was a test for inverted indexes. Useless we have choosen zebra

This commit is contained in:
tipaul 2005-08-12 13:39:51 +00:00
parent 0275f18980
commit 35e4cf543c

View file

@ -1,549 +0,0 @@
package C4::SearchMarcTest;
# Copyright 2000-2002 Katipo Communications
#
# This file is part of Koha.
#
# Koha is free software; you can redistribute it and/or modify it under the
# terms of the GNU General Public License as published by the Free Software
# Foundation; either version 2 of the License, or (at your option) any later
# version.
#
# Koha is distributed in the hope that it will be useful, but WITHOUT ANY
# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License along with
# Koha; if not, write to the Free Software Foundation, Inc., 59 Temple Place,
# Suite 330, Boston, MA 02111-1307 USA
use strict;
require Exporter;
use DBI;
use C4::Context;
use C4::Biblio;
use C4::Date;
use Date::Manip;
use vars qw($VERSION @ISA @EXPORT @EXPORT_OK %EXPORT_TAGS);
# set the version for version checking
$VERSION = 0.02;
=head1 NAME
C4::Search - Functions for searching the Koha MARC catalog
=head1 FUNCTIONS
This module provides the searching facilities for the Koha MARC catalog
A COPY of official SearchMarc, with some tests for inverted index table
works only with 1 MARC tag/subfield
=cut
@ISA = qw(Exporter);
@EXPORT = qw(&catalogsearch &findseealso &findsuggestion &getMARCnotes &getMARCsubjects);
=head1 findsuggestion($dbh,$values);
=head2 $dbh is a link to the DB handler.
use C4::Context;
my $dbh =C4::Context->dbh;
=head2 $values is a word
Searches words with the same soundex, ordered by frequency of use.
Useful to suggest other searches to the users.
=cut
sub findsuggestion {
my ($dbh,$values) = @_;
my $sth = $dbh->prepare("SELECT count( * ) AS total, word FROM marc_word WHERE sndx_word = soundex( ? ) AND word <> ? GROUP BY word ORDER BY total DESC");
my @results;
for(my $i = 0 ; $i <= $#{$values} ; $i++) {
if (length(@$values[$i]) >=5) {
$sth->execute(@$values[$i],@$values[$i]);
my $resfound = 1;
my @resline;
while ((my ($count,$word) = $sth->fetchrow) and $resfound <=10) {
push @results, "@$values[$i]|$word|$count";
# $results{@$values[$i]} = \@resline;
$resfound++;
}
}
}
return \@results;
}
=head1 findseealso($dbh,$fields);
=head2 $dbh is a link to the DB handler.
use C4::Context;
my $dbh =C4::Context->dbh;
=head2 $fields is a reference to the fields array
This function modify the @$fields array and add related fields to search on.
=cut
sub findseealso {
my ($dbh, $fields) = @_;
my $tagslib = MARCgettagslib ($dbh,1);
for (my $i=0;$i<=$#{$fields};$i++) {
my ($tag) =substr(@$fields[$i],1,3);
my ($subfield) =substr(@$fields[$i],4,1);
@$fields[$i].=','.$tagslib->{$tag}->{$subfield}->{seealso} if ($tagslib->{$tag}->{$subfield}->{seealso});
}
}
=head1 my ($count, @results) = catalogsearch($dbh, $tags, $and_or, $excluding, $operator, $value, $offset,$length,$orderby);
=head2 $dbh is a link to the DB handler.
use C4::Context;
my $dbh =C4::Context->dbh;
$tags,$and_or, $excluding, $operator, $value are references to array
=head2 $tags
contains the list of tags+subfields (for example : $@tags[0] = '200a')
A field can be a list of fields : '200f','700a','700b','701a','701b'
Example
=head2 $and_or
contains a list of strings containing and or or. The 1st value is useless.
=head2 $excluding
contains 0 or 1. If 1, then the request is negated.
=head2 $operator
contains contains,=,start,>,>=,<,<= the = and start work on the complete subfield. The contains operator works on every word in the subfield.
examples :
contains home, search home anywhere.
= home, search a string being home.
=head2 $value
contains the value to search
If it contains a * or a %, then the search is partial.
=head2 $offset and $length
returns $length results, beginning at $offset
=head2 $orderby
define the field used to order the request. Any field in the biblio/biblioitem tables can be used. DESC is possible too
(for example title, title DESC,...)
=head2 RETURNS
returns an array containing hashes. The hash contains all biblio & biblioitems fields and a reference to an item hash. The "item hash contains one line for each callnumber & the number of items related to the callnumber.
=cut
=head2 my $marcnotesarray = &getMARCnotes($dbh,$bibid,$marcflavour);
Returns a reference to an array containing all the notes stored in the MARC database for the given bibid.
$marcflavour ("MARC21" or "UNIMARC") determines which tags are used for retrieving subjects.
=head2 my $marcsubjctsarray = &getMARCsubjects($dbh,$bibid,$marcflavour);
Returns a reference to an array containing all the subjects stored in the MARC database for the given bibid.
$marcflavour ("MARC21" or "UNIMARC") determines which tags are used for retrieving subjects.
=cut
sub catalogsearch {
my ($dbh, $tags, $and_or, $excluding, $operator, $value, $offset,$length,$orderby,$desc_or_asc) = @_;
# "Normal" statements
my @normal_tags = ();
my @normal_and_or = ();
my @normal_operator = ();
my @normal_value = ();
# Extracts the NOT statements from the list of statements
my @not_tags = ();
my @not_and_or = ();
my @not_operator = ();
my @not_value = ();
my $any_not = 0;
$orderby = "biblio.title" unless $orderby;
$desc_or_asc = "ASC" unless $desc_or_asc;
# the item.notforloan contains an integer. Every value <>0 means "book unavailable for loan".
# but each library can have it's own table of meaning for each value. Get them
# 1st search if there is a list of authorised values connected to items.notforloan
my $sth = $dbh->prepare('select authorised_value from marc_subfield_structure where kohafield="items.notforloan"');
$sth->execute;
my %notforloanstatus;
my ($authorised_valuecode) = $sth->fetchrow;
if ($authorised_valuecode) {
$sth = $dbh->prepare("select authorised_value,lib from authorised_values where category=?");
$sth->execute($authorised_valuecode);
while (my ($authorised_value,$lib) = $sth->fetchrow) {
$notforloanstatus{$authorised_value} = $lib?$lib:$authorised_value;
}
}
#
#
# marc_T_word PROOF OF CONCEPT BEGINNING
#
# fixme : only do a search on "contains every word"
# misses :
# - begins or is equal to
# - excluding
# - or
# the global array result.
my @result;
for(my $i = 0 ; $i <= $#{$value} ; $i++)
{
# replace * by %
@$value[$i] =~ s/\*/%/g;
# remove % at the beginning
@$value[$i] =~ s/^%//g;
@$value[$i] =~ s/(\.|\?|\:|\!|\'|,|\-|\"|\(|\)|\[|\]|\{|\}|\/)/ /g if @$operator[$i] eq "contains";
foreach my $word (split(/ /, @$value[$i])) # if operator is contains, splits the words in separate requests
{
# the array from this word
my @thiswordresults;
my $Tquery = $dbh->prepare("select tagsubfield,usedin from marc_Tword where tagsubfield in (@$tags[$i]) and word like ?");
$Tquery->execute($word);
warn "EXECUTING select tagsubfield,usedin from marc_Tword where tagsubfield in (@$tags[$i]) and word like $word";
# get the list of biblionumber - title
while (my ($tagsubfield,$usedin) = $Tquery->fetchrow) {
# warn "$word with ".@$tags[$i]." used in $usedin";
# split it in an array
my @lines = split /,/,$usedin;
# and copy it to an hash.
foreach my $line (@lines) {
# warn "PUSHING $line" if $line; # the if $line avoid pushing the 1st entry, that is empty (usedin begins by a ,)
push @thiswordresults, $line if $line;
}
}
# now, as it's a AND, merge %results & %thiswordresults in 1 hash
@result = @thiswordresults if $#result<0; #for the 1st loop, fill the global array
my %intersect;
my %union;
my $x; # temp variable
foreach $x (@result, @thiswordresults) {
$union{$x}++ && $intersect{$x}++;
}
@result = keys %intersect;
}
}
# we have biblionumber array.
# now, sort it
my @result = sort @result;
#Now, loads title and author from [offset] to [offset]+[length]
my $counter = $offset;
# HINT : biblionumber as bn is important. The hash is fills biblionumber with items.biblionumber.
# so if you dont' has an item, you get a not nice empty value.
$sth = $dbh->prepare("SELECT biblio.biblionumber as bn,biblioitems.*,biblio.*, marc_biblio.bibid,itemtypes.notforloan,itemtypes.description
FROM biblio, marc_biblio
LEFT JOIN biblioitems on biblio.biblionumber = biblioitems.biblionumber
LEFT JOIN itemtypes on itemtypes.itemtype=biblioitems.itemtype
WHERE biblio.biblionumber = marc_biblio.biblionumber AND biblio.biblionumber = ?");
#
#
# marc_Tword Proof of concept
#
#
my $subtitle;
my $sth_subtitle = $dbh->prepare("SELECT subtitle FROM bibliosubtitle WHERE biblionumber=?"); # Added BY JF for Subtitles
my @finalresult = ();
my @CNresults=();
my $totalitems=0;
my $oldline;
my ($oldbibid, $oldauthor, $oldtitle);
my $sth_itemCN = $dbh->prepare("select items.* from items where biblionumber=? and (itemlost = 0 or itemlost is NULL)");
my $sth_issue = $dbh->prepare("select date_due,returndate from issues where itemnumber=?");
# parse all biblios between start & end.
warn "RESULT SIZE : ".$#result;
while (($counter <= $#result) && ($counter <= ($offset + $length))) {
# search & parse all items & note itemcallnumber
# 1st, get the biblionumber
$result[$counter] =~ /(.*)-(.*)/;
$sth->execute($2);
warn "EXECUTING SELECT biblio.biblionumber as bn,biblioitems.*,biblio.*, marc_biblio.bibid,itemtypes.notforloan,itemtypes.description FROM biblio, marc_biblio LEFT JOIN biblioitems on biblio.biblionumber = biblioitems.biblionumber LEFT JOIN itemtypes on itemtypes.itemtype=biblioitems.itemtype WHERE biblio.biblionumber = marc_biblio.biblionumber AND biblio.biblionumber = $2";
my $continue=1;
my $line = $sth->fetchrow_hashref;
my $biblionumber=$line->{bn};
# Return subtitles first ADDED BY JF
$sth_subtitle->execute($biblionumber);
warn "EXECUTING SELECT subtitle FROM bibliosubtitle WHERE biblionumber=$biblionumber";
my $subtitle_here.= $sth_subtitle->fetchrow." ";
chop $subtitle_here;
$subtitle = $subtitle_here;
# /ADDED BY JF
# $continue=0 unless $line->{bn};
# my $lastitemnumber;
$sth_itemCN->execute($biblionumber);
warn "EXECUTING itemCN select items.* from items where biblionumber=$biblionumber and (itemlost = 0 or itemlost is NULL)";
my @CNresults = ();
my $notforloan=1; # to see if there is at least 1 item that can be issued
while (my $item = $sth_itemCN->fetchrow_hashref) {
# parse the result, putting holdingbranch & itemcallnumber in separate array
# then all other fields in the main array
# search if item is on loan
my $date_due;
$sth_issue->execute($item->{itemnumber});
warn "EXECUTING ISSUES select date_due,returndate from issues where itemnumber=".$item->{itemnumber};
while (my $loan = $sth_issue->fetchrow_hashref) {
if ($loan->{date_due} and !$loan->{returndate}) {
$date_due = $loan->{date_due};
}
}
# store this item
my %lineCN;
$lineCN{holdingbranch} = $item->{holdingbranch};
$lineCN{itemcallnumber} = $item->{itemcallnumber};
$lineCN{location} = $item->{location};
$lineCN{date_due} = format_date($date_due);
$lineCN{notforloan} = $notforloanstatus{$line->{notforloan}} if ($line->{notforloan}); # setting not forloan if itemtype is not for loan
$lineCN{notforloan} = $notforloanstatus{$item->{notforloan}} if ($item->{notforloan}); # setting not forloan it this item is not for loan
$notforloan=0 unless ($item->{notforloan} or $item->{wthdrawn} or $item->{itemlost});
push @CNresults,\%lineCN;
$totalitems++;
}
# save the biblio in the final array, with item and item issue status
my %newline;
%newline = %$line;
$newline{totitem} = $totalitems;
# if $totalitems == 0, check if it's being ordered.
if ($totalitems == 0) {
my $sth = $dbh->prepare("select count(*) from aqorders where biblionumber=? and datecancellationprinted is NULL");
$sth->execute($biblionumber);
warn "EXECUTING select count(*) from aqorders where biblionumber=$biblionumber and datecancellationprinted is NULL";
my ($ordered) = $sth->fetchrow;
$newline{onorder} = 1 if $ordered;
}
$newline{biblionumber} = $biblionumber;
$newline{norequests} = 0;
$newline{norequests} = 1 if ($line->{notforloan}); # itemtype not issuable
$newline{norequests} = 1 if (!$line->{notforloan} && $notforloan); # itemtype issuable but all items not issuable for instance
$newline{subtitle} = $subtitle; # put the subtitle in ADDED BY JF
my @CNresults2= @CNresults;
$newline{CN} = \@CNresults2;
$newline{'even'} = 1 if $#finalresult % 2 == 0;
$newline{'odd'} = 1 if $#finalresult % 2 == 1;
$newline{'timestamp'} = format_date($newline{timestamp});
@CNresults = ();
push @finalresult, \%newline;
$totalitems=0;
$counter++;
}
my $nbresults = $#result+1;
return (\@finalresult, $nbresults);
}
# Creates the SQL Request
sub create_request {
my ($dbh,$tags, $and_or, $operator, $value) = @_;
my $sql_tables; # will contain marc_subfield_table as m1,...
my $sql_where1; # will contain the "true" where
my $sql_where2 = "("; # will contain m1.bibid=m2.bibid
my $nb_active=0; # will contain the number of "active" entries. an entry is active if a value is provided.
my $nb_table=1; # will contain the number of table. ++ on each entry EXCEPT when an OR is provided.
my $maxloop=8; # the maximum number of words to avoid a too complex search.
$maxloop = @$value if @$value<$maxloop;
for(my $i=0; $i<=$maxloop;$i++) {
if (@$value[$i]) {
$nb_active++;
if ($nb_active==1) {
if (@$operator[$i] eq "start") {
$sql_tables .= "marc_subfield_table as m$nb_table,";
$sql_where1 .= "(m1.subfieldvalue like ".$dbh->quote("@$value[$i]%");
if (@$tags[$i]) {
$sql_where1 .=" and concat(m1.tag,m1.subfieldcode) in (@$tags[$i])";
}
$sql_where1.=")";
} elsif (@$operator[$i] eq "contains") {
$sql_tables .= "marc_word as m$nb_table,";
$sql_where1 .= "(m1.word like ".$dbh->quote("@$value[$i]");
if (@$tags[$i]) {
$sql_where1 .=" and m1.tagsubfield in (@$tags[$i])";
}
$sql_where1.=")";
} else {
$sql_tables .= "marc_subfield_table as m$nb_table,";
$sql_where1 .= "(m1.subfieldvalue @$operator[$i] ".$dbh->quote("@$value[$i]");
if (@$tags[$i]) {
$sql_where1 .=" and concat(m1.tag,m1.subfieldcode) in (@$tags[$i])";
}
$sql_where1.=")";
}
} else {
if (@$operator[$i] eq "start") {
$nb_table++;
$sql_tables .= "marc_subfield_table as m$nb_table,";
$sql_where1 .= "@$and_or[$i] (m$nb_table.subfieldvalue like ".$dbh->quote("@$value[$i]%");
if (@$tags[$i]) {
$sql_where1 .=" and concat(m$nb_table.tag,m$nb_table.subfieldcode) in (@$tags[$i])";
}
$sql_where1.=")";
$sql_where2 .= "m1.bibid=m$nb_table.bibid and ";
} elsif (@$operator[$i] eq "contains") {
if (@$and_or[$i] eq 'and') {
$nb_table++;
$sql_tables .= "marc_word as m$nb_table,";
$sql_where1 .= "@$and_or[$i] (m$nb_table.word like ".$dbh->quote("@$value[$i]");
if (@$tags[$i]) {
$sql_where1 .=" and m$nb_table.tagsubfield in(@$tags[$i])";
}
$sql_where1.=")";
$sql_where2 .= "m1.bibid=m$nb_table.bibid and ";
} else {
$sql_where1 .= "@$and_or[$i] (m$nb_table.word like ".$dbh->quote("@$value[$i]");
if (@$tags[$i]) {
$sql_where1 .=" and m$nb_table.tagsubfield in (@$tags[$i])";
}
$sql_where1.=")";
$sql_where2 .= "m1.bibid=m$nb_table.bibid and ";
}
} else {
$nb_table++;
$sql_tables .= "marc_subfield_table as m$nb_table,";
$sql_where1 .= "@$and_or[$i] (m$nb_table.subfieldvalue @$operator[$i] ".$dbh->quote(@$value[$i]);
if (@$tags[$i]) {
$sql_where1 .=" and concat(m$nb_table.tag,m$nb_table.subfieldcode) in (@$tags[$i])";
}
$sql_where2 .= "m1.bibid=m$nb_table.bibid and ";
$sql_where1.=")";
}
}
}
}
if($sql_where2 ne "(") # some datas added to sql_where2, processing
{
$sql_where2 = substr($sql_where2, 0, (length($sql_where2)-5)); # deletes the trailing ' and '
$sql_where2 .= ")";
}
else # no sql_where2 statement, deleting '('
{
$sql_where2 = "";
}
chop $sql_tables; # deletes the trailing ','
return ($sql_tables, $sql_where1, $sql_where2);
}
sub getMARCnotes {
my ($dbh, $bibid, $marcflavour) = @_;
my ($mintag, $maxtag);
if ($marcflavour eq "MARC21") {
$mintag = "500";
$maxtag = "599";
} else { # assume unimarc if not marc21
$mintag = "300";
$maxtag = "399";
}
my $sth=$dbh->prepare("SELECT subfieldvalue,tag FROM marc_subfield_table WHERE bibid=? AND tag BETWEEN ? AND ? ORDER BY tagorder");
$sth->execute($bibid,$mintag,$maxtag);
my @marcnotes;
my $note = "";
my $tag = "";
my $marcnote;
while (my $data=$sth->fetchrow_arrayref) {
my $value=$data->[0];
my $thistag=$data->[1];
if ($value=~/\.$/) {
$value=$value . " ";
}
if ($thistag ne $tag && $note ne "") {
$marcnote = {marcnote => $note,};
push @marcnotes, $marcnote;
$note=$value;
$tag=$thistag;
}
if ($note ne $value) {
$note = $note." ".$value;
}
}
if ($note) {
$marcnote = {marcnote => $note};
push @marcnotes, $marcnote; #load last tag into array
}
$sth->finish;
$dbh->disconnect;
my $marcnotesarray=\@marcnotes;
return $marcnotesarray;
} # end getMARCnotes
sub getMARCsubjects {
my ($dbh, $bibid, $marcflavour) = @_;
my ($mintag, $maxtag);
if ($marcflavour eq "MARC21") {
$mintag = "600";
$maxtag = "699";
} else { # assume unimarc if not marc21
$mintag = "600";
$maxtag = "619";
}
my $sth=$dbh->prepare("SELECT subfieldvalue,subfieldcode FROM marc_subfield_table WHERE bibid=? AND tag BETWEEN ? AND ? ORDER BY tagorder");
$sth->execute($bibid,$mintag,$maxtag);
my @marcsubjcts;
my $subjct = "";
my $subfield = "";
my $marcsubjct;
while (my $data=$sth->fetchrow_arrayref) {
my $value = $data->[0];
my $subfield = $data->[1];
if ($subfield eq "a" && $value ne $subjct) {
$marcsubjct = {MARCSUBJCT => $value,};
push @marcsubjcts, $marcsubjct;
$subjct = $value;
}
}
$sth->finish;
$dbh->disconnect;
my $marcsubjctsarray=\@marcsubjcts;
return $marcsubjctsarray;
} #end getMARCsubjects
END { } # module clean-up code here (global destructor)
1;
__END__
=back
=head1 AUTHOR
Koha Developement team <info@koha.org>
=cut