From 39e37f6b279b0aaa2e9737542bde8aa654a2d9d4 Mon Sep 17 00:00:00 2001 From: Julian Maurice Date: Fri, 27 Jul 2012 12:30:41 +0200 Subject: [PATCH 2/2] Bug 7419: New script dedup_authorities.pl This script allow to deduplicate authorities automatically. It works this way: 1) authorities are fetched from the database. You can limit fetched results by authtypecode, or directly by specifying WHERE clause 2) for each authority: 2.1) build a Zebra query, depending on --match parameter(s) 2.2) run the query, retrieve the results 2.3) for each result check if result is really a duplicate. check is customizable with option --check. 2.4) among real duplicates, choose the one we want to keep (use --choose-method option). 2.5) use C4::Authorities::merge to merge authorities 3) index updated biblios and authorities, and remove from index removed authorities Use --help for more informations on options Script is in misc/migration_tools/ --- C4/AuthoritiesMarc.pm | 114 ++---- misc/migration_tools/dedup_authorities.pl | 584 +++++++++++++++++++++++++++++ 2 files changed, 617 insertions(+), 81 deletions(-) create mode 100755 misc/migration_tools/dedup_authorities.pl diff --git a/C4/AuthoritiesMarc.pm b/C4/AuthoritiesMarc.pm index 90c253a..8b44891 100644 --- a/C4/AuthoritiesMarc.pm +++ b/C4/AuthoritiesMarc.pm @@ -1302,7 +1302,7 @@ Then we should add some new parameter : bibliotargettag, authtargettag sub merge { my ($mergefrom,$MARCfrom,$mergeto,$MARCto) = @_; - my ($counteditedbiblio,$countunmodifiedbiblio,$counterrors)=(0,0,0); + my @editedbiblios; my $dbh=C4::Context->dbh; my $authtypecodefrom = GetAuthTypeCode($mergefrom); my $authtypecodeto = GetAuthTypeCode($mergeto); @@ -1316,17 +1316,17 @@ sub merge { my ($auth_tag_to_report_from) = $sth->fetchrow; $sth->execute($authtypecodeto); my ($auth_tag_to_report_to) = $sth->fetchrow; - + my @record_to; @record_to = $MARCto->field($auth_tag_to_report_to)->subfields() if $MARCto->field($auth_tag_to_report_to); my @record_from; @record_from = $MARCfrom->field($auth_tag_to_report_from)->subfields() if $MARCfrom->field($auth_tag_to_report_from); - + my @reccache; # search all biblio tags using this authority. #Getting marcbiblios impacted by the change. if (C4::Context->preference('NoZebra')) { - #nozebra way + #nozebra way my $dbh=C4::Context->dbh; my $rq=$dbh->prepare(qq(SELECT biblionumbers from nozebra where indexname="an" and server="biblioserver" and value="$mergefrom" )); $rq->execute; @@ -1375,41 +1375,41 @@ sub merge { while (my ($tagfield) = $sth->fetchrow) { push @tags_using_authtype,$tagfield ; } - my $tag_to=0; - if ($authtypecodeto ne $authtypecodefrom){ + my $tag_to=0; + if ($authtypecodeto ne $authtypecodefrom){ # If many tags, take the first - $sth->execute($authtypecodeto); + $sth->execute($authtypecodeto); $tag_to=$sth->fetchrow; - #warn $tag_to; - } + } # BulkEdit marc records - # May be used as a template for a bulkedit field + # May be used as a template for a bulkedit field foreach my $marcrecord(@reccache){ - my $update; + my $update; foreach my $tagfield (@tags_using_authtype){ -# warn "tagfield : $tagfield "; foreach my $field ($marcrecord->field($tagfield)){ my $auth_number=$field->subfield("9"); - my $tag=$field->tag(); + my $tag=$field->tag(); if ($auth_number==$mergefrom) { - my $field_to=MARC::Field->new(($tag_to?$tag_to:$tag),$field->indicator(1),$field->indicator(2),"9"=>$mergeto); - my $exclude='9'; - foreach my $subfield (@record_to) { - $field_to->add_subfields($subfield->[0] =>$subfield->[1]); - $exclude.= $subfield->[0]; - } - $exclude='['.$exclude.']'; -# add subfields in $field not included in @record_to - my @restore= grep {$_->[0]!~/$exclude/} $field->subfields(); - foreach my $subfield (@restore) { - $field_to->add_subfields($subfield->[0] =>$subfield->[1]); - } - $marcrecord->delete_field($field); - $marcrecord->insert_grouped_field($field_to); - $update=1; + my $field_to = MARC::Field->new(($tag_to?$tag_to:$tag), + $field->indicator(1),$field->indicator(2),"9"=>$mergeto); + my $exclude='9'; + foreach my $subfield (@record_to) { + $field_to->add_subfields($subfield->[0] =>$subfield->[1]); + $exclude.= $subfield->[0]; + } + $exclude='['.$exclude.']'; + #add subfields in $field not included in @record_to + my @restore= grep {$_->[0]!~/$exclude/} $field->subfields(); + foreach my $subfield (@restore) { + $field_to->add_subfields($subfield->[0] =>$subfield->[1]); + } + $marcrecord->delete_field($field); + $marcrecord->insert_grouped_field($field_to); + $update=1; } }#for each tag }#foreach tagfield + my ($bibliotag,$bibliosubf) = GetMarcFromKohaField("biblio.biblionumber","") ; my $biblionumber; if ($bibliotag<10){ @@ -1424,60 +1424,12 @@ sub merge { } if ($update==1){ &ModBiblio($marcrecord,$biblionumber,GetFrameworkCode($biblionumber)) ; - $counteditedbiblio++; - warn $counteditedbiblio if (($counteditedbiblio % 10) and $ENV{DEBUG}); - } + push @editedbiblios, $biblionumber; + } }#foreach $marc - return $counteditedbiblio; - # now, find every other authority linked with this authority - # now, find every other authority linked with this authority -# my $oConnection=C4::Context->Zconn("authorityserver"); -# my $query; -# # att 9210 Auth-Internal-authtype -# # att 9220 Auth-Internal-LN -# # ccl.properties to add for authorities -# $query= "= ".$mergefrom; -# my $oResult = $oConnection->search(new ZOOM::Query::CCL2RPN( $query, $oConnection )); -# my $count=$oResult->size() if ($oResult); -# my @reccache; -# my $z=0; -# while ( $z<$count ) { -# my $rec; -# $rec=$oResult->record($z); -# my $marcdata = $rec->raw(); -# push @reccache, $marcdata; -# $z++; -# } -# $oResult->destroy(); -# foreach my $marc(@reccache){ -# my $update; -# my $marcrecord; -# $marcrecord = MARC::File::USMARC::decode($marc); -# foreach my $tagfield (@tags_using_authtype){ -# $tagfield=substr($tagfield,0,3); -# my @tags = $marcrecord->field($tagfield); -# foreach my $tag (@tags){ -# my $tagsubs=$tag->subfield("9"); -# #warn "$tagfield:$tagsubs:$mergefrom"; -# if ($tagsubs== $mergefrom) { -# $tag->update("9" =>$mergeto); -# foreach my $subfield (@record_to) { -# # warn "$subfield,$subfield->[0],$subfield->[1]"; -# $tag->update($subfield->[0] =>$subfield->[1]); -# }#for $subfield -# } -# $marcrecord->delete_field($tag); -# $marcrecord->add_fields($tag); -# $update=1; -# }#for each tag -# }#foreach tagfield -# my $authoritynumber = TransformMarcToKoha($dbh,$marcrecord,"") ; -# if ($update==1){ -# &ModAuthority($marcrecord,$authoritynumber,GetAuthTypeCode($authoritynumber)) ; -# } -# -# }#foreach $marc -}#sub + + return @editedbiblios; +} =head2 get_auth_type_location diff --git a/misc/migration_tools/dedup_authorities.pl b/misc/migration_tools/dedup_authorities.pl new file mode 100755 index 0000000..47ccda4 --- /dev/null +++ b/misc/migration_tools/dedup_authorities.pl @@ -0,0 +1,584 @@ +#!/usr/bin/perl +use Modern::Perl; +use C4::Context; +use C4::AuthoritiesMarc; +use C4::Biblio; +use C4::Search; +use C4::Charset; +use C4::Debug; + +use Getopt::Long; +use YAML; +use List::MoreUtils qw/uniq/; + +my @matchstrings; +my $choosemethod = "u"; # by default, choose to keep the most used authority +my ($verbose, $all, $help, $wherestring, $test); + +# default is to check all these subfields for 'auth_tag_to_report' +my $check = 'TAGabcdefghitxyz'; + +my $result = GetOptions ( + "match=s" => \@matchstrings + , "verbose+" => \$verbose + , "all|a" => \$all + , "test|t" => \$test + , "help|h" => \$help + , "where=s" => \$wherestring + , "choose-method=s" => \$choosemethod + , "check=s" => \$check +); + +if ( $help || ! (@ARGV || $all)) { + print_usage(); + exit 0; +} + +my @choose_subs; +foreach (split //, $choosemethod) { + given($_) { + when ('d') { + push @choose_subs, \&_get_date; + } + when ('p') { + push @choose_subs, \&_has_ppn; + } + when ('u') { + push @choose_subs, \&_get_usage; + } + default { + warn "Choose method '$_' is not supported"; + } + } +} + +my @checks; +foreach my $c (split /,/, $check) { + my $field = substr $c, 0, 3; + my $subfields = substr $c, 3; + if ($field =~ /^[0-9][0-9.]{2}$/ or $field eq "TAG") { + my @subf = grep /^[0-9a-z]$/, uniq split //, $subfields; + if (@subf == 0 and not MARC::Field->is_controlfield_tag($field)) { + die "Field '$field' is not a control field and no subfields are given"; + } + if (@subf > 0 and MARC::Field->is_controlfield_tag($field)) { + die "Field '$field' is a control field but you have given subfields"; + } + push @checks, { + field => $field, + subfields => \@subf + }; + } else { + die "'$field' is not a valid tag"; + } +} + +my $dbh = C4::Context->dbh; + +$verbose and logger("Fetching authtypecodes..."); +my $authtypes_ref = $dbh->selectcol_arrayref( + qq{SELECT authtypecode, auth_tag_to_report FROM auth_types}, + { Columns=>[1,2] } +); +my %authtypes=@$authtypes_ref; +$verbose and logger("Fetching authtypecodes done."); + +my %biblios; +my %authorities; +my @authtypecodes = @ARGV ; +@authtypecodes = keys %authtypes unless(@authtypecodes); + +unless (@matchstrings) { + @matchstrings = ('at/152b##he/2..abxyz'); +} + +$verbose and logger("Preparing matchstrings..."); +my @attempts = prepare_matchstrings(@matchstrings); +$verbose and logger("Preparing matchstrings done."); + +for my $authtypecode (@authtypecodes) { + $verbose and logger("Deduping authtype '$authtypecode'"); + $verbose and logger("Fetching all authids for '$authtypecode'... "); + my @sqlparams; + my $strselect + = qq{SELECT authid, NULL FROM auth_header where authtypecode=?}; + push @sqlparams, $authtypecode; + if ($wherestring) { + $strselect .= " and $wherestring"; + } + my $auth_ref + = $dbh->selectcol_arrayref( $strselect, { Columns => [ 1, 2 ] }, + @sqlparams ); + $verbose and logger("Fetching authids done."); + my $auth_tag = $authtypes{$authtypecode}; + my %hash_authorities_authtypecode; + if ($auth_ref){ + %hash_authorities_authtypecode=@$auth_ref; + $verbose and logger("Building authorities hash..."); + @authorities{keys %hash_authorities_authtypecode} = values %hash_authorities_authtypecode; + $verbose and logger("Building authorities hash done."); + } + $verbose and logger("Start deduping for authtype '$authtypecode'"); + my $size = scalar keys %hash_authorities_authtypecode; + my $i = 1; + for my $authid ( keys %hash_authorities_authtypecode ) { + if ($verbose >= 2) { + my $percentage = sprintf("%.2f", $i * 100 / $size); + logger("Processing authority $authid ($i/$size $percentage%)"); + } elsif ($verbose and ($i % 100) == 0) { + my $percentage = sprintf("%.2f", $i * 100 / $size); + logger("Progression for authtype '$authtypecode': $i/$size ($percentage%)"); + } + + #authority was marked as duplicate + next if $authorities{$authid}; + my $authrecord = GetAuthority($authid); + + # next if cannot take authority + next unless $authrecord; + SetUTF8Flag($authrecord); + my $success; + for my $attempt (@attempts) { + ($verbose >= 2) and logger("Building query..."); + my $query = _build_query( $authrecord, $attempt ); + ($verbose >= 2) and logger("Building query done."); + $debug and $query and warn $query; + # This prevent too general queries to be executed + # TODO: This should allow more than these 3 indexes + next unless $query =~ /(he|he-main|ident)(,\S*)*(=|:)/; + + ($verbose >= 2) and logger("Searching..."); + my ($error, $results) = SimpleSearch($query, undef, undef, ["authorityserver"]); + if ( $error || !$results ) { + $debug and warn $@; + $debug and warn $error; + $debug and warn YAML::Dump($query); + $debug and warn $auth_tag; + next; + } + $debug and warn YAML::Dump($results); + next if ( !$results or scalar( @$results ) < 1 ); + my @recordids = map { + _get_id( MARC::Record->new_from_usmarc($_) ) + } @$results; + ($verbose >= 2) and logger("Searching done."); + + ($verbose >= 2) and logger("Choosing records..."); + my ( $recordid_to_keep, @recordids_to_merge ) + = _choose_records( $authid, @recordids ); + ($verbose >= 2) and logger("Choosing records done."); + unless ($test or @recordids_to_merge == 0) { + ($verbose >= 2) and logger("Merging ". join(',',@recordids_to_merge) ." into $recordid_to_keep."); + for my $localauthid (@recordids_to_merge) { + my @editedbiblios = eval { + merge( $localauthid, undef, $recordid_to_keep, undef ) + }; + if ($@) { + warn "merging $localauthid into $recordid_to_keep failed :", $@; + } else { + for my $biblio (@editedbiblios){ + $biblios{$biblio} = 1; + } + $authorities{$localauthid} = 2; + } + } + ($verbose >= 2) and logger("Merge done."); + $authorities{$recordid_to_keep} = 1; + } elsif ($verbose >= 2) { + if (@recordids_to_merge > 0) { + logger('Would merge ' + . join(',', @recordids_to_merge) + . " into $recordid_to_keep."); + } else { + logger("No duplicates found for $recordid_to_keep"); + } + } + } + $i++; + } + $verbose and logger("End of deduping for authtype '$authtypecode'"); +} + +# Update biblios +my @biblios_to_update = grep {defined $biblios{$_} and $biblios{$_} == 1} + keys %biblios; +if( @biblios_to_update > 0 ) { + logger("Updating biblios index (" . scalar(@biblios_to_update) . " biblios to update)"); + C4::Search::IndexRecord("biblio", \@biblios_to_update, + { as_xml => 1, record_format => "marcxml" }); +} else { + logger("No biblios to update"); +} + +# Update authorities +my @authorities_to_update = grep{defined $authorities{$_} and $authorities{$_} == 1} + keys %authorities; +if( @authorities_to_update > 0 ) { + logger("Updating authorities index (" . scalar(@authorities_to_update) . " authorities to update)"); + C4::Search::IndexRecord("authority", \@authorities_to_update, + { as_xml => 1, record_format => "marcxml" }); +} else { + logger("No autorities to update"); +} + +# Delete authorities +my @authorities_to_delete = grep{defined $authorities{$_} and $authorities{$_} == 2} + keys %authorities; +if( @authorities_to_delete > 0 ) { + logger("Deleting authorities from index (" . scalar(@authorities_to_delete) . " authorities to delete)"); + C4::Search::DeleteRecordIndex("authority", \@authorities_to_delete, + { as_xml => 1, record_format => "marcxml" }); +} else { + logger("No authorities to delete"); +} + +exit 0; + +sub compare_arrays{ + my ($arrayref1,$arrayref2)=@_; + return 0 if scalar(@$arrayref1)!=scalar(@$arrayref2); + my $compare=1; + for (my $i=0;$i[$i] eq $arrayref2->[$i])); + } + return $compare; +} + +sub prepare_matchstrings { + my @structure; + for my $matchstring (@_) { + my @andstrings = split /##/,$matchstring; + my @matchelements = map { + my $hash; + @$hash{qw(index subfieldtag)} = split '/', $_; + if ($$hash{subfieldtag}=~m/([0-9\.]{3})(.*)/) { + @$hash{qw(tag subfields)}=($1,[ split (//, $2) ]); + } + delete $$hash{subfieldtag}; + $hash + } @andstrings; + push @structure , \@matchelements; + } + return @structure; +} + +#trims the spaces before and after a string +#and normalize the number of spaces inside +sub trim{ + map{ + my $value=$_; + $value=~s/\s+$//g; + $value=~s/^\s+//g; + $value=~s/\s+/ /g; + $value; + }@_ +} + +sub prepare_strings{ + my $authrecord=shift; + my $attempt=shift; + my @stringstosearch; + for my $field ($authrecord->field($attempt->{tag})){ + if ($attempt->{tag} le '009'){ + if ($field->data()){ + push @stringstosearch, trim($field->data()); + } + } + else { + if ($attempt->{subfields}){ + for my $subfield (@{$attempt->{subfields}}){ + push @stringstosearch, trim(map { NormalizeString($_, undef, 1) } $field->subfield($subfield)); + } + } + else { + push @stringstosearch, trim(map { NormalizeString($_, undef, 1) } $field->as_string()); + } + + } + } + return map { + ( $_ + ? qq<$$attempt{'index'}=\"$_\"> + : () ) + }@stringstosearch; +} + +sub _build_query{ + my $authrecord = shift; + my $attempt = shift; + if ($attempt) { + my $query = join ' AND ', map { + prepare_strings($authrecord, $_) + } @$attempt; + return $query; + } + return; +} + +sub _get_id { + my $record = shift; + + if($record and (my $field = $record->field('001'))) { + return $field->data(); + } + return 0; +} + +sub _has_ppn { + my $record = shift; + + if($record and (my $field = $record->field('009'))) { + return $field->data() ? 1 : 0; + } + return 0; +} + +sub _get_date { + my $record = shift; + + if($record and (my $field = $record->field('005'))) { + return $field->data(); + } + return 0; +} + +sub _get_usage { + my $record = shift; + + if($record and (my $field = $record->field('001'))) { + return CountUsage($field->data()); + } + return 0; +} + +=head2 _choose_records + this function takes input of candidate record ids to merging + and returns + first the record to merge to + and list of records to merge from +=cut +sub _choose_records { + my @recordids = @_; + + my @records = map { GetAuthority($_) } @recordids; + my @candidate_auths = + grep { _is_duplicate( $recordids[0], $records[0], _get_id($_), $_ ) } + ( @records[ 1 .. $#records ] ); + + # See http://www.sysarch.com/Perl/sort_paper.html Schwartzian transform + my @candidate_authids = + map $_->[0] => + sort { + $b->[1] <=> $a->[1] || + $b->[2] <=> $a->[2] || + $b->[3] <=> $a->[3] + } + map [ + _get_id($_), + $choose_subs[0] ? $choose_subs[0]->($_) : 0, + $choose_subs[1] ? $choose_subs[1]->($_) : 0, + $choose_subs[2] ? $choose_subs[2]->($_) : 0 + ] => + ( $records[0], @candidate_auths ); + + return @candidate_authids; +} + +sub compare_subfields { + my ($field1, $field2, $subfields) = @_; + + my $certainty = 1; + foreach my $subfield (@$subfields) { + $debug && warn "Comparing ". $field1->tag() ."\$$subfield ", + "to ". $field2->tag(). "\$$subfield"; + + my @subfields1 = $field1->subfield($subfield); + my @subfields2 = $field2->subfield($subfield); + + if (compare_arrays([ trim(@subfields1) ], [ trim(@subfields2) ])) { + $debug && warn "certainty 1 ", @subfields1, " ", @subfields2; + } else { + $debug && warn "certainty 0 ", @subfields1, " ", @subfields2; + $certainty = 0; + last; + } + } + + return $certainty; +} + +sub _is_duplicate { + my ( $authid1, $authrecord, $authid2, $marc ) = @_; + return 0 if ( $authid1 == $authid2 ); + $authrecord ||= GetAuthority($authid1); + $marc ||= GetAuthority($authid2); + if (!$authrecord){ + warn "no or bad record for $authid1"; + return 0; + } + if (!$marc){ + warn "no or bad record for $authid2"; + return 0; + } + my $at1 = GetAuthTypeCode($authid1); + if (!$at1){ + warn "no or bad authoritytypecode for $authid1"; + return 0; + } + my $auth_tag = $authtypes{$at1} if (exists $authtypes{$at1}); + my $at2 = GetAuthTypeCode($authid2); + if (!$at2){ + warn "no or bad authoritytypecode for $authid2"; + return 0; + } + my $auth_tag2 = $authtypes{$at2} if (exists $authtypes{$at2}); + $debug and warn YAML::Dump($authrecord); + $debug and warn YAML::Dump($marc); + SetUTF8Flag($authrecord); + SetUTF8Flag($marc); + my $certainty = 1; + + $debug and warn "_is_duplicate ($authid1, $authid2)"; + if ( $marc->field($auth_tag2) + and !( $authrecord->field($auth_tag) ) ) + { + $debug && warn "certainty 0 "; + return 0; + } + elsif ( $authrecord->field($auth_tag) + && !( $marc->field($auth_tag2) ) ) + { + $debug && warn "certainty 0 "; + return 0; + } + + foreach my $check (@checks) { + last if ($certainty == 0); + my $field = $check->{field}; + my $subfields = $check->{subfields}; + my ($tag1, $tag2) = ($field, $field); + if ($field eq "TAG") { + $tag1 = $auth_tag; + $tag2 = $auth_tag2; + } + + my @fields1 = $marc->field($tag1); + my @fields2 = $authrecord->field($tag2); + if (scalar @fields1 != scalar @fields2) { + $debug && warn "Not the same number of fields: ", + "id $authid1: ". scalar @fields1 ." fields '$tag1', ", + "id $authid2: ". scalar @fields2 ." fields '$tag2'."; + $certainty = 0; + last; + } + + for (my $i=0; $i<@fields1; $i++) { + if (@$subfields > 0) { + $certainty = compare_subfields($fields1[$i], $fields2[$i], $subfields); + last if ($certainty == 0); + } else { + $debug && warn "Comparing ".$fields1[$i]->tag()." data ", + "to ".$fields2[$i]->tag()." data"; + if ($fields1[$i]->data() ne $fields2[$i]->data()) { + $debug && warn "certainty 0 ", $fields1[$i]->data(), " ", + $fields2[$i]->data(); + $certainty = 0; + last; + } + } + } + } + + return $certainty; +} + +sub logger { + my ($sec, $min, $hour, $mday, $mon, $year) = localtime; + my $time_msg = sprintf "[%04d-%02d-%02d %02d:%02d:%02d]", + 1900+$year, 1+$mon, $mday, $hour, $min, $sec; + say $time_msg, ' ', @_; +} + +sub print_usage { + print <<_USAGE_; +$0: deduplicate authorities + +Use this batch job to remove duplicate authorities + +Parameters: + --match + + matchstring is composed of : index1/tagsubfield1[##index2/tagsubfield2] + the matching will be done with an AND between all elements of ONE + matchstring. + tagsubfield can be 123a or 123abc or 123. + + If multiple match parameters are sent, then it will try to match in the + order it is provided to the script. + + Examples: + at/152b##he-main/2..a##he/2..bxyzt##ident/009 + authtype/152b##he-main,ext/2..a##he,ext/2..bxyz + sn,ne,st-numeric/001##authtype/152b##he-main,ext/2..a##he,ext/2..bxyz + + Match strings MUST contains at least one of the + following indexes: he, he-main and ident + + + --check [,[,...]] + + Check only these fields and subfields to determine whether two + authorities are duplicates. + + is a three-digit number which represents the field tag, or the + string 'TAG' which represents auth_tag_to_report and depends on + authtypecode. You can use '.' as wildcards. E.g. '2..' + + is a list of subfield codes, matching [0-9a-z] + + Examples: + 200abxyz will check subfields a,b,x,y,z of 200 fields + 009,152b will check 009 data and 152\$b subfields + TAGab will check 250\$a and 250\$b if auth_tag_to_report is 250 for + the type of the authority currently being checked + + Default: TAGabcdefghitxyz + + + --choose-method + + Method(s) used to choose which authority to keep in case we found + duplicates. + is a string composed of letters describing what methods to use + and in which order. + Letters can be: + d: date, keep the most recent authority (based on 005 field) + u: usage, keep the most used authority + p: PPN (UNIMARC only), keep the authority with a ppn (when some + authorities don't have one, based on 009 field) + + Examples: + 'pdu': Among the authorities that have a PPN, keep the most recent, + and if two (or more) have the same date in 005, keep the + most used. + + Default is 'u' + You cannot type more than 3 letters + + + --where limit the deduplication to SOME authorities only + + --verbose display logs + + --all deduplicate all authority type + + --help or -h show this message. + +Other paramters : + Authtypecode to deduplicate authorities on +exemple: + $0 --match ident/009 --match he-main,ext/200a##he,ext/200bxz NC + +Note : If launched with DEBUG=1 it will print more messages +_USAGE_ +} -- 1.7.10.4