X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FText%2FTradition%2FCollation%2FRelationshipStore.pm;h=3d98174fa5420692e639e73299ee8fdacd5838f1;hb=13e893dc2c0d9e482e4a8576d7c7d59e88866669;hp=2479d08193e92240f2e3f163bf4c790b7d100e79;hpb=83d5ac3af5c70f09ee081885efcba827fcef4cb4;p=scpubgit%2Fstemmatology.git diff --git a/lib/Text/Tradition/Collation/RelationshipStore.pm b/lib/Text/Tradition/Collation/RelationshipStore.pm index 2479d08..3d98174 100644 --- a/lib/Text/Tradition/Collation/RelationshipStore.pm +++ b/lib/Text/Tradition/Collation/RelationshipStore.pm @@ -2,13 +2,16 @@ package Text::Tradition::Collation::RelationshipStore; use strict; use warnings; +use Text::Tradition::Error; use Text::Tradition::Collation::Relationship; +use TryCatch; use Moose; =head1 NAME -Text::Tradition::Collation::Reading - represents a reading (usually a word) in a collation. +Text::Tradition::Collation::RelationshipStore - Keeps track of the relationships +between readings in a given collation =head1 DESCRIPTION @@ -17,6 +20,39 @@ texts, particularly medieval ones. The RelationshipStore is an internal object of the collation, to keep track of the defined relationships (both specific and 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 =head2 new( collation => $collation ); @@ -49,6 +85,35 @@ has 'graph' => ( }, ); +=head2 get_relationship + +Return the relationship object, if any, that exists between two readings. + +=cut + +sub get_relationship { + 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; +} + +sub _set_relationship { + my( $self, $relationship, @vector ) = @_; + $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. @@ -61,13 +126,15 @@ sub create { # Check to see if a relationship exists between the two given readings my $source = delete $options->{'orig_a'}; my $target = delete $options->{'orig_b'}; - my $rel; - if( $self->graph->has_edge( $source, $target ) ) { - $rel = $self->graph->get_edge_attribute( $source, $target, 'object' ); - if( $rel->type ne $options->type ) { - warn "Relationship of type " . $rel->type - . "already exists between $source and $target"; - return; + my $rel = $self->get_relationship( $source, $target ); + if( $rel ) { + 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 { return $rel; } @@ -79,8 +146,7 @@ sub create { if( $rel && $rel->type eq $options->{'type'} ) { return $rel; } elsif( $rel ) { - warn sprintf( "Relationship of type %s with scope %s already defined for readings %s and %s", $rel->type, $rel->scope, $options->{'reading_a'}, $options->{'reading_b'} ); - return; + throw( sprintf( "Relationship of type %s with scope %s already defined for readings %s and %s", $rel->type, $rel->scope, $options->{'reading_a'}, $options->{'reading_b'} ) ); } else { $rel = Text::Tradition::Collation::Relationship->new( $options ); $self->add_scoped_relationship( $rel ) if $rel->nonlocal; @@ -97,13 +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 ) @@ -138,68 +207,165 @@ add_relationship. sub add_relationship { my( $self, $source, $source_rdg, $target, $target_rdg, $options ) = @_; - # Check the options - $options->{'scope'} = 'local' unless $options->{'scope'}; - - my( $is_valid, $reason ) = - $self->relationship_valid( $source, $target, $options->{'type'} ); - unless( $is_valid ) { - return ( undef, $reason ); + my $relationship; + my $thispaironly; + 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'} ); + unless( $is_valid ) { + throw( "Invalid relationship: $reason" ); + } + + # 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; + 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 ); - return( undef, "Relationship creation failed" ) unless $relationship; + # 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 = [ $source, $target ]; + if( $relationship->colocated && $relationship->nonlocal && !$thispaironly ) { + push( @vectors, $self->_find_applicable( $relationship ) ); } - + # Now set the relationship(s). my @pairs_set; foreach my $v ( @vectors ) { - if( $self->graph->has_edge( @$v ) ) { - # Is it locally scoped? - my $rel = $self->graph->get_edge_attribute( @$v ); + my $rel = $self->get_relationship( @$v ); + if( $rel && $rel ne $relationship ) { if( $rel->nonlocal ) { - # TODO I think we should not be able to get here. - warn "Found conflicting relationship at @$v"; - } else { - warn "Not overriding local relationship set at @$v"; - next; + throw( "Found conflicting relationship at @$v" ); + } 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, $relationship->type, + @$v, $rel->reading_a, $rel->reading_b ); + next; + } } } - $self->graph->add_edge( @$v ); - $self->graph->set_edge_attribute( @$v, 'object', $relationship ); + map { $self->_drop_collations( $_ ) } @$v; + $self->_set_relationship( $relationship, @$v ); push( @pairs_set, $v ); } - return( 1, @pairs_set ); + 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 @@ -214,6 +380,10 @@ sub relationship_valid { 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. + # If we haven't made reading paths yet, take it on faith. + return( 1, "no paths yet" ) unless $c->sequence->successors( $c->start ); + + # We have some paths, so carry on. my %seen_wits; map { $seen_wits{$_} = 1 } $c->reading_witnesses( $source ); foreach my $w ( $c->reading_witnesses( $target ) ) { @@ -221,61 +391,154 @@ sub relationship_valid { return ( 0, "Readings both occur in witness $w" ) if $rel eq 'transposition'; return ( 1, "ok" ) if $rel eq 'repetition'; + } + } + # For transpositions, there should also be a path from one reading + # to the other. + if( $rel eq 'transposition' ) { + my( %sourceseq, %targetseq ); + map { $sourceseq{$_} = 1 } $c->sequence->all_successors( $source ); + map { $targetseq{$_} = 1 } $c->sequence->all_successors( $target ); + return( 0, "Readings are parallel" ) + unless $sourceseq{$target} || $targetseq{$source}; } return $rel eq 'transposition' ? ( 1, "ok" ) : ( 0, "Readings occur only in distinct witnesses" ); - } - } else { + } + if( $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 + # to a path loop for any witness. If they have the same rank then + # they are parallel by definition. + # For transpositions, we want the opposite result: it is only valid if + # the readings cannot be parallel. + my $sourcerank = $c->reading( $source )->has_rank + ? $c->reading( $source )->rank : undef; + my $targetrank = $c->reading( $target )->has_rank + ? $c->reading( $target )->rank : undef; + if( $sourcerank && $targetrank && $sourcerank == $targetrank ) { + return( 0, "Cannot transpose readings of same rank" ) + if $rel eq 'transposition'; + return( 1, "ok" ); + } + + # Otherwise, 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' ) ); + # Drop the collation links of source and target, unless we want to + # add a collation relationship. + my @dropped; + foreach my $r ( ( $source, $target ) ) { + push( @dropped, $self->_drop_collations( $r ) ) + unless $rel eq 'collated'; + push( @proposed_related, $self->related_readings( $r, 'colocated' ) ); + } + # Also drop any collation links at intermediate ranks. + foreach my $rank ( $sourcerank+1 .. $targetrank-1 ) { + map { push( @dropped, $self->_drop_collations( $_ ) ) } + $c->readings_at_rank( $rank ); + } 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. + # The cumulative predecessors and successors of the proposed-related readings + # should not overlap. + my %all_pred; + my %all_succ; 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}; + map { $all_pred{$_} = 1 } $c->sequence->all_predecessors( $pr ); + map { $all_succ{$_} = 1 } $c->sequence->all_successors( $pr ); + } + foreach my $k ( keys %all_pred ) { + if( exists $all_succ{$k} ) { + $self->_restore_collations( @dropped ); + return( 1, "ok" ) if $rel eq 'transposition'; + return( 0, "Relationship would create witness loop" ); + } + } + foreach my $k ( keys %pr_ids ) { + if( exists $all_pred{$k} || exists $all_succ{$k} ) { + $self->_restore_collations( @dropped ); + return( 1, "ok" ) if $rel eq 'transposition'; + return( 0, "Relationship would create witness loop" ); } - } + } + if( $rel eq 'transposition' ) { + $self->_restore_collations( @dropped ); + return ( 0, "Cannot transpose parallel readings" ); + } return ( 1, "ok" ); } } -=head2 related_readings( $reading, $colocated_only ) +sub _drop_collations { + my( $self, $reading ) = @_; + my @deleted; + foreach my $n ( $self->graph->neighbors( $reading ) ) { + if( $self->get_relationship( $reading, $n )->type eq 'collated' ) { + $self->del_relationship( $reading, $n ); + push( @deleted, [ $reading, $n ] ); + } + } + return @deleted; +} + +sub _restore_collations { + my( $self, @vectors ) = @_; + foreach my $v ( @vectors ) { + try { + $self->add_relationship( @$v, { 'type' => 'collated' } ); + } catch ( Text::Tradition::Error $e ) { + warn "Could not restore collation " . join( ' -> ', @$v ); + } + } +} + +=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 @related = $self->graph->all_reachable( $reading ); - if( $colocated ) { - my @colo; - foreach my $r ( @related ) { - my $obj = $self->graph->get_edge_attribute( $reading, $r, 'object' ); - push( @colo, $r ) if $obj->colocated; + my @answer; + if( $filter ) { + # Backwards compat + if( $filter eq 'colocated' ) { + $filter = sub { $_[0]->colocated }; + } + my %found = ( $reading => 1 ); + my $check = [ $reading ]; + my $iter = 0; + while( @$check ) { + my $more = []; + foreach my $r ( @$check ) { + foreach my $nr ( $self->graph->neighbors( $r ) ) { + if( &$filter( $self->get_relationship( $r, $nr ) ) ) { + push( @$more, $nr ) unless exists $found{$nr}; + $found{$nr} = 1; + } + } + } + $check = $more; } - @related = @colo; + delete $found{$reading}; + @answer = keys %found; + } else { + @answer = $self->graph->all_reachable( $reading ); } if( $return_object ) { my $c = $self->collation; - return map { $c->reading( $_ ) } @related; + return map { $c->reading( $_ ) } @answer; } else { - return @related; + return @answer; } } @@ -288,6 +551,8 @@ stops tracking the to-be-deleted reading. sub merge_readings { my( $self, $kept, $deleted, $combined ) = @_; + # Delete any relationship between kept and deleted + $self->del_relationship( $kept, $deleted ); foreach my $edge ( $self->graph->edges_at( $deleted ) ) { # Get the pair of kept / rel my @vector = ( $kept ); @@ -297,22 +562,76 @@ 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( $self->graph->has_edge( @vector ) ) { - warn sprintf( "Readings %s and %s have existing relationship; dropping link with %s", @vector, $deleted ); - next; - } + # If kept / rel already has a relationship, just keep the old + my $rel = $self->get_relationship( @vector ); + next if $rel; # Otherwise, adopt the relationship that would be deleted. - my $rel = $self->graph->get_edge_attribute( @$edge, 'object' ); - $self->graph->add_edge( @vector ); - $self->graph->set_edge_attribute( @vector, 'object', $rel ); + $rel = $self->get_relationship( @$edge ); + $self->_set_relationship( $rel, @vector ); } $self->delete_reading( $deleted ); } -sub as_graphml { ## TODO - return; +sub _as_graphml { + my( $self, $graphml_ns, $xmlroot, $node_hash, $nodeid_key, $edge_keys ) = @_; + + my $rgraph = $xmlroot->addNewChild( $graphml_ns, 'graph' ); + $rgraph->setAttribute( 'edgedefault', 'directed' ); + $rgraph->setAttribute( 'id', 'relationships', ); + $rgraph->setAttribute( 'parse.edgeids', 'canonical' ); + $rgraph->setAttribute( 'parse.edges', scalar($self->graph->edges) ); + $rgraph->setAttribute( 'parse.nodeids', 'canonical' ); + $rgraph->setAttribute( 'parse.nodes', scalar($self->graph->vertices) ); + $rgraph->setAttribute( 'parse.order', 'nodesfirst' ); + + # Add the vertices according to their XML IDs + my %rdg_lookup = ( reverse %$node_hash ); + foreach my $n ( sort _by_xmlid keys( %rdg_lookup ) ) { + my $n_el = $rgraph->addNewChild( $graphml_ns, 'node' ); + $n_el->setAttribute( 'id', $n ); + _add_graphml_data( $n_el, $nodeid_key, $rdg_lookup{$n} ); + } + + # 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. + 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 ); + foreach my $key ( keys %$edge_keys ) { + my $value = $rel_obj->$key; + _add_graphml_data( $edge_el, $edge_keys->{$key}, $value ) + if defined $value; + } + } +} + +sub _by_xmlid { + my $tmp_a = $a; + my $tmp_b = $b; + $tmp_a =~ s/\D//g; + $tmp_b =~ s/\D//g; + return $tmp_a <=> $tmp_b; +} + +sub _add_graphml_data { + my( $el, $key, $value ) = @_; + return unless defined $value; + my $data_el = $el->addNewChild( $el->namespaceURI, 'data' ); + $data_el->setAttribute( 'key', $key ); + $data_el->appendText( $value ); +} + +sub throw { + Text::Tradition::Error->throw( + 'ident' => 'Relationship error', + 'message' => $_[0], + ); } no Moose;