home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
PC World 2001 August
/
PCWorld_2001-08_cd.bin
/
Komunikace
/
sambar
/
_setup.1
/
search.pl
< prev
next >
Wrap
Text File
|
2000-10-17
|
3KB
|
165 lines
#
# Perl-based Search Engine
#
# Copyright 1998 Tod Sambar
# All rights reserved.
#
my $docroot = $ENV{'DOCUMENT_ROOT'};
my $docrootlen = length($docroot);
#
# PARSE THE CGI FORM
#
# Buffer the POST content
read(STDIN, $buffer, $ENV{'CONTENT_LENGTH'});
# Process the name=value argument pairs
my $pair;
my $name;
my $value;
my @args = split(/&/, $buffer);
foreach $pair (@args)
{
($name, $value) = split(/=/, $pair);
# Unescape the argument value
$value =~ tr/+/ /;
$value =~ s/%([a-fA-F0-9][a-fA-F0-9])/pack("C", hex($1))/eg;
# Save the name=value pair for use below.
$FORM{$name} = $value;
}
#
# BUILD THE FILE SEARCH LIST
#
# NOTE: We only search *.htm, *.html and *.txt files
#
# WARNING: This search engine is a serious memory hog
# for large sites with many files.
# Change to the Server's document root
chdir($docroot);
# Get a list of all matching files
my $list;
$list = `dir /b/S *.htm`;
$list .= `dir /b/S *.html`;
$list .= `dir /b/S *.txt`;
my @LIST = split(/\s+/, $list);
#
# SEARCH THE FILES
#
my $file;
my $data;
my $match;
my $found;
my $query = $FORM{'query'};
print "Content-type: text/html\n\n";
print "<HTML><HEAD><TITLE>PERL Search Results</TITLE></HEAD>\n";
print "<BODY bgcolor=#FFFFFF>\n";
print "<CENTER>\n";
print "<FONT SIZE=6 COLOR=#99003><B>Sambar Server</B></FONT><BR>\n";
print "<FONT SIZE=6 COLOR=#99003><I>PERL Search Results</I></FONT>\n";
print "</CENTER><P>\n";
print "<B>Query</B>: <I>".$query."</I><P>\n";
print "<HR><UL>\n";
my @query = split(/\s+/, $query);
$found = 'no';
foreach $file (@LIST)
{
# Read the file
open(FILE, "<$file");
@LINES = <FILE>;
close(FILE);
# Merge the lines of the file together
$data = join(' ', @LINES);
$data =~ s/\n//g;
$match = 'no';
if ($FORM{'logic'} eq 'and')
{
foreach $term (@query)
{
# Perform case insensitive comparison
if (!($data =~ /$term/i))
{
# Term did not match.
$match = 'no';
last;
}
else
{
# Term matched
$match = 'yes';
}
}
}
elsif ($FORM{'logic'} eq 'or')
{
foreach $term (@query)
{
if ($data =~ /$term/i)
{
# Term matched
$match = 'yes';
last;
}
}
}
else
{
print "Unrecognized query logic...\n";
}
if ($match eq 'yes')
{
$found = 'yes';
# Strip off the document root
$file = substr($file, $docrootlen - 1);
# Fixup the directory slashes
$file =~ s/\\/\//g;
if ($data =~ /<title>(.*)<\/title>/i)
{
print "<LI><A HREF=\"$file\"> ".$1."</A><BR>\n";
}
else
{
print "<LI><A HREF=\"$file\"> ".$file."</A><BR>\n";
}
}
}
if ($found eq 'no')
{
print "<I><B>No search results.</B></I><BR>\n";
}
print "\n</UL><HR>\n";
print "</BODY></HTML>\n";
#
# DONE
#
exit(0);