source: main/trunk/greenstone2/perllib/plugins/PDFPlugin.pm@ 32277

###########################################################################
#
# PDFPlugin.pm -- reasonably with-it pdf plugin
# A component of the Greenstone digital library software
# from the New Zealand Digital Library Project at the 
# University of Waikato, New Zealand.
#
# Copyright (C) 1999-2001 New Zealand Digital Library Project
#
# This program 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.
#
# This program 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 this program; if not, write to the Free Software
# Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
#
###########################################################################
package PDFPlugin;

use strict;
no strict 'refs'; # so we can use a var for filehandles (e.g. STDERR)
no strict 'subs'; # allow filehandles to be variables and viceversa

use ReadTextFile;
use unicode;
use Mojo::DOM; # for HTML parsing

use AutoLoadConverters;
use ConvertBinaryFile;

@PDFPlugin::ISA = ('ConvertBinaryFile', 'AutoLoadConverters', 'ReadTextFile');


my $convert_to_list =
    [ { 'name' => "auto",
    'desc' => "{ConvertBinaryFile.convert_to.auto}" },
      { 'name' => "html",
    'desc' => "{ConvertBinaryFile.convert_to.html}" },
      { 'name' => "text",
    'desc' => "{ConvertBinaryFile.convert_to.text}" },
      { 'name' => "paged_html",
    'desc' => "{PDFPlugin.convert_to.paged_html}"},
      { 'name' => "pagedimg_jpg",
    'desc' => "{ConvertBinaryFile.convert_to.pagedimg_jpg}"},
      { 'name' => "pagedimg_gif",
    'desc' => "{ConvertBinaryFile.convert_to.pagedimg_gif}"},
      { 'name' => "pagedimg_png",
    'desc' => "{ConvertBinaryFile.convert_to.pagedimg_png}"}, 
      ];


my $arguments = 
    [
     { 'name' => "convert_to",
       'desc' => "{ConvertBinaryFile.convert_to}",
       'type' => "enum",
       'reqd' => "yes",
       'list' => $convert_to_list, 
       'deft' => "html" },   
     { 'name' => "process_exp",
       'desc' => "{BaseImporter.process_exp}",
       'type' => "regexp",
       'deft' => &get_default_process_exp(),
       'reqd' => "no" },
     { 'name' => "block_exp",
       'desc' => "{CommonUtil.block_exp}",
       'type' => "regexp",
       'deft' => &get_default_block_exp() },
     { 'name' => "metadata_fields",
       'desc' => "{HTMLPlugin.metadata_fields}",
       'type' => "string",
       'deft' => "Title,Author,Subject,Keywords" },
      { 'name' => "metadata_field_separator",
    'desc' => "{HTMLPlugin.metadata_field_separator}",
    'type' => "string",
    'deft' => "" },
     { 'name' => "noimages",
       'desc' => "{PDFPlugin.noimages}",
       'type' => "flag" },
     { 'name' => "allowimagesonly",
       'desc' => "{PDFPlugin.allowimagesonly}",
       'type' => "flag" },
     { 'name' => "complex",
       'desc' => "{PDFPlugin.complex}",
       'type' => "flag" },
     { 'name' => "nohidden",
       'desc' => "{PDFPlugin.nohidden}",
       'type' => "flag" },
     { 'name' => "zoom",
       'desc' => "{PDFPlugin.zoom}",
       'deft' => "2",
#       'range' => "1,3", # actually the range is 0.5-3 
       'type' => "string" },
     { 'name' => "use_sections",
       'desc' => "{PDFPlugin.use_sections}",
       'type' => "flag" },
     { 'name' => "description_tags",
       'desc' => "{HTMLPlugin.description_tags}",
       'type' => "flag" },
      { 'name' => "use_realistic_book",
        'desc' => "{PDFPlugin.use_realistic_book}",
    'type' => "flag"}
     ];

my $options = { 'name'     => "PDFPlugin",
        'desc'     => "{PDFPlugin.desc}",
        'abstract' => "no",
        'inherits' => "yes",
        'srcreplaceable' => "yes", # Source docs in PDF can be replaced with GS-generated html      
        'args'     => $arguments };

sub new {
    my ($class) = shift (@_);
    my ($pluginlist,$inputargs,$hashArgOptLists) = @_;
    push(@$pluginlist, $class);

    push(@$inputargs,"-title_sub");
    push(@$inputargs,'^(Page\s+\d+)?(\s*1\s+)?');

    push(@{$hashArgOptLists->{"ArgList"}},@{$arguments});
    push(@{$hashArgOptLists->{"OptList"}},$options);

    my $auto_converter_self = new AutoLoadConverters($pluginlist,$inputargs,$hashArgOptLists,["PDFBoxConverter"],1);
    my $cbf_self = new ConvertBinaryFile($pluginlist, $inputargs, $hashArgOptLists);
    my $self = BaseImporter::merge_inheritance($auto_converter_self, $cbf_self);
    
    if ($self->{'info_only'}) {
    # don't worry about any options etc
    return bless $self, $class;
    }
    
    $self = bless $self, $class;
    $self->{'file_type'} = "PDF";

    # PDFPlugin is deprecated and migrating users should hereafter choose between
    # PDFv1Plugin, if they want to use the old pdftohtml tool's capabilities,
    # and PDFv2Plugin, if they want to use pdfbox or the new xpdftools capabilities.
    &gsprintf::gsprintf(STDERR, "{PDFPlugin.deprecated_plugin}\n");
    
    # these are passed through to gsConvert.pl by ConvertBinaryFile.pm
    my $zoom = $self->{"zoom"};
    # By default, PDFPlugin assumes gsConvert.pl will run the old pdftohtml conversion tool,
    # But if pdfbox conversion is turned on, the tool used is pdfbox (which is presently an
    # AutoLoadConverter and therefore bypasses gsConvert.pl)
    $self->{'convert_options'} = "-pdf_tool pdftohtml"; 
    $self->{'convert_options'} .= " -pdf_zoom $zoom";
    $self->{'convert_options'} .= " -pdf_complex" if $self->{"complex"};
    $self->{'convert_options'} .= " -pdf_nohidden" if $self->{"nohidden"};
    $self->{'convert_options'} .= " -pdf_ignore_images" if $self->{"noimages"};
    $self->{'convert_options'} .= " -pdf_allow_images_only" if $self->{"allowimagesonly"};

    # check convert_to
    if ($self->{'convert_to'} eq "text" && $ENV{'GSDLOS'} =~ /^windows$/i) {
    &gsprintf::gsprintf(STDERR, "{PDFPlugin.win_old_pdftotext_unsupported}\n");
    $self->{'convert_to'} = "html";
    }
    elsif ($self->{'convert_to'} eq "auto") {
    # choose html ?? is this the best option
    $self->{'convert_to'} = "paged_html";
    }
    if ($self->{'use_realistic_book'}) {
    if ($self->{'convert_to'} ne "html") {
        print STDERR "PDFs will be converted to HTML for realistic book functionality\n";
        $self->{'convert_to'} = "html";
    }
    }
    # set convert_to_plugin and convert_to_ext
    $self->set_standard_convert_settings();

    my $secondary_plugin_name = $self->{'convert_to_plugin'};
    my $secondary_plugin_options = $self->{'secondary_plugin_options'};

    if (!defined $secondary_plugin_options->{$secondary_plugin_name}) {
    $secondary_plugin_options->{$secondary_plugin_name} = [];
    }
    my $specific_options = $secondary_plugin_options->{$secondary_plugin_name};

    # following title_sub removes "Page 1" added by pdftohtml, and a leading
    # "1", which is often the page number at the top of the page. Bad Luck
    # if your document title actually starts with "1 " - is there a better way?
    push(@$specific_options , "-title_sub", '^(Page\s+\d+)?(\s*1\s+)?');
    my $associate_tail_re = $self->{'associate_tail_re'};
    if ((defined $associate_tail_re) && ($associate_tail_re ne "")) {
    push(@$specific_options, "-associate_tail_re", $associate_tail_re);
    }
    push(@$specific_options, "-file_rename_method", "none");
    
    if ($secondary_plugin_name eq "HTMLPlugin") {
    # pdftohtml always produces utf8 - What about pdfbox???
    # push(@$specific_options, "-input_encoding", "utf8");
    push(@$specific_options, "-extract_language") if $self->{'extract_language'};
    push(@$specific_options, "-processing_tmp_files");
    # Instruct HTMLPlug (when eventually accessed through read_into_doc_obj) 
    # to extract these metadata fields from the HEAD META fields
    if (defined $self->{'metadata_fields'} && $self->{'metadata_fields'} =~ /\S/) {
        push(@$specific_options,"-metadata_fields",$self->{'metadata_fields'});
    } else {
        push(@$specific_options,"-metadata_fields","Title,GENERATOR,date,author<Creator>");
    }
    if (defined $self->{'metadata_field_separator'} && $self->{'metadata_field_separator'} =~ /\S/) {
        push(@$specific_options,"-metadata_field_separator",$self->{'metadata_field_separator'});
    }
    if ($self->{'use_sections'} || $self->{'description_tags'}) {
        $self->{'description_tags'} = 1;
        push(@$specific_options, "-description_tags");
    }
    if ($self->{'use_realistic_book'}) {
        push(@$specific_options, "-use_realistic_book");
    }
        if($self->{'convert_to'} eq "paged_html") { # for paged html, the default should be to sectionalise on headings the single superpage containing divs representing individual pages as section
            push(@$specific_options, "sectionalise_using_h_tags");
        }
    }
    elsif ($secondary_plugin_name eq "PagedImagePlugin") {
    push(@$specific_options, "-screenviewsize", "1000");
    push(@$specific_options, "-enable_cache");
    push(@$specific_options, "-processing_tmp_files");
    }

    $self = bless $self, $class;
    $self->load_secondary_plugins($class,$secondary_plugin_options,$hashArgOptLists);
    return $self;
}

sub get_default_process_exp {
    my $self = shift (@_);

    return q^(?i)\.pdf$^;
}

# so we don't inherit HTMLPlug's block exp...
sub get_default_block_exp {
    return "";
}

sub init {
    my $self = shift (@_);

    # ConvertBinaryFile init
    $self->SUPER::init(@_);
    $self->AutoLoadConverters::init(@_);

}

sub begin {
    my $self = shift (@_);

    $self->AutoLoadConverters::begin(@_);
    $self->SUPER::begin(@_);

}

sub deinit {
    my $self = shift (@_);
    
    $self->AutoLoadConverters::deinit(@_);
    $self->SUPER::deinit(@_);

}
  
# By setting hashing to be on ga xml this ensures that two 
# PDF files that are identical except for the metadata
# to hash to different values. Without this, when each PDF
# file is converted to HTML there is a chance that they 
# will both be *identical* if the conversion utility does
# not embed the metadata in the generated HTML. This is
# certainly the case when PDFBOX is being used. 

# This change makes this convert to based plugin more 
# consistent with the original vision that the same document 
# with different metadata should
# be seen as different.

sub get_oid_hash_type {
    my $self = shift (@_);
    return "hash_on_ga_xml";
}
  
  
sub tmp_area_convert_file {

    my $self = shift (@_);
    return $self->AutoLoadConverters::tmp_area_convert_file(@_);

}

# Overriding to do some extra handling for paged_html output mode
sub run_conversion_command {
    my $self = shift (@_);
    my ($tmp_dirname, $tmp_inputPDFname, $utf8_tailname, $lc_suffix, $tailname, $suffix) = @_;    

    if($self->{'convert_to'} ne "paged_html") {
    return $self->ConvertBinaryFile::run_conversion_command(@_);
    }

    # if output mode is paged_html, we use Xpdf tools' pdftohtml and tell it
    # to create a subdir called "pages" in the tmp area to puts its products
    # in there. (Xpdf's pdftohtml needs to be passed a *non-existent* directory
    # parameter, the "pages" subdir). If Xpdf's pdftohtml has successfully run,
    # the intermediary output file tmp/<random-num>/pages/index.html should
    # exist (besides other output products there)

    # We let ConvertBinaryFile proceed normally, but the return value should reflect
    # that on success it should expect the intermediary product tmpdir/pages/index.html
    # (which is the product of xpdftohtml conversion).
    my $output_filename = $self->ConvertBinaryFile::run_conversion_command(@_);
    $output_filename = &FileUtils::filenameConcatenate($tmp_dirname, "pages", "index.html");

    # However, when convert_post_process() is done, it should have output the final
    # product of the paged_html conversion: an html file of the same name and in the
    # same tmp location as the input PDF file.

    my ($name_prefix, $output_dir, $ext)
    = &File::Basename::fileparse($tmp_inputPDFname, "\\.[^\\.]+\$");
    $self->{'conv_filename_after_post_process'} = &FileUtils::filenameConcatenate($output_dir, $name_prefix.".html");
#    print STDERR "@@@@@ final paged html file will be: " . $self->{'conv_filename_after_post_process'} . "\n";

    return $output_filename;
}

sub convert_post_process
{
    my $self = shift (@_);
    my ($conv_filename) = @_;

    my $outhandle=$self->{'outhandle'};

    if($self->{'convert_to'} eq "paged_html") {
    # special post-processing for paged_html mode, as HTML pages generated
    # by xpdf's pdftohtml need to be massaged into the form we want 
    $self->xpdftohtml_convert_post_process($conv_filename);
    }
    else { # use PDFPlugin's usual post processing
    $self->default_convert_post_process($conv_filename);
    }
}

# Called after gsConvert.pl has been run to convert a PDF to paged_html
# using Xpdftools' pdftohtml
# This method will do some cleanup of the HTML files produced after XPDF has produced
# an HTML doc for each PDF page: it first gets rid of the default index.html.
# Instead, it constructs a single html page containing each original HTML page
# <body> nested as divs instead, with simple section information inserted at the top
# of each 'page' <div> and some further styling customisation. This HTML manipulation
# is to be done with the Mojo::DOM perl package.
# Note that since xpdf's pdftohtml would have failed if the output dir already
# existed and for simpler naming, the output files are created in a new "pages"
# subdirectory of the tmp location parent of $conv_filename instead
sub xpdftohtml_convert_post_process
{
    my $self = shift (@_);
    my ($pages_index_html) = @_; # = tmp/<rand>/pages/index.html for paged_html output mode
    my $output_filename = $self->{'conv_filename_after_post_process'};

    # Read in all the html files in tmp's "pages" subdir, except for index.html.
    # and use it to create a new html file called $self->{'conv_filename_after_post_process'}
    # which will consist of a slightly modified version of
    # each of the other html files concatenated together.

    my $outhandle=$self->{'outhandle'};

    my ($tailname, $pages_subdir, $suffix)
    = &File::Basename::fileparse($pages_index_html, "\\.[^\\.]+\$");

    # Code from util::create_itemfile()
    # Read in all the files
    opendir(DIR, $pages_subdir) || die "can't opendir $pages_subdir: $!";
    my @page_files = grep {-f "$pages_subdir/$_"} readdir(DIR);
    closedir DIR;
    # Sort files in the directory by page_num
    # files are named index.html, page1.html, page2.html, ..., pagen.html
    sub page_number {
    my ($dir) = @_;
    my ($pagenum) =($dir =~ m/^page(\d+)\.html?$/i);
    $pagenum = 0 unless defined $pagenum; # index.html will be given pagenum=0
    return $pagenum;
    }
    # sort the files in the directory in the order of page_num rather than lexically.
    @page_files = sort { page_number($a) <=> page_number($b) } @page_files;

    #my $num_html_pages = (scalar(@page_files) - 1)/2; # skip index file.
              # For every html file there's an img file, so halve the total num.
              # What about other file types that may potentially be there too???
    my $num_html_pages = 0;
    foreach my $pagefile (@page_files) {
    $num_html_pages++ if $pagefile =~ m/\.html?$/ && $pagefile !~ /^index\.html?/i; 
    }

    # Prepare to create our new html page that will contain all the individual
    # htmls generated by xpdf's pdftohtml in sequence.
    # First write the opening html tags out to the output file. These are the
    # same tags and their contents, including <meta>, as is generated by 
    # Xpdf's pdftohtml for each of its individual html pages.
    my $start_text = "<html>\n<head>\n";
    my ($output_tailname, $tmp_subdir, $html_suffix)
    = &File::Basename::fileparse($output_filename, "\\.[^\\.]+\$");
    $start_text .= "<title>$output_tailname</title>\n";
    $start_text .= "<meta http-equiv=\"Content-Type\" content=\"text/html; charset=UTF-8\">\n";
    $start_text .= "</head>\n<body>\n\n";
    $start_text .= "<h1>$output_tailname</h1>\n\n";

    #handle content encodings the same way that default_convert_post_process does
    # $self->utf8_write_file ($start_text, $conv_filename); # will close file after write    
    # Don't want to build a giant string in memory of all the pages concatenated
    # and then write it out in one go. Instead, build up the final single page
    # by writing each modified paged_html file out to it as this is processed.
    # Copying file open/close code from CommonUtil::utf8_write_file()
    if (!open (OUTFILE, ">:utf8", $output_filename)) {
    gsprintf(STDERR, "PDFPlugin::xpdftohtml_convert_post_process {CommonUtil.could_not_open_for_writing} ($!)\n", $output_filename);
    die "\n";
    }
    print OUTFILE $start_text;

    # Get the contents of each individual HTML page generated by Xpdf, after first
    # modifying each, and write each out into our single all-encompassing html
    foreach my $pagefile (@page_files) {
    if ($pagefile =~ m/\.html?$/ && $pagefile !~ /^index\.html?/i) {
        my $page_num = page_number($pagefile);    
        # get full path to pagefile
        $pagefile = &FileUtils::filenameConcatenate($pages_subdir, $pagefile);
#       print STDERR "@@@ About to process html file $pagefile (num $page_num)\n";
        my $modified_page_contents = $self->_process_paged_html_page($pagefile, $page_num, $num_html_pages);
        print OUTFILE "$modified_page_contents\n\n";
    }
    }

    # we've now created a single HTML file by concatenating (a modified version)
    # of each paged html file
    print OUTFILE "</body>\n</html>\n"; # write out closing tags
    close OUTFILE; # done

    # Get rid of all the htm(l) files incl index.html in the associated "pages"
    # subdir, since we've now processed them all into a single html file
    # one folder level up and we don't want HTMLPlugin to process all of them next.
    &FileUtils::removeFilesFiltered($pages_subdir, "\.html?\$"); #  no specific whitelist, but blacklist htm(l)

    # now the tmp area should contain a single html file contain all the html pages'
    # contents in sequence, and a "pages" subdir containing the screenshot images
    # of each page.    
    # HTMLPlugin will process these further in the plugin pipeline
}

# For whatever reason, most html <tags> don't get printed out in GLI
# So when debugging, use this function to print them out as [tags] instead.
sub _debug_print_html
{
    my $self = shift (@_);
    my ($string_or_dom) = @_;

    # can't seem to determine type of string with ref/reftype
    # https://stackoverflow.com/questions/1731333/how-do-i-tell-what-type-of-value-is-in-a-perl-variable
    # Not needed, as $dom objects seem to get correctly stringified in string contexts
    # $dom.to_string/$dom.stringify seem to get called, no need to call them
    # https://stackoverflow.com/questions/5214543/what-is-stringification-in-perl
    my $escapedTxt = $string_or_dom; 
    $escapedTxt =~ s@\<@[@sg;
    $escapedTxt =~ s@\>@]@sg;

    print STDERR "#### $escapedTxt\n";
}

# Helper function to read in each paged_html generated by Xpdf's pdftohtml
# then modify the html suitably using the HTML parsing functions offered by 
# Mojo::DOM, then return the modified HTML content as a string
# See https://mojolicious.org/perldoc/Mojo/DOM
sub _process_paged_html_page
{
    my $self = shift (@_);
    my ($pagefile, $page_num, $num_html_pages) = @_;

    my $text = "";

    # handling content encoding the same way default_convert_post_process does
    $self->read_file ($pagefile, "utf8", "", \$text);

    my $dom = Mojo::DOM->new($text);

#    $self->_debug_print_html($dom);

    # there's a <style> element on the <html>, we need to shift it into the <div>
    # tag that we'll be creating. We'll first slightly modify the <style> element
    # store the first style element, which is the only one and in the <body>
    # we'll later insert it as child of an all-encompassing div that we'll create
    my $page_style_tag_str = $dom->at('html')->at('style')->to_string;
    # In the style tag, convert id style references to class style references
    my $css_class = ".p".$page_num."f";
    $page_style_tag_str =~ s@\#f@$css_class@sg;
    my $style_element = Mojo::DOM->new($page_style_tag_str)->at('style'); # modified    
#$self->_debug_print_html($style_element);

    # need to know the image's height to set the height of the surrounding
    # div that's to replace this page's <body>:
    my $img_height = $dom->find('img')->[0]{height};

    # 2. Adjust the img#background src attribute to point to the pages subdir for imgs
    # 3. Set that img tag's class=background, and change its id to background+$page_num
    my $bg_img_tag=$dom->find('img#background')->[0];
    my $img_src_str = $bg_img_tag->{src};
    $img_src_str = "pages/$img_src_str";
    $bg_img_tag->attr(src => $img_src_str); # reset
#$self->_debug_print_html($bg_img_tag);
    # set both class and modified id attributes in one step:
    $bg_img_tag->attr({class => "background", id => "background".$page_num});
#$self->_debug_print_html($bg_img_tag);

    # get all the <span> nested inside <div class="txt"> elements and
    # 1. set their class attr to be "p + page_num + id-of-the-span",
    # 2. then delete the id, because the span ids have been reused when element
    # ids ought to be unique. Which is why we set the modified ids to be the
    # value of the class attribute instead
    $dom->find('div.txt span')->each(sub { 
    $_->attr(class => "p". $page_num. $_->{id});
    delete $_->{id};
                     }); # both changes done in one find() operation
#$self->_debug_print_html($dom->find('div.txt span')->last);

    # Finally can create our new dom, starting with a div tag for the current page
    # Must be: <div id="$page_num" style="position:relative; height:$img_height;"/>
#    my $new_dom = Mojo::DOM->new_tag('div', id => "page".$page_num, style => "position: relative; height: ".$img_height."px;" )
    my $new_dom = Mojo::DOM->new_tag('div', style => "position: relative; height: ".$img_height."px;" );
#$self->_debug_print_html($new_dom);
    $new_dom->at('div')->append_content($style_element)->root;


#$self->_debug_print_html($new_dom);
    # Copy across all the old html's body tag's child nodes into the new dom's new div tag
    $dom->at('body')->child_nodes->each(sub { $new_dom->at('div')->append_content($_)}); #$_->to_string
#$self->_debug_print_html($new_dom);


    # build up the outer div with the <h>tags for sectionalising
    my $inner_div_str = $new_dom->to_string;

    my $page_div = "<div id=\"page".$page_num."\">\n";
    # Append a page range bucket heading if applicable: if we have more than 10 pages
    # to display in the current bucket AND we're on the first page of each bucket of 10 pages.
    # Dr Bainbridge thinks for now we need only consider PDFs where the
    # total number of pages < 1000 and create buckets of size 10 (e.g. 1-10, ... 51-60, ...)
    # If number of remaining pages >= 10, then create new bucket heading
    # e.g. "Pages 30-40"
    if(($page_num % 10) == 1 && ($num_html_pages - $page_num) > 10) {
    # Double-digit page numbers that start with 2
    # i.e. 21 to 29 (and 30) should be in 21 to 30 range
    my $start_range = $page_num - ($page_num % 10) + 1;
    my $end_range = $page_num + 10 - ($page_num % 10);
    $page_div .= "<h2 style=\"font-size:1em;font-weight:normal;\">Pages ".$start_range . "-" . $end_range."</h2>\n";
    }

    # No sectionalising for 10 pages or under. Otherwise, every page is a section too, not just buckets
    if($num_html_pages > 10) {
        # Whether we're starting a new bucket or not, add a simpler heading: just the pagenumber, "Page #"  
        $page_div .= "<h3 style=\"font-size:1em;font-weight:normal;\">Page ".$page_num."</h3>\n";       
    }

    $page_div .= $inner_div_str;
    $page_div .= "\n</div>";

    # Finished processing a single html page of the paged_html output generated by
    # Xpdf's pdftohtml: finished massaging that single html page into the right form
    return $page_div;
}

# This subroutine is called to do the PDFPlugin post-processing for all cases
# except the "paged_html" conversion mode. This is what PDFPlugin always used to do:
sub default_convert_post_process
{
    my $self = shift (@_);
    my ($conv_filename) = @_;
    my $outhandle=$self->{'outhandle'};

    #$self->{'input_encoding'} = "utf8"; # The output is always in utf8 (is it?? it is for html, but what about other types?)
    #my ($language, $encoding) = $self->textcat_get_language_encoding ($conv_filename);

    # read in file ($text will be in utf8)
    my $text = "";
    # encoding will be utf8 for html files - what about other types? will we do this step for them anyway?
    $self->read_file ($conv_filename, "utf8", "", \$text);

    # To support the use_sections option with PDFBox: Greenstone splits PDFs into pages for 
    # sections. The PDFPlugin code wants each new page to be prefixed with <a name=pagenum></a>,
    # which it then splits on to generate page-based sections. However, that's not what PDFBox 
    # generates in its HTML output. Fortunately, PDFBox does have its own page-separator: it
    # embeds each page in an extra div. The div opener is: 
    # <div style=\"page-break-before:always; page-break-after:always\">
    # The PDFPlugin now looks for this and prefixes <a name=0></a> to each such div. (The 
    # pagenumber is fixed at 0 since I'm unable to work out how to increment the pagenum during 
    # a regex substitution even with regex extensions on.) Later, when we process each section 
    # to get the pagenum, PDFBox's output for this is pre-processed by having a loopcounter 
    # that increments the pagenum for each subsequent section.

    #$pdfbox_pageheader="\<div style=\"page-break-before:always; page-break-after:always\">";
    my $loopcounter = 0; # used later on!
    $text =~ s@\<div style=\"page-break-before:always; page-break-after:always\">@<a name=$loopcounter></a><div style=\"page-break-before:always; page-break-after:always\">@g;


    # Calculate number of pages based on <a ...> tags (we have a <a name=1> etc
    # for each page).  Metadata based on this calculation not set until process()
    # 
    # Note: this is done even if we are not breaking the document into pages as it might
    # be useful to give an indication of document length in browser through setting
    # num_pages as metadata.
    # Clean html from low and hight surrogates D800–DFFF
    $text =~ s@[\N{U+D800}-\N{U+DFFF}]@\ @g;
    my @pages = ($text =~ m/\<[Aa] name=\"?\w+\"?>/ig); #<div style=\"?page-break-before:always; page-break-after:always\"?>
    my $num_pages = scalar(@pages);
    $self->{'num_pages'} = $num_pages;

    if ($self->{'use_sections'}
    && $self->{'converted_to'} eq "HTML") {

    print $outhandle "PDFPlugin: Calculating sections...\n";

    # we have "<a name=1></a>" etc for each page
    # it may be <A name=
    my @sections = split('<[Aa] name=', $text);

    my $top_section = "";

    if (scalar (@sections) == 1) { #only one section - no split!
        print $outhandle "PDFPlugin: warning - no sections found\n";
    } else {
        $top_section .= shift @sections; # keep HTML header etc as top_section
    }

    # handle first section specially for title? Or all use first 100...
    
    my $title = $sections[0];
    $title =~ s/^\"?\w+\"?>//; # specific for pdftohtml...
    $title =~ s/<\/([^>]+)><\1>//g; # (eg) </b><b> - no space
    $title =~ s/<[^>]*>/ /g;
    $title =~ s/(?:&nbsp;|\xc2\xa0)/ /g; # utf-8 for nbsp...
    $title =~ s/^\s+//s;
    $title =~ s/\s+$//;
    $title =~ s/\s+/ /gs;
    $title =~ s/^$self->{'title_sub'}// if ($self->{'title_sub'});
    $title =~ s/^\s+//s; # in case title_sub introduced any...
    $title = substr ($title, 0, 100);
    $title =~ s/\s\S*$/.../;


    if (scalar (@sections) == 1) { # no sections found
        $top_section .= $sections[0];
        @sections=();
    } else {
        $top_section .= "<!--<Section>\n<Metadata name=\"Title\">$title</Metadata>\n-->\n <!--</Section>-->\n";
    }

    # add metadata per section...
    foreach my $section (@sections) {
        # section names are not always just digits, may be like "outline"
        $section =~ s@^\"?(\w+)\"?></a>@@; # leftover from split expression...

        $title = $1; # Greenstone does magic if sections are titled digits

        # A title of pagenum=0 means use_sections is being applied on output from PDFBox,
        # which didn't originally have a <a name=incremented pagenumber></a> to split each page. 
        # Our Perl code then prefixed <a name=0></a> to it. Now need to increment the pagenum here:
        if($loopcounter > 0 || ($title eq 0 && $loopcounter == 0)) { # implies use_sections with PDFBox
        $title = ++$loopcounter;
        }

        if (! defined($title) ) {
        print STDERR "no title: $section\n";
        $title = " "; # get rid of the undefined warning in next line
        }
        my $newsection = "<!-- from PDFPlugin -->\n<!-- <Section>\n";
        $newsection .= "<Metadata name=\"Title\">" . $title
        . "</Metadata>\n--><br />\n";
        $newsection .= $section;
        $newsection .= "<!--</Section>-->\n";
        $section = $newsection;
    }

    $text=join('', ($top_section, @sections));
    }

    if ($self->{'use_sections'}
    && $self->{'converted_to'} eq "text") {
    print STDERR "**** When converting PDF to text, cannot apply use_sections\n";
    }


    # The following should no longer be needed, now that strings
    # read in are Unicode aware (in the Perl sense) rather than
    # raw binary strings that just happen to be UTF-8 compliant

    # turn any high bytes that aren't valid utf-8 into utf-8.
##    unicode::ensure_utf8(\$text);

    # Write it out again!
    $self->utf8_write_file (\$text, $conv_filename);
}


# do plugin specific processing of doc_obj for HTML type
sub process {
    my $self = shift (@_);
    my ($pluginfo, $base_dir, $file, $metadata, $doc_obj, $gli) = @_;

    my $result = $self->process_type($base_dir,$file,$doc_obj);

    # fix up the extracted date metadata to be in Greenstone date format,
    # and fix the capitalisation of 'date'
    my $cursection = $doc_obj->get_top_section();
    foreach my $datemeta (@{$doc_obj->get_metadata($cursection, "date")}) {
    $doc_obj->delete_metadata($cursection, "date", $datemeta);

    # We're just interested in the date bit, not the time
    # some pdf creators (eg "Acrobat 5.0 Scan Plug-in for Windows")
    # set a /CreationDate, and set /ModDate to 000000000. pdftohtml
    # extracts the ModDate, so it is 0...
    $datemeta =~ /(\d+)-(\d+)-(\d+)/;
    my ($year, $month, $day) = ($1,$2,$3);
    if (defined($year) && defined($month) && defined($day)) {
        if ($year == 0) {next}
        if ($year < 100) {$year += 1900} # just to be safe
        if ($month =~ /^\d$/) {$month="0$month"} # single digit
        if ($day =~ /^\d$/) {$day="0$day"} # single digit
        my $date="$year$month$day";
        $doc_obj->add_utf8_metadata($cursection, "Date", $date);
    }
    }

    $doc_obj->add_utf8_metadata($cursection, "NumPages", $self->{'num_pages'}) if defined $self->{'num_pages'};
    
    if ($self->{'use_sections'} && $self->{'converted_to'} eq "HTML") {
    # For gs2 we explicitly make it a paged document, cos greenstone won't get it
    # right if any section has an empty title, or one with letters in it
    if (&util::is_gs3()) {
        # but for gs3, paged docs currently use image slider which is ugly if there are no images
        $doc_obj->set_utf8_metadata_element ($cursection, "gsdlthistype", "Hierarchy");
    } else {
        $doc_obj->set_utf8_metadata_element ($cursection, "gsdlthistype", "Paged");
    }
    }

    return $result;
}

1;