File:  [LON-CAPA] / loncom / interface / loncommon.pm
Revision 1.50: download - view: text, annotated - select for diffs
Thu Aug 8 13:06:22 2002 UTC (21 years, 9 months ago) by matthew
Branches: MAIN
CVS tags: HEAD
Attempt to prevent the lousyness with lonindexer remembering incorrectly
to only show sequences.  I have not been able to reproduce the error...

# The LearningOnline Network with CAPA
# a pile of common routines
#
# $Id: loncommon.pm,v 1.50 2002/08/08 13:06:22 matthew Exp $
#
# Copyright Michigan State University Board of Trustees
#
# This file is part of the LearningOnline Network with CAPA (LON-CAPA).
#
# LON-CAPA 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.
#
# LON-CAPA 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 LON-CAPA; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
#
# /home/httpd/html/adm/gpl.txt
#
# http://www.lon-capa.org/
#
# YEAR=2001
# 2/13-12/7 Guy Albertelli
# 12/11,12/12,12/17 Scott Harrison
# 12/21 Gerd Kortemeyer
# 12/21 Scott Harrison
# 12/25,12/28 Gerd Kortemeyer
# YEAR=2002
# 1/4 Gerd Kortemeyer
# 6/24,7/2 H. K. Ng

# Makes a table out of the previous attempts
# Inputs result_from_symbread, user, domain, course_id
# Reads in non-network-related .tab files

# POD header:

=pod

=head1 NAME

Apache::loncommon - pile of common routines

=head1 SYNOPSIS

Referenced by other mod_perl Apache modules.

Invocation:
 &Apache::loncommon::SUBROUTINENAME(ARGUMENTS);

=head1 INTRODUCTION

Common collection of used subroutines.  This collection helps remove
redundancy from other modules and increase efficiency of memory usage.

Current things done:

 Makes a table out of the previous homework attempts
 Inputs result_from_symbread, user, domain, course_id
 Reads in non-network-related .tab files

This is part of the LearningOnline Network with CAPA project
described at http://www.lon-capa.org.

=head2 General Subroutines

=over 4

=cut 

# End of POD header
package Apache::loncommon;

use strict;
use Apache::lonnet();
use GDBM_File;
use POSIX qw(strftime);
use Apache::Constants qw(:common);
use Apache::lonmsg();
my $readit;

=pod 

=item Global Variables

=over 4

=cut
# ----------------------------------------------- Filetypes/Languages/Copyright
my %language;
my %cprtag;
my %fe; my %fd;
my %category_extensions;

# ---------------------------------------------- Thesaurus variables

=pod

=item %Keywords  

A hash used by &keyword to determine if a word is considered a keyword.

=item $thesaurus_db_file

Scalar containing the full path to the thesaurus database.                 

=cut

my %Keywords;
my $thesaurus_db_file;


=pod

=back

=cut

# ----------------------------------------------------------------------- BEGIN

=pod

=item BEGIN() 

Initialize values from language.tab, copyright.tab, filetypes.tab,
thesaurus.tab, and filecategories.tab.

=cut

# ----------------------------------------------------------------------- BEGIN

BEGIN {
    # Variable initialization
    $thesaurus_db_file = $Apache::lonnet::perlvar{'lonTabDir'}."/thesaurus.db";
    #
    unless ($readit) {
# ------------------------------------------------------------------- languages
    {
	my $fh=Apache::File->new($Apache::lonnet::perlvar{'lonTabDir'}.
				 '/language.tab');
	if ($fh) {
	    while (<$fh>) {
		next if /^\#/;
		chomp;
		my ($key,$val)=(split(/\s+/,$_,2));
		$language{$key}=$val;
	    }
	}
    }
# ------------------------------------------------------------------ copyrights
    {
	my $fh=Apache::File->new($Apache::lonnet::perlvar{'lonIncludes'}.
				  '/copyright.tab');
	if ($fh) {
	    while (<$fh>) {
		next if /^\#/;
		chomp;
		my ($key,$val)=(split(/\s+/,$_,2));
		$cprtag{$key}=$val;
	    }
	}
    }
# ------------------------------------------------------------- file categories
    {
	my $fh=Apache::File->new($Apache::lonnet::perlvar{'lonTabDir'}.
				  '/filecategories.tab');
	if ($fh) {
	    while (<$fh>) {
		next if /^\#/;
		chomp;
		my ($extension,$category)=(split(/\s+/,$_,2));
		push @{$category_extensions{lc($category)}},$extension;
	    }
	}
    }
# ------------------------------------------------------------------ file types
    {
	my $fh=Apache::File->new($Apache::lonnet::perlvar{'lonTabDir'}.
	       '/filetypes.tab');
	if ($fh) {
            while (<$fh>) {
		next if (/^\#/);
		chomp;
		my ($ending,$emb,$descr)=split(/\s+/,$_,3);
		if ($descr ne '') { 
		    $fe{$ending}=lc($emb);
		    $fd{$ending}=$descr;
		}
	    }
	}
    }
    &Apache::lonnet::logthis(
              "<font color=yellow>INFO: Read file types</font>");
    $readit=1;
    }  # end of unless($readit) 
    
}
# ============================================================= END BEGIN BLOCK
###############################################################
##           HTML and Javascript Helper Functions            ##
###############################################################

=pod 

=item browser_and_searcher_javascript 

Returns scalar containing javascript to open a browser window
or a searcher window.  Also creates 

=over 4

=item openbrowser(formname,elementname,only,omit) [javascript]

inputs: formname, elementname, only, omit

formname and elementname indicate the name of the html form and name of
the element that the results of the browsing selection are to be placed in. 

Specifying 'only' will restrict the browser to displaying only files
with the given extension.  Can be a comma seperated list.

Specifying 'omit' will restrict the browser to NOT displaying files
with the given extension.  Can be a comma seperated list.

=item opensearcher(formname, elementname) [javascript]

Inputs: formname, elementname

formname and elementname specify the name of the html form and the name
of the element the selection from the search results will be placed in.

=back

=cut

###############################################################
sub browser_and_searcher_javascript {
    return <<END;
    var editbrowser = null;
    function openbrowser(formname,elementname,only,omit) {
        var url = '/res/?';
        if (editbrowser == null) {
            url += 'launch=1&';
        }
        url += 'catalogmode=interactive&';
        url += 'mode=edit&';
        url += 'form=' + formname + '&';
        if (only != null) {
            url += 'only=' + only + '&';
        } 
        if (omit != null) {
            url += 'omit=' + omit + '&';
        }
        url += 'element=' + elementname + '';
        var title = 'Browser';
        var options = 'scrollbars=1,resizable=1,menubar=0';
        options += ',width=700,height=600';
        editbrowser = open(url,title,options,'1');
        editbrowser.focus();
    }
    var editsearcher;
    function opensearcher(formname,elementname) {
        var url = '/adm/searchcat?';
        if (editsearcher == null) {
            url += 'launch=1&';
        }
        url += 'catalogmode=interactive&';
        url += 'mode=edit&';
        url += 'form=' + formname + '&';
        url += 'element=' + elementname + '';
        var title = 'Search';
        var options = 'scrollbars=1,resizable=1,menubar=0';
        options += ',width=700,height=600';
        editsearcher = open(url,title,options,'1');
        editsearcher.focus();
    }
END
}



###############################################################

=pod

=item linked_select_forms(...)

linked_select_forms returns a string containing a <script></script> block
and html for two <select> menus.  The select menus will be linked in that
changing the value of the first menu will result in new values being placed
in the second menu.  The values in the select menu will appear in alphabetical
order.

linked_select_forms takes the following ordered inputs:

=over 4

=item $formname, the name of the <form> tag

=item $middletext, the text which appears between the <select> tags

=item $firstdefault, the default value for the first menu

=item $firstselectname, the name of the first <select> tag

=item $secondselectname, the name of the second <select> tag

=item $hashref, a reference to a hash containing the data for the menus.

=back 

Below is an example of such a hash.  Only the 'text', 'default', and 
'select2' keys must appear as stated.  keys(%menu) are the possible 
values for the first select menu.  The text that coincides with the 
first menu value is given in $menu{$choice1}->{'text'}.  The values 
and text for the second menu are given in the hash pointed to by 
$menu{$choice1}->{'select2'}.  

my %menu = ( A1 => { text =>"Choice A1" ,
                      default => "B3",
                      select2 => { 
                          B1 => "Choice B1",
                          B2 => "Choice B2",
                          B3 => "Choice B3",
                          B4 => "Choice B4"
                          }
                  },
              A2 => { text =>"Choice A2" ,
                      default => "C2",
                      select2 => { 
                          C1 => "Choice C1",
                          C2 => "Choice C2",
                          C3 => "Choice C3"
                          }
                  },
              A3 => { text =>"Choice A3" ,
                      default => "D6",
                      select2 => { 
                          D1 => "Choice D1",
                          D2 => "Choice D2",
                          D3 => "Choice D3",
                          D4 => "Choice D4",
                          D5 => "Choice D5",
                          D6 => "Choice D6",
                          D7 => "Choice D7"
                          }
                  }
              );

=cut

# ------------------------------------------------

sub linked_select_forms {
    my ($formname,
        $middletext,
        $firstdefault,
        $firstselectname,
        $secondselectname, 
        $hashref
        ) = @_;
    my $second = "document.$formname.$secondselectname";
    my $first = "document.$formname.$firstselectname";
    # output the javascript to do the changing
    my $result = '';
    $result.="<script>\n";
    $result.="var select2data = new Object();\n";
    $" = '","';
    my $debug = '';
    foreach my $s1 (sort(keys(%$hashref))) {
        $result.="select2data.d_$s1 = new Object();\n";        
        $result.="select2data.d_$s1.def = new String('".
            $hashref->{$s1}->{'default'}."');\n";
        $result.="select2data.d_$s1.values = new Array(";        
        my @s2values = sort(keys( %{ $hashref->{$s1}->{'select2'} } ));
        $result.="\"@s2values\");\n";
        $result.="select2data.d_$s1.texts = new Array(";        
        my @s2texts;
        foreach my $value (@s2values) {
            push @s2texts, $hashref->{$s1}->{'select2'}->{$value};
        }
        $result.="\"@s2texts\");\n";
    }
    $"=' ';
    $result.= <<"END";

function select1_changed() {
    // Determine new choice
    var newvalue = "d_" + $first.value;
    // update select2
    var values     = select2data[newvalue].values;
    var texts      = select2data[newvalue].texts;
    var select2def = select2data[newvalue].def;
    var i;
    // out with the old
    for (i = 0; i < $second.options.length; i++) {
        $second.options[i] = null;
    }
    // in with the nuclear
    for (i=0;i<values.length; i++) {
        $second.options[i] = new Option(values[i]);
        $second.options[i].text = texts[i];
        if (values[i] == select2def) {
            $second.options[i].selected = true;
        }
    }
}
</script>
END
    # output the initial values for the selection lists
    $result .= "<select size=\"1\" name=\"$firstselectname\" onchange=\"select1_changed()\">\n";
    foreach my $value (sort(keys(%$hashref))) {
        $result.="    <option value=\"$value\" ";
        $result.=" selected=\"true\" " if ($value eq $firstdefault);
        $result.=">$hashref->{$value}->{'text'}</option>\n";
    }
    $result .= "</select>\n";
    my %select2 = %{$hashref->{$firstdefault}->{'select2'}};
    $result .= $middletext;
    $result .= "<select size=\"1\" name=\"$secondselectname\">\n";
    my $seconddefault = $hashref->{$firstdefault}->{'default'};
    foreach my $value (sort(keys(%select2))) {
        $result.="    <option value=\"$value\" ";        
        $result.=" selected=\"true\" " if ($value eq $seconddefault);
        $result.=">$select2{$value}</option>\n";
    }
    $result .= "</select>\n";
    #    return $debug;
    return $result;
}   #  end of sub linked_select_forms {

###############################################################

=pod

=item help_open_topic($topic, $text, $stayOnPage, $width, $height)

Returns a string corresponding to an HTML link to the given help $topic, where $topic corresponds to the name of a .tex file in /home/httpd/html/adm/help/tex, with underscores replaced by spaces.

$text will optionally be linked to the same topic, allowing you to link text in addition to the graphic. If you do not want to link text, but wish to specify one of the later parameters, pass an empty string.

$stayOnPage is a value that will be interpreted as a boolean. If true, the link will not open a new window. If false, the link will open a new window using Javascript. (Default is false.)

$width and $height are optional numerical parameters that will override the width and height of the popped up window, which may be useful for certain help topics with big pictures included.

=cut

sub help_open_topic {
    my ($topic, $text, $stayOnPage, $width, $height) = @_;
    $text = "" if (not defined $text);
    $stayOnPage = 0 if (not defined $stayOnPage);
    $width = 350 if (not defined $width);
    $height = 400 if (not defined $height);
    my $filename = $topic;
    $filename =~ s/ /_/g;

    my $template = "";
    my $link;

    if (!$stayOnPage)
    {
	$link = "javascript:void(open('/adm/help/${filename}.hlp', 'Help_for_$topic', 'menubar=0,toolbar=1,scrollbars=1,width=$width,height=$height'))";
    }
    else
    {
	$link = "/adm/help/${filename}.hlp";
    }

    # Add the text
    if ($text ne "")
    {
	$template .= "<a href=\"$link\">$text</a> ";
    }

    # Add the graphic
    $template .= <<"ENDTEMPLATE";
<a href="$link"><image src="/adm/help/gif/smallHelp.gif" border="0" alt="(Help: $topic)"></a>
ENDTEMPLATE

    return $template;

}

=pod

=item csv_translate($text) 

Translate $text to allow it to be output as a 'comma seperated values' 
format.

=cut

sub csv_translate {
    my $text = shift;
    $text =~ s/\"/\"\"/g;
    $text =~ s/\n//g;
    return $text;
}

###############################################################
##        Home server <option> list generating code          ##
###############################################################
#-------------------------------------------

=pod

=item get_domains()

Returns an array containing each of the domains listed in the hosts.tab
file.

=cut

#-------------------------------------------
sub get_domains {
    # The code below was stolen from "The Perl Cookbook", p 102, 1st ed.
    my @domains;
    my %seen;
    foreach (sort values(%Apache::lonnet::hostdom)) {
        push (@domains,$_) unless $seen{$_}++;
    }
    return @domains;
}

#-------------------------------------------

=pod

=item select_dom_form($defdom,$name)

Returns a string containing a <select name='$name' size='1'> form to 
allow a user to select the domain to preform an operation in.  
See loncreateuser.pm for an example invocation and use.

=cut

#-------------------------------------------
sub select_dom_form {
    my ($defdom,$name) = @_;
    my @domains = get_domains();
    my $selectdomain = "<select name=\"$name\" size=\"1\">\n";
    foreach (@domains) {
        $selectdomain.="<option value=\"$_\" ".
            ($_ eq $defdom ? 'selected' : '').
                ">$_</option>\n";
    }
    $selectdomain.="</select>";
    return $selectdomain;
}

#-------------------------------------------

=pod

=item get_home_servers($domain)

Returns a hash which contains keys like '103l3' and values like 
'kirk.lite.msu.edu'.  All of the keys will be for machines in the
given $domain.

=cut

#-------------------------------------------
sub get_home_servers {
    my $domain = shift;
    my %home_servers;
    foreach (keys(%Apache::lonnet::libserv)) {
        if ($Apache::lonnet::hostdom{$_} eq $domain) {
            $home_servers{$_} = $Apache::lonnet::hostname{$_};
        }
    }
    return %home_servers;
}

#-------------------------------------------

=pod

=item home_server_option_list($domain)

returns a string which contains an <option> list to be used in a 
<select> form input.  See loncreateuser.pm for an example.

=cut

#-------------------------------------------
sub home_server_option_list {
    my $domain = shift;
    my %servers = &get_home_servers($domain);
    my $result = '';
    foreach (sort keys(%servers)) {
        $result.=
            '<option value="'.$_.'">'.$_.' '.$servers{$_}."</option>\n";
    }
    return $result;
}
###############################################################
##    End of home server <option> list generating code       ##
###############################################################

###############################################################
##    Authentication changing form generation subroutines    ##
###############################################################
##
## All of the authform_xxxxxxx subroutines take their inputs in a
## hash, and have reasonable default values.
##
##    formname = the name given in the <form> tag.
#-------------------------------------------

=pod

=item authform_xxxxxx

The authform_xxxxxx subroutines provide javascript and html forms which 
handle some of the conveniences required for authentication forms.  
This is not an optimal method, but it works.  

See loncreateuser.pm for invocation and use examples.

=over 4

=item authform_header

=item authform_authorwarning

=item authform_nochange

=item authform_kerberos

=item authform_internal

=item authform_filesystem

=back

=cut

#-------------------------------------------
sub authform_header{  
    my %in = (
        formname => 'cu',
        kerb_def_dom => 'MSU.EDU',
        @_,
    );
    $in{'formname'} = 'document.' . $in{'formname'};
    my $result='';
    $result.=<<"END";
var current = new Object();
current.radiovalue = 'nochange';
current.argfield = null;

function changed_radio(choice,currentform) {
    var choicearg = choice + 'arg';
    // If a radio button in changed, we need to change the argfield
    if (current.radiovalue != choice) {
        current.radiovalue = choice;
        if (current.argfield != null) {
            currentform.elements[current.argfield].value = '';
        }
        if (choice == 'nochange') {
            current.argfield = null;
        } else {
            current.argfield = choicearg;
            switch(choice) {
                case 'krb': 
                    currentform.elements[current.argfield].value = 
                        "$in{'kerb_def_dom'}";
                break;
              default:
                break;
            }
        }
    }
    return;
}

function changed_text(choice,currentform) {
    var choicearg = choice + 'arg';
    if (currentform.elements[choicearg].value !='') {
        switch (choice) {
            case 'krb': currentform.elements[choicearg].value =
                currentform.elements[choicearg].value.toUpperCase();
                break;
            default:
        }
        // clear old field
        if ((current.argfield != choicearg) && (current.argfield != null)) {
            currentform.elements[current.argfield].value = '';
        }
        current.argfield = choicearg;
    }
    set_auth_radio_buttons(choice,currentform);
    return;
}

function set_auth_radio_buttons(newvalue,currentform) {
    var i=0;
    while (i < currentform.login.length) {
        if (currentform.login[i].value == newvalue) { break; }
        i++;
    }
    if (i == currentform.login.length) {
        return;
    }
    current.radiovalue = newvalue;
    currentform.login[i].checked = true;
    return;
}
END
    return $result;
}

sub authform_authorwarning{
    my $result='';
    $result=<<"END";
<i>As a general rule, only authors or co-authors should be filesystem
authenticated (which allows access to the server filesystem).</i>
END
    return $result;
}

sub authform_nochange{  
    my %in = (
              formname => 'document.cu',
              kerb_def_dom => 'MSU.EDU',
              @_,
          );
    my $result='';
    $result.=<<"END";
<input type="radio" name="login" value="nochange" checked="checked"
       onclick="javascript:changed_radio('nochange',$in{'formname'});">
Do not change login data
END
    return $result;
}

sub authform_kerberos{  
    my %in = (
              formname => 'document.cu',
              kerb_def_dom => 'MSU.EDU',
              @_,
              );
    my $result='';
    $result.=<<"END";
<input type="radio" name="login" value="krb" 
       onclick="javascript:changed_radio('krb',$in{'formname'});"
       onchange="javascript:changed_radio('krb',$in{'formname'});">
Kerberos authenticated with domain
<input type="text" size="10" name="krbarg" value=""
       onchange="javascript:changed_text('krb',$in{'formname'});">
END
    return $result;
}

sub authform_internal{  
    my %args = (
                formname => 'document.cu',
                kerb_def_dom => 'MSU.EDU',
                @_,
                );
    my $result='';
    $result.=<<"END";
<input type="radio" name="login" value="int"
       onchange="javascript:changed_radio('int',$args{'formname'});"
       onclick="javascript:changed_radio('int',$args{'formname'});">
Internally authenticated (with initial password 
<input type="text" size="10" name="intarg" value=""
       onchange="javascript:changed_text('int',$args{'formname'});">
END
    return $result;
}

sub authform_local{  
    my %in = (
              formname => 'document.cu',
              kerb_def_dom => 'MSU.EDU',
              @_,
              );
    my $result='';
    $result.=<<"END";
<input type="radio" name="login" value="loc"
       onchange="javascript:changed_radio('loc',$in{'formname'});"
       onclick="javascript:changed_radio('loc',$in{'formname'});"> 
Local Authentication with argument
<input type="text" size="10" name="locarg" value=""
       onchange="javascript:changed_text('loc',$in{'formname'});">
END
    return $result;
}

sub authform_filesystem{  
    my %in = (
              formname => 'document.cu',
              kerb_def_dom => 'MSU.EDU',
              @_,
              );
    my $result='';
    $result.=<<"END";
<input type="radio" name="login" value="fsys" 
       onchange="javascript:changed_radio('fsys',$in{'formname'});"
       onclick="javascript:changed_radio('fsys',$in{'formname'});"> 
Filesystem authenticated (with initial password 
<input type="text" size="10" name="fsysarg" value=""
       onchange="javascript:changed_text('fsys',$in{'formname'});">
END
    return $result;
}

###############################################################
##   End Authentication changing form generation functions   ##
###############################################################

###############################################################
##                Thesaurus Functions                        ##
###############################################################

=pod

=item initialize_keywords

Initializes the package variable %Keywords if it is empty.  Uses the
package variable $thesaurus_db_file.

=cut

###################################################

sub initialize_keywords {
    return 1 if (scalar keys(%Keywords));
    # If we are here, %Keywords is empty, so fill it up
    #   Make sure the file we need exists...
    if (! -e $thesaurus_db_file) {
        &Apache::lonnet::logthis("Attempt to access $thesaurus_db_file".
                                 " failed because it does not exist");
        return 0;
    }
    #   Set up the hash as a database
    my %thesaurus_db;
    if (! tie(%thesaurus_db,'GDBM_File',
              $thesaurus_db_file,&GDBM_READER,0640)){
        &Apache::lonnet::logthis("Could not tie \%thesaurus_db to ".
                                 $thesaurus_db_file);
        return 0;
    } 
    #  Get the average number of appearances of a word.
    my $avecount = $thesaurus_db{'average.count'};
    #  Put keywords (those that appear > average) into %Keywords
    while (my ($word,$data)=each (%thesaurus_db)) {
        my ($count,undef) = split /:/,$data;
        $Keywords{$word}++ if ($count > $avecount);
    }
    untie %thesaurus_db;
    # Remove special values from %Keywords.
    foreach ('total.count','average.count') {
        delete($Keywords{$_}) if (exists($Keywords{$_}));
    }
    return 1;
}

###################################################

=pod

=item keyword($word)

Returns true if $word is a keyword.  A keyword is a word that appears more 
than the average number of times in the thesaurus database.  Calls 
&initialize_keywords

=cut

###################################################

sub keyword {
    return if (!&initialize_keywords());
    my $word=lc(shift());
    $word=~s/\W//g;
    return exists($Keywords{$word});
}

###################################################
#         Old code, to be removed soon            #
###################################################
# -------------------------------------------------------- Return related words
#sub related {
#    my $newword=shift;
#    $newword=~s/\W//g;
#    $newword=~tr/A-Z/a-z/;
#    my $tindex=$theindex{$newword};
#    if ($tindex) {
#        my %found=();
#        foreach (split(/\,/,$therelated[$tindex])) {
## - Related word found
#            my ($ridx,$rcount)=split(/\:/,$_);
## - Direct relation index
#            my $directrel=$rcount/$thecount[$tindex];
#            if ($directrel>$thethreshold) {
#               foreach (split(/\,/,$therelated[$ridx])) {
#                  my ($rridx,$rrcount)=split(/\:/,$_);
#                  if ($rridx==$tindex) {
## - Determine reverse relation index
#                     my $revrel=$rrcount/$thecount[$ridx];
## - Calculate full index
#                     $found{$ridx}=$directrel*$revrel;
#                     if ($found{$ridx}>$thethreshold) {
#                        foreach (split(/\,/,$therelated[$ridx])) {
#                            my ($rrridx,$rrrcount)=split(/\:/,$_);
#                            unless ($found{$rrridx}) {
#                               my $revrevrel=$rrrcount/$thecount[$ridx];
#                               if (
#                          $directrel*$revrel*$revrevrel>$thethreshold
#                               ) {
#                                  $found{$rrridx}=
#                                       $directrel*$revrel*$revrevrel;
#                               }
#                            }
#                        }
#                     }
#                  }
#               }
#            }
#        }
#    }
#    return ();
#}

###############################################################

=pod 

=item get_related_words

Look up a word in the thesaurus.  Takes a scalar arguement and returns
an array of words.  If the keyword is not in the thesaurus, an empty array
will be returned.  The order of the words returned is determined by the
database which holds them.

Uses global $thesaurus_db_file.

=cut

###############################################################

sub get_related_words {
    my $keyword = shift;
    my %thesaurus_db;
    if (! -e $thesaurus_db_file) {
        &Apache::lonnet::logthis("Attempt to access $thesaurus_db_file ".
                                 "failed because the file does not exist");
        return ();
    }
    if (! tie(%thesaurus_db,'GDBM_File',
              $thesaurus_db_file,&GDBM_READER,0640)){
        return ();
    } 
    my @Words=();
    if (exists($thesaurus_db{$keyword})) {
        $_ = $thesaurus_db{$keyword};
        (undef,@Words) = split/:/;  # The first element is the number of times
                                    # the word appears.  We do not need it now.
        for (my $i=0;$i<=$#Words;$i++) {
            ($Words[$i],undef)= split/\,/,$Words[$i];
        }
    }
    untie %thesaurus_db;
    return @Words;
}

###############################################################
##              End Thesaurus Functions                      ##
###############################################################

# ---------------------------------------------------------------- Language IDs
sub languageids {
    return sort(keys(%language));
}

# -------------------------------------------------------- Language Description
sub languagedescription {
    return $language{shift(@_)};
}

# --------------------------------------------------------------- Copyright IDs
sub copyrightids {
    return sort(keys(%cprtag));
}

# ------------------------------------------------------- Copyright Description
sub copyrightdescription {
    return $cprtag{shift(@_)};
}

# ------------------------------------------------------------- File Categories
sub filecategories {
    return sort(keys(%category_extensions));
}

# -------------------------------------- File Types within a specified category
sub filecategorytypes {
    return @{$category_extensions{lc($_[0])}};
}

# ------------------------------------------------------------------ File Types
sub fileextensions {
    return sort(keys(%fe));
}

# ------------------------------------------------------------- Embedding Style
sub fileembstyle {
    return $fe{lc(shift(@_))};
}

# ------------------------------------------------------------ Description Text
sub filedescription {
    return $fd{lc(shift(@_))};
}

# ------------------------------------------------------------ Description Text
sub filedescriptionex {
    my $ex=shift;
    return '.'.$ex.' '.$fd{lc($ex)};
}

# ---- Retrieve attempts by students
# input
# $symb             - problem including path
# $username,$domain - that of the student
# $course           - course name
# $getattempt       - leave blank if want all attempts, else put something.
# $regexp           - regular expression. If string matches regexp send to
# $gradesub         - routine that process the string if it matches regexp
# 
# output
# formatted as a table all the attempts, if any.
#
sub get_previous_attempt {
  my ($symb,$username,$domain,$course,$getattempt,$regexp,$gradesub)=@_;
  my $prevattempts='';
  no strict 'refs';
  if ($symb) {
    my (%returnhash)=
      &Apache::lonnet::restore($symb,$course,$domain,$username);
    if ($returnhash{'version'}) {
      my %lasthash=();
      my $version;
      for ($version=1;$version<=$returnhash{'version'};$version++) {
        foreach (sort(split(/\:/,$returnhash{$version.':keys'}))) {
	  $lasthash{$_}=$returnhash{$version.':'.$_};
        }
      }
      $prevattempts='<table border="0" width="100%"><tr><td bgcolor="#777777">';
      $prevattempts.='<table border="0" width="100%"><tr bgcolor="#e6ffff"><td>History</td>';
      foreach (sort(keys %lasthash)) {
	my ($ign,@parts) = split(/\./,$_);
	if ($#parts > 0) {
	  my $data=$parts[-1];
	  pop(@parts);
	  $prevattempts.='<td>Part '.join('.',@parts).'<br />'.$data.'&nbsp;</td>';
	} else {
	  if ($#parts == 0) {
	    $prevattempts.='<th>'.$parts[0].'</th>';
	  } else {
	    $prevattempts.='<th>'.$ign.'</th>';
	  }
	}
      }
      if ($getattempt eq '') {
	for ($version=1;$version<=$returnhash{'version'};$version++) {
	  $prevattempts.='</tr><tr bgcolor="#ffffe6"><td>Transaction '.$version.'</td>';
	    foreach (sort(keys %lasthash)) {
	       my $value;
	       if ($_ =~ /timestamp/) {
		  $value=scalar(localtime($returnhash{$version.':'.$_}));
	       } else {
		  $value=$returnhash{$version.':'.$_};
	       }
	       $prevattempts.='<td>'.$value.'&nbsp;</td>';   
	    }
	 }
      }
      $prevattempts.='</tr><tr bgcolor="#ffffe6"><td>Current</td>';
      foreach (sort(keys %lasthash)) {
	my $value;
	if ($_ =~ /timestamp/) {
	  $value=scalar(localtime($lasthash{$_}));
	} else {
	  $value=$lasthash{$_};
	}
	if ($_ =~/$regexp$/ && (defined &$gradesub)) {$value = &$gradesub($value)}
	$prevattempts.='<td>'.$value.'&nbsp;</td>';
      }
      $prevattempts.='</tr></table></td></tr></table>';
    } else {
      $prevattempts='Nothing submitted - no attempts.';
    }
  } else {
    $prevattempts='No data.';
  }
}

sub get_student_view {
  my ($symb,$username,$domain,$courseid) = @_;
  my ($map,$id,$feedurl) = split(/___/,$symb);
  my (%old,%moreenv);
  my @elements=('symb','courseid','domain','username');
  foreach my $element (@elements) {
    $old{$element}=$ENV{'form.grade_'.$element};
    $moreenv{'form.grade_'.$element}=eval '$'.$element #'
  }
  &Apache::lonnet::appenv(%moreenv);
  my $userview=&Apache::lonnet::ssi('/res/'.$feedurl);
  &Apache::lonnet::delenv('form.grade_');
  foreach my $element (@elements) {
    $ENV{'form.grade_'.$element}=$old{$element};
  }
  $userview=~s/\<body[^\>]*\>//gi;
  $userview=~s/\<\/body\>//gi;
  $userview=~s/\<html\>//gi;
  $userview=~s/\<\/html\>//gi;
  $userview=~s/\<head\>//gi;
  $userview=~s/\<\/head\>//gi;
  $userview=~s/action\s*\=/would_be_action\=/gi;
  return $userview;
}

sub get_student_answers {
  my ($symb,$username,$domain,$courseid) = @_;
  my ($map,$id,$feedurl) = split(/___/,$symb);
  my (%old,%moreenv);
  my @elements=('symb','courseid','domain','username');
  foreach my $element (@elements) {
    $old{$element}=$ENV{'form.grade_'.$element};
    $moreenv{'form.grade_'.$element}=eval '$'.$element #'
  }
  $moreenv{'form.grade_target'}='answer';
  &Apache::lonnet::appenv(%moreenv);
  my $userview=&Apache::lonnet::ssi('/res/'.$feedurl);
  &Apache::lonnet::delenv('form.grade_');
  foreach my $element (@elements) {
    $ENV{'form.grade_'.$element}=$old{$element};
  }
  return $userview;
}

###############################################

###############################################

sub get_unprocessed_cgi {
  my ($query,$possible_names)= @_;
  # $Apache::lonxml::debug=1;
  foreach (split(/&/,$query)) {
    my ($name, $value) = split(/=/,$_);
    $name = &Apache::lonnet::unescape($name);
    if (!defined($possible_names) || (grep {$_ eq $name} @$possible_names)) {
      $value =~ tr/+/ /;
      $value =~ s/%([a-fA-F0-9][a-fA-F0-9])/pack("C",hex($1))/eg;
      &Apache::lonxml::debug("Seting :$name: to :$value:");
      unless (defined($ENV{'form.'.$name})) { &add_to_env('form.'.$name,$value) };
    }
  }
}

sub cacheheader {
  unless ($ENV{'request.method'} eq 'GET') { return ''; }
  my $date=strftime("%a, %d %b %Y %H:%M:%S GMT",gmtime);
  my $output .='<meta HTTP-EQUIV="Expires" CONTENT="'.$date.'" />
                <meta HTTP-EQUIV="Cache-control" CONTENT="no-cache" />
                <meta HTTP-EQUIV="Pragma" CONTENT="no-cache" />';
  return $output;
}

sub no_cache {
  my ($r) = @_;
  unless ($ENV{'request.method'} eq 'GET') { return ''; }
  #my $date=strftime("%a, %d %b %Y %H:%M:%S GMT",gmtime);
  $r->no_cache(1);
  $r->header_out("Pragma" => "no-cache");
  #$r->header_out("Expires" => $date);
}

sub add_to_env {
  my ($name,$value)=@_;
  if (defined($ENV{$name})) {
    if (ref($ENV{$name})) {
      #already have multiple values
      push(@{ $ENV{$name} },$value);
    } else {
      #first time seeing multiple values, convert hash entry to an arrayref
      my $first=$ENV{$name};
      undef($ENV{$name});
      push(@{ $ENV{$name} },$first,$value);
    }
  } else {
    $ENV{$name}=$value;
  }
}

=pod

=back 

=head2 CSV Upload/Handling functions

=over 4

=item  upfile_store($r)

Store uploaded file, $r should be the HTTP Request object,
needs $ENV{'form.upfile'}
returns $datatoken to be put into hidden field

=cut

sub upfile_store {
    my $r=shift;
    $ENV{'form.upfile'}=~s/\r/\n/gs;
    $ENV{'form.upfile'}=~s/\f/\n/gs;
    $ENV{'form.upfile'}=~s/\n+/\n/gs;
    $ENV{'form.upfile'}=~s/\n+$//gs;

    my $datatoken=$ENV{'user.name'}.'_'.$ENV{'user.domain'}.
	'_enroll_'.$ENV{'request.course.id'}.'_'.time.'_'.$$;
    {
	my $fh=Apache::File->new('>'.$r->dir_config('lonDaemons').
				 '/tmp/'.$datatoken.'.tmp');
	print $fh $ENV{'form.upfile'};
    }
    return $datatoken;
}

=item load_tmp_file($r)

Load uploaded file from tmp, $r should be the HTTP Request object,
needs $ENV{'form.datatoken'},
sets $ENV{'form.upfile'} to the contents of the file

=cut

sub load_tmp_file {
    my $r=shift;
    my @studentdata=();
    {
	my $fh;
	if ($fh=Apache::File->new($r->dir_config('lonDaemons').
				  '/tmp/'.$ENV{'form.datatoken'}.'.tmp')) {
	    @studentdata=<$fh>;
	}
    }
    $ENV{'form.upfile'}=join('',@studentdata);
}

=item upfile_record_sep()

Separate uploaded file into records
returns array of records,
needs $ENV{'form.upfile'} and $ENV{'form.upfiletype'}

=cut

sub upfile_record_sep {
    if ($ENV{'form.upfiletype'} eq 'xml') {
    } else {
	return split(/\n/,$ENV{'form.upfile'});
    }
}

=item record_sep($record)

Separate a record into fields $record should be an item from the upfile_record_sep(), needs $ENV{'form.upfiletype'}

=cut

sub record_sep {
    my $record=shift;
    my %components=();
    if ($ENV{'form.upfiletype'} eq 'xml') {
    } elsif ($ENV{'form.upfiletype'} eq 'space') {
        my $i=0;
        foreach (split(/\s+/,$record)) {
            my $field=$_;
            $field=~s/^(\"|\')//;
            $field=~s/(\"|\')$//;
            $components{$i}=$field;
            $i++;
        }
    } elsif ($ENV{'form.upfiletype'} eq 'tab') {
        my $i=0;
        foreach (split(/\t+/,$record)) {
            my $field=$_;
            $field=~s/^(\"|\')//;
            $field=~s/(\"|\')$//;
            $components{$i}=$field;
            $i++;
        }
    } else {
        my @allfields=split(/\,/,$record);
        my $i=0;
        my $j;
        for ($j=0;$j<=$#allfields;$j++) {
            my $field=$allfields[$j];
            if ($field=~/^\s*(\"|\')/) {
		my $delimiter=$1;
                while (($field!~/$delimiter$/) && ($j<$#allfields)) {
		    $j++;
		    $field.=','.$allfields[$j];
		}
                $field=~s/^\s*$delimiter//;
                $field=~s/$delimiter\s*$//;
            }
            $components{$i}=$field;
	    $i++;
        }
    }
    return %components;
}

=item upfile_select_html()

return HTML code to select file and specify its type

=cut

sub upfile_select_html {
    return (<<'ENDUPFORM');
<input type="file" name="upfile" size="50">
<br />Type: <select name="upfiletype">
<option value="csv">CSV (comma separated values, spreadsheet)</option>
<option value="space">Space separated</option>
<option value="tab">Tabulator separated</option>
<option value="xml">HTML/XML</option>
</select>
ENDUPFORM
}

=item csv_print_samples($r,$records)

Prints a table of sample values from each column uploaded $r is an
Apache Request ref, $records is an arrayref from
&Apache::loncommon::upfile_record_sep

=cut

sub csv_print_samples {
    my ($r,$records) = @_;
    my (%sone,%stwo,%sthree);
    %sone=&record_sep($$records[0]);
    if (defined($$records[1])) {%stwo=&record_sep($$records[1]);}
    if (defined($$records[2])) {%sthree=&record_sep($$records[2]);}

    $r->print('Samples<br /><table border="2"><tr>');
    foreach (sort({$a <=> $b} keys(%sone))) { $r->print('<th>Column&nbsp;'.($_+1).'</th>'); }
    $r->print('</tr>');
    foreach my $hash (\%sone,\%stwo,\%sthree) {
	$r->print('<tr>');
	foreach (sort({$a <=> $b} keys(%sone))) {
	    $r->print('<td>');
	    if (defined($$hash{$_})) { $r->print($$hash{$_}); }
	    $r->print('</td>');
	}
	$r->print('</tr>');
    }
    $r->print('</tr></table><br />'."\n");
}

=item csv_print_select_table($r,$records,$d)

Prints a table to create associations between values and table columns.
$r is an Apache Request ref,
$records is an arrayref from &Apache::loncommon::upfile_record_sep,
$d is an array of 2 element arrays (internal name, displayed name)

=cut

sub csv_print_select_table {
    my ($r,$records,$d) = @_;
    my $i=0;my %sone;
    %sone=&record_sep($$records[0]);
    $r->print('Associate columns with student attributes.'."\n".
	     '<table border="2"><tr><th>Attribute</th><th>Column</th></tr>'."\n");
    foreach (@$d) {
	my ($value,$display)=@{ $_ };
	$r->print('<tr><td>'.$display.'</td>');

	$r->print('<td><select name=f'.$i.
		  ' onchange="javascript:flip(this.form,'.$i.');">');
	$r->print('<option value="none"></option>');
	foreach (sort({$a <=> $b} keys(%sone))) {
	    $r->print('<option value="'.$_.'">Column '.($_+1).'</option>');
	}
	$r->print('</select></td></tr>'."\n");
	$i++;
    }
    $i--;
    return $i;
}

=item csv_samples_select_table($r,$records,$d)

Prints a table of sample values from the upload and can make associate samples to internal names.

$r is an Apache Request ref,
$records is an arrayref from &Apache::loncommon::upfile_record_sep,
$d is an array of 2 element arrays (internal name, displayed name)

=cut

sub csv_samples_select_table {
    my ($r,$records,$d) = @_;
    my %sone; my %stwo; my %sthree;
    my $i=0;

    $r->print('<table border=2><tr><th>Field</th><th>Samples</th></tr>');
    %sone=&record_sep($$records[0]);
    if (defined($$records[1])) {%stwo=&record_sep($$records[1]);}
    if (defined($$records[2])) {%sthree=&record_sep($$records[2]);}

    foreach (sort keys %sone) {
	$r->print('<tr><td><select name=f'.$i.
		  ' onchange="javascript:flip(this.form,'.$i.');">');
	foreach (@$d) {
	    my ($value,$display)=@{ $_ };
	    $r->print('<option value='.$value.'>'.$display.'</option>');
	}
	$r->print('</select></td><td>');
	if (defined($sone{$_})) { $r->print($sone{$_}."</br>\n"); }
	if (defined($stwo{$_})) { $r->print($stwo{$_}."</br>\n"); }
	if (defined($sthree{$_})) { $r->print($sthree{$_}."</br>\n"); }
	$r->print('</td></tr>');
	$i++;
    }
    $i--;
    return($i);
}
1;
__END__;

=pod

=back

=head2 Access .tab File Data

=over 4

=item languageids() 

returns list of all language ids

=item languagedescription() 

returns description of a specified language id

=item copyrightids() 

returns list of all copyrights

=item copyrightdescription() 

returns description of a specified copyright id

=item filecategories() 

returns list of all file categories

=item filecategorytypes() 

returns list of file types belonging to a given file
category

=item fileembstyle() 

returns embedding style for a specified file type

=item filedescription() 

returns description for a specified file type

=item filedescriptionex() 

returns description for a specified file type with
extra formatting

=back

=head2 Alternate Problem Views

=over 4

=item get_previous_attempt() 

return string with previous attempt on problem

=item get_student_view() 

show a snapshot of what student was looking at

=item get_student_answers() 

show a snapshot of how student was answering problem

=back

=head2 HTTP Helper

=over 4

=item get_unprocessed_cgi($query,$possible_names)

Modify the %ENV hash to contain unprocessed CGI form parameters held in
$query.  The parameters listed in $possible_names (an array reference),
will be set in $ENV{'form.name'} if they do not already exist.

Typically called with $ENV{'QUERY_STRING'} as the first parameter.  
$possible_names is an ref to an array of form element names.  As an example:
get_unprocessed_cgi($ENV{'QUERY_STRING'},['uname','udom']);
will result in $ENV{'form.uname'} and $ENV{'form.udom'} being set.

=item cacheheader() 

returns cache-controlling header code

=item nocache() 

specifies header code to not have cache

=item add_to_env($name,$value) 

adds $name to the %ENV hash with value
$value, if $name already exists, the entry is converted to an array
reference and $value is added to the array.

=back

=cut

FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>