = $self->scoped_relationship( $rdga, $rdgb );
+                       if( $otherrel && $otherrel->type eq $options->{type}
+                               && $otherrel->scope eq $options->{scope} ) {
+                               warn "Applying existing scoped relationship";
+                               $relationship = $otherrel;
+                       }
+       }
+               $relationship = $self->create( $options ) unless $relationship;  # Will throw on error
     }
-    
-    # Try to create the relationship object.
-    $options->{'reading_a'} = $source_rdg->text;
-    $options->{'reading_b'} = $target_rdg->text;
-    $options->{'orig_a'} = $source;
-    $options->{'orig_b'} = $target;
-    my $relationship = $self->create( $options );  # Will throw on error
+
 
        # Find all the pairs for which we need to set the relationship.
-       my @vectors = ( [ $source, $target ] ); 
-    if( $relationship->colocated && $relationship->nonlocal ) {
-       my $c = $self->collation;
-       # Set the same relationship everywhere we can, throughout the graph.
-       my @identical_readings = grep { $_->text eq $relationship->reading_a }
-               $c->readings;
-       foreach my $ir ( @identical_readings ) {
-               next if $ir->id eq $source;
-               # Check to see if there is a target reading with the same text at
-               # the same rank.
-               my @itarget = grep 
-                       { $_->rank == $ir->rank && $_->text eq $relationship->reading_b }
-                       $c->readings;
-               if( @itarget ) {
-                       # We found a hit.
-                       warn "More than one reading with text " . $target_rdg->text
-                               . " at rank " . $ir->rank . "!" if @itarget > 1;
-                       push( @vectors, [ $ir->id, $itarget[0]->id ] );
-               }
-       }       
+       my @vectors;
+    if( $relationship->colocated && $relationship->nonlocal && !$thispaironly ) {
+       push( @vectors, $self->_find_applicable( $relationship ) );
     }
-    
+        
     # Now set the relationship(s).
     my @pairs_set;
+       my $rel = $self->get_relationship( $source, $target );
+       my $skip;
+       if( $rel && $rel ne $relationship ) {
+               if( $rel->nonlocal ) {
+                       throw( "Found conflicting relationship at $source - $target" );
+               } elsif( $rel->type ne 'collated' ) {
+                       # Replace a collation relationship; leave any other sort in place.
+                       my $r1ann = $rel->has_annotation ? $rel->annotation : '';
+                       my $r2ann = $relationship->has_annotation ? $relationship->annotation : '';
+                       unless( $rel->type eq $relationship->type && $r1ann eq $r2ann ) {
+                               warn sprintf( "Not overriding local relationship %s with global %s " 
+                                       . "set at %s -> %s (%s -> %s)", $rel->type, $relationship->type,
+                                       $source, $target, $rel->reading_a, $rel->reading_b );
+                               $skip = 1;
+                       }
+               }
+       }
+       $self->_set_relationship( $relationship, $source, $target ) unless $skip;
+       push( @pairs_set, [ $source, $target ] );
+    
+    # Set any additional relationships that might be in @vectors.
     foreach my $v ( @vectors ) {
-               my $rel = $self->get_relationship( @$v );
-       if( $rel ) {
-               if( $rel->nonlocal ) {
-                       throw( "Found conflicting relationship at @$v" );
-               } else {
-                       warn "Not overriding local relationship set at @$v";
-               }
-               next;
-       }
-       $self->_set_relationship( $relationship, @$v );
-       push( @pairs_set, $v );
+       next if $v->[0] eq $source && $v->[1] eq $target;
+       next if $v->[1] eq $source && $v->[0] eq $target;
+       my @added = $self->add_relationship( @$v, $relationship );
+       push( @pairs_set, @added );
     }
     
+    # Finally, restore whatever collations we can, and return.
+    $self->_restore_collations( @$droppedcolls );
     return @pairs_set;
 }
 
+=head2 del_scoped_relationship( $reading_a, $reading_b )
+
+Returns the general (document-level or global) relationship that has been defined 
+between the two reading strings. Returns undef if there is no general relationship.
+
+=cut
+
+sub del_scoped_relationship {
+       my( $self, $rdga, $rdgb ) = @_;
+       my( $first, $second ) = sort( $rdga, $rdgb );
+       return delete $self->scopedrels->{$first}->{$second};
+}
+
+sub _find_applicable {
+       my( $self, $rel ) = @_;
+       my $c = $self->collation;
+       # TODO Someday we might use a case sensitive language.
+       my $lang = $c->tradition->language;
+       my @vectors;
+       my @identical_readings;
+       if( $rel->type eq 'orthographic' ) {
+               @identical_readings = grep { $_->text eq $rel->reading_a } 
+                       $c->readings;
+       } else {
+               @identical_readings = grep { lc( $_->text ) eq lc( $rel->reading_a ) }
+                       $c->readings;
+       }
+       foreach my $ir ( @identical_readings ) {
+               my @itarget;
+               if( $rel->type eq 'orthographic' ) {
+                       @itarget = grep { $_->rank == $ir->rank 
+                                                         && $_->text eq $rel->reading_b } $c->readings;
+               } else {
+                       @itarget = grep { $_->rank == $ir->rank 
+                                                         && lc( $_->text ) eq lc( $rel->reading_b ) } $c->readings;
+               }
+               if( @itarget ) {
+                       # Warn if there is more than one hit with no orth link between them.
+                       my $itmain = shift @itarget;
+                       if( @itarget ) {
+                               my %all_targets;
+                               map { $all_targets{$_} = 1 } @itarget;
+                               map { delete $all_targets{$_} } 
+                                       $self->related_readings( $itmain, 
+                                               sub { $_[0]->type eq 'orthographic' } );
+                       warn "More than one unrelated reading with text " . $itmain->text
+                               . " at rank " . $ir->rank . "!" if keys %all_targets;
+                       }
+                       push( @vectors, [ $ir->id, $itmain->id ] );
+               }
+       }
+       return @vectors;
+}
+
+=head2 del_relationship( $source, $target )
+
+Removes the relationship between the given readings. If the relationship is
+non-local, removes the relationship everywhere in the graph.
+
+=cut
+
+sub del_relationship {
+       my( $self, $source, $target ) = @_;
+       my $rel = $self->get_relationship( $source, $target );
+       return () unless $rel; # Nothing to delete; return an empty set.
+       my @vectors = ( [ $source, $target ] );
+       $self->_remove_relationship( $source, $target );
+       if( $rel->nonlocal ) {
+               # Remove the relationship wherever it occurs.
+               # Remove the relationship wherever it occurs.
+               my @rel_edges = grep { $self->get_relationship( @$_ ) == $rel }
+                       $self->relationships;
+               foreach my $re ( @rel_edges ) {
+                       $self->_remove_relationship( @$re );
+                       push( @vectors, $re );
+               }
+               $self->del_scoped_relationship( $rel->reading_a, $rel->reading_b );
+       }
+       return @vectors;
+}
+
+sub _remove_relationship {
+       my( $self, @vector ) = @_;
+       $self->graph->delete_edge( @vector );
+}
+       
 =head2 relationship_valid( $source, $target, $type )
 
 Checks whether a relationship of type $type may exist between the readings given
@@ -233,11 +479,13 @@ a yes/no boolean and, if the answer is no, message gives the reason why.
 =cut
 
 sub relationship_valid {
-    my( $self, $source, $target, $rel ) = @_;
+    my( $self, $source, $target, $rel, $mustdrop ) = @_;
+    $mustdrop = [] unless $mustdrop; # in case we were passed nothing
     my $c = $self->collation;
     if ( $rel eq 'transposition' || $rel eq 'repetition' ) {
                # Check that the two readings do (for a repetition) or do not (for
                # a transposition) appear in the same witness.
+               # TODO this might be called before witness paths are set...
                my %seen_wits;
                map { $seen_wits{$_} = 1 } $c->reading_witnesses( $source );
                foreach my $w ( $c->reading_witnesses( $target ) ) {
@@ -245,49 +493,144 @@ sub relationship_valid {
                                return ( 0, "Readings both occur in witness $w" ) 
                                        if $rel eq 'transposition';
                                return ( 1, "ok" ) if $rel eq 'repetition';
+                       }
                }
-               return $rel eq 'transposition' ? ( 1, "ok" )
-                       : ( 0, "Readings occur only in distinct witnesses" );
+               return ( 0, "Readings occur only in distinct witnesses" )
+                       if $rel eq 'repetition';
+       } 
+       if ( $rel eq 'transposition' ) {
+               # We also need to check both that the readings occur in distinct
+               # witnesses, and that they are not in the same place. That is,
+               # proposing to link them should cause a witness loop.
+               my $map = {};
+               my( $startrank, $endrank );
+               if( $c->end->has_rank ) {
+                       my $cpred = $c->common_predecessor( $source, $target );
+                       my $csucc = $c->common_successor( $source, $target );
+                       $startrank = $cpred->rank;
+                       $endrank = $csucc->rank;
                }
-       } else {
+               my $eqgraph = $c->equivalence_graph( $map, $startrank, $endrank, 
+                       $source, $target );
+               if( $eqgraph->has_a_cycle ) {
+                       return ( 1, "ok" );
+               } else {
+                       return ( 0, "Readings appear to be colocated, not transposed" );
+               }
+               
+       } elsif( $rel ne 'repetition' ) {
                # Check that linking the source and target in a relationship won't lead
-               # to a path loop for any witness.  First make a lookup table of all the
-               # readings related to either the source or the target.
-               my @proposed_related = ( $source, $target );
-               push( @proposed_related, $self->related_readings( $source, 'colocated' ) );
-               push( @proposed_related, $self->related_readings( $target, 'colocated' ) );
-               my %pr_ids;
-               map { $pr_ids{ $_ } = 1 } @proposed_related;
-       
-               # None of these proposed related readings should have a neighbor that
-               # is also in proposed_related.
-               foreach my $pr ( keys %pr_ids ) {
-                       foreach my $neighbor( $c->sequence->neighbors( $pr ) ) {
-                               return( 0, "Would relate neighboring readings $pr and $neighbor" )
-                                       if exists $pr_ids{$neighbor};
+               # to a path loop for any witness. 
+               # First, drop/stash any collations that might interfere
+               my $sourceobj = $c->reading( $source );
+               my $targetobj = $c->reading( $target );
+               my $sourcerank = $sourceobj->has_rank ? $sourceobj->rank : -1;
+               my $targetrank = $targetobj->has_rank ? $targetobj->rank : -1;
+               unless( $rel eq 'collated' || $sourcerank == $targetrank ) {
+                       push( @$mustdrop, $self->_drop_collations( $source ) );
+                       push( @$mustdrop, $self->_drop_collations( $target ) );
+               }
+               my $map = {};
+               my( $startrank, $endrank );
+               if( $c->end->has_rank ) {
+                       my $cpred = $c->common_predecessor( $source, $target );
+                       my $csucc = $c->common_successor( $source, $target );
+                       $startrank = $cpred->rank;
+                       $endrank = $csucc->rank;
+                       unless( $rel eq 'collated' || $sourcerank == $targetrank ) {
+                               foreach my $rk ( $startrank+1 .. $endrank-1 ) {
+                                       map { push( @$mustdrop, $self->_drop_collations( $_->id ) ) }
+                                               $c->readings_at_rank( $rk );
+                               }
                        }
-               }               
+               }
+               my $eqgraph = $c->equivalence_graph( $map, $startrank, $endrank, 
+                       $source, $target );
+               if( $eqgraph->has_a_cycle ) {
+                       $self->_restore_collations( @$mustdrop );
+                       return( 0, "Relationship would create witness loop" );
+               }
                return ( 1, "ok" );
        }
 }
 
-=head2 related_readings( $reading, $colocated_only )
+sub _drop_collations {
+       my( $self, $reading ) = @_;
+       my @dropped;
+       foreach my $n ( $self->graph->neighbors( $reading ) ) {
+               if( $self->get_relationship( $reading, $n )->type eq 'collated' ) {
+                       push( @dropped, [ $reading, $n ] );
+                       $self->del_relationship( $reading, $n );
+               }
+       }
+       return @dropped;
+}
+
+sub _restore_collations {
+       my( $self, @vectors ) = @_;
+       foreach my $v ( @vectors ) {
+               try {
+                       $self->add_relationship( @$v, { 'type' => 'collated' } );
+               } catch {
+                       print STDERR $v->[0] . " - " . $v->[1] . " no longer collate\n";
+               }
+       }
+}
+
+=head2 filter_collations()
+
+Utility function. Removes any redundant 'collated' relationships from the graph.
+A collated relationship is redundant if the readings in question would occupy
+the same rank regardless of the existence of the relationship.
+
+=cut
+
+sub filter_collations {
+       my $self = shift;
+       my $c = $self->collation;
+       foreach my $r ( 1 .. $c->end->rank - 1 ) {
+               my $anchor;
+               my @need_collations;
+               foreach my $rdg ( $c->readings_at_rank( $r ) ) {
+                       next if $rdg->is_meta;
+                       my $ip = 0;
+                       foreach my $pred ( $rdg->predecessors ) {
+                               if( $pred->rank == $r - 1 ) {
+                                       $ip = 1;
+                                       $anchor = $rdg unless( $anchor );
+                                       last;
+                               }
+                       }
+                       push( @need_collations, $rdg ) unless $ip;
+                       $c->relations->_drop_collations( "$rdg" );
+               }
+               $anchor
+                       ? map { $c->add_relationship( $anchor, $_, { 'type' => 'collated' } ) } @need_collations
+                       : warn "No anchor found at $r";
+       }
+}
+
+=head2 related_readings( $reading, $filter )
 
 Returns a list of readings that are connected via relationship links to $reading.
-If $colocated_only is true, restricts the list to those readings that are in the
-same logical location (and therefore have the same rank in the collation graph.)
+If $filter is set to a subroutine ref, returns only those related readings where
+$filter( $relationship ) returns a true value.
 
 =cut
 
 sub related_readings {
-       my( $self, $reading, $colocated ) = @_;
+       my( $self, $reading, $filter ) = @_;
        my $return_object;
        if( ref( $reading ) eq 'Text::Tradition::Collation::Reading' ) {
                $reading = $reading->id;
                $return_object = 1;
        }
        my @answer;
-       if( $colocated ) {
+       if( $filter ) {
+               # Backwards compat
+               if( $filter eq 'colocated' ) {
+                       $filter = sub { $_[0]->colocated };
+               }
                my %found = ( $reading => 1 );
                my $check = [ $reading ];
                my $iter = 0;
@@ -295,7 +638,7 @@ sub related_readings {
                        my $more = [];
                        foreach my $r ( @$check ) {
                                foreach my $nr ( $self->graph->neighbors( $r ) ) {
-                                       if( $self->get_relationship( $r, $nr )->colocated ) {
+                                       if( &$filter( $self->get_relationship( $r, $nr ) ) ) {
                                                push( @$more, $nr ) unless exists $found{$nr};
                                                $found{$nr} = 1;
                                        }
@@ -303,6 +646,7 @@ sub related_readings {
                        }
                        $check = $more;
                }
+               delete $found{$reading};
                @answer = keys %found;
        } else {
                @answer = $self->graph->all_reachable( $reading );
@@ -333,12 +677,9 @@ sub merge_readings {
                # If kept changes its text, drop the relationship.
                next if $combined;
                        
-               # If kept / rel already has a relationship, warn and keep the old
+               # If kept / rel already has a relationship, just keep the old
                my $rel = $self->get_relationship( @vector );
-               if( $rel ) {
-                       warn sprintf( "Readings %s and %s have existing relationship; dropping link with %s", @vector, $deleted );
-                       next;
-               }
+               next if $rel;
                
                # Otherwise, adopt the relationship that would be deleted.
                $rel = $self->get_relationship( @$edge );
@@ -354,36 +695,40 @@ sub _as_graphml {
        $rgraph->setAttribute( 'edgedefault', 'directed' );
     $rgraph->setAttribute( 'id', 'relationships', );
     $rgraph->setAttribute( 'parse.edgeids', 'canonical' );
-    $rgraph->setAttribute( 'parse.edges', scalar($self->graph->edges) );
+    $rgraph->setAttribute( 'parse.edges', 0 );
     $rgraph->setAttribute( 'parse.nodeids', 'canonical' );
-    $rgraph->setAttribute( 'parse.nodes', scalar($self->graph->vertices) );
+    $rgraph->setAttribute( 'parse.nodes', 0 );
     $rgraph->setAttribute( 'parse.order', 'nodesfirst' );
     
     # Add the vertices according to their XML IDs
     my %rdg_lookup = ( reverse %$node_hash );
-    foreach my $n ( sort _by_xmlid keys( %rdg_lookup ) ) {
+    # my @nlist = sort _by_xmlid keys( %rdg_lookup ); ## CAUSES SEGFAULT
+    my @nlist = sort keys( %rdg_lookup );
+    foreach my $n ( @nlist ) {
        my $n_el = $rgraph->addNewChild( $graphml_ns, 'node' );
        $n_el->setAttribute( 'id', $n );
        _add_graphml_data( $n_el, $nodeid_key, $rdg_lookup{$n} );
     }
+       $rgraph->setAttribute( 'parse.nodes', scalar @nlist );
     
     # Add the relationship edges, with their object information
     my $edge_ctr = 0;
     foreach my $e ( sort { $a->[0] cmp $b->[0] } $self->graph->edges ) {
        # Add an edge and fill in its relationship info.
+       next unless( exists $node_hash->{$e->[0]} && exists $node_hash->{$e->[1]} );
                my $edge_el = $rgraph->addNewChild( $graphml_ns, 'edge' );
                $edge_el->setAttribute( 'source', $node_hash->{$e->[0]} );
                $edge_el->setAttribute( 'target', $node_hash->{$e->[1]} );
                $edge_el->setAttribute( 'id', 'e'.$edge_ctr++ );
 
                my $rel_obj = $self->get_relationship( @$e );
-               _add_graphml_data( $edge_el, $edge_keys->{'relationship'}, $rel_obj->type );
-               _add_graphml_data( $edge_el, $edge_keys->{'scope'}, $rel_obj->scope );
-               _add_graphml_data( $edge_el, $edge_keys->{'non_correctable'}, 
-                       $rel_obj->non_correctable ) if $rel_obj->noncorr_set;
-               _add_graphml_data( $edge_el, $edge_keys->{'non_independent'}, 
-                       $rel_obj->non_independent ) if $rel_obj->nonind_set;
+               foreach my $key ( keys %$edge_keys ) {
+                       my $value = $rel_obj->$key;
+                       _add_graphml_data( $edge_el, $edge_keys->{$key}, $value ) 
+                               if defined $value;
+               }
        }
+       $rgraph->setAttribute( 'parse.edges', $edge_ctr );
 }
 
 sub _by_xmlid {