* @license GNU/GPL */ /** * Defines the "OpenofficeText" child of class "learnpath" */ require_once 'openoffice_document.class.php'; if (api_get_setting('search_enabled') == 'true') { require_once api_get_path(LIBRARY_PATH).'specific_fields_manager.lib.php'; require_once api_get_path(LIBRARY_PATH).'search/ChamiloIndexer.class.php'; require_once api_get_path(LIBRARY_PATH).'search/IndexableChunk.class.php'; } /** * @package chamilo.learnpath.OpenofficeDocument */ class OpenofficeText extends OpenofficeDocument { public $split_steps; /** * Class constructor. Calls the parent class and initialises the local attribute split_steps * @param boolean Whether to split steps (true) or make one large page (false) * @param string Course code * @param integer Resource ID * @param integer Creator user id */ public function __construct($split_steps = false, $course_code = null, $resource_id = null, $user_id = null) { $this -> split_steps = $split_steps; parent::__construct($course_code, $resource_id, $user_id); } /** * Gets html pages and compose them into a learning path * @param array The files that will compose the generated learning path. Unused so far. * @return boolean False if file does not exit. Nothing otherwise. */ public function make_lp($files = array()) { $_course = api_get_course_info(); // We get a content where ||page_break|| indicates where the page is broken. if (!file_exists($this->base_work_dir.'/'.$this->created_dir.'/'.$this->file_name.'.html')) { return false; } $content = file_get_contents($this->base_work_dir.'/'.$this->created_dir.'/'.$this->file_name.'.html'); unlink($this->base_work_dir.'/'.$this->file_path); unlink($this->base_work_dir.'/'.$this->created_dir.'/'.$this->file_name.'.html'); // The file is utf8 encoded and it seems to make problems with special quotes. // Then we htmlentities that, we replace these quotes and html_entity_decode that in good charset. $charset = api_get_system_encoding(); $content = api_htmlentities($content, ENT_COMPAT, $this->original_charset); $content = str_replace('’', '\'', $content); $content = api_convert_encoding($content, $charset, $this->original_charset); $content = str_replace($this->original_charset, $charset, $content); $content = api_html_entity_decode($content, ENT_COMPAT, $charset); // Set the path to pictures to absolute (so that it can be modified in fckeditor). $content = preg_replace("|src=\"([^\"]*)|i", "src=\"".api_get_path(REL_COURSE_PATH).$_course['path'].'/document'.$this->created_dir."/\\1", $content); list($header, $body) = explode('\r\n"; $style_to_import .= '@import "'.api_get_path(WEB_CODE_PATH).'css/'.$my_style.'/default.css";'."\n"; $style_to_import .= "\r\n"; $header = preg_replace("||i", "\r\n$style_to_import\r\n\\0", $header); // Line break before and after picture. $header = str_replace('p {', 'p {clear:both;', $header); $header = str_replace('absolute', 'relative', $header); switch ($this->split_steps) { case 'per_page': $this -> dealPerPage($header, $body); break; case 'per_chapter': $this -> dealPerChapter($header, $body); break; } } /** * Manages dir/chapter splitting * @param string Chapter header * @param string Content * @return void */ public function dealPerChapter($header, $content) { $_course = api_get_course_info(); $content = str_replace('||page_break||', '', $content); // Get all the h1. preg_match_all("|]*>([^(h1)+]*)|is", $content, $matches_temp); // Empty the fake chapters. $new_index = 0; for ($i = 0; $i < count($matches_temp[0]); $i++) { if (trim($matches_temp[1][$i]) !== '') { $matches[0][$new_index] = $matches_temp[0][$i]; $matches[1][$new_index] = $matches_temp[1][$i]; $new_index++; } } // Add intro item. $intro_content = substr($content, 0, strpos($content, $matches[0][0])); $items_to_create[get_lang('Introduction')] = $intro_content; for ($i = 0; $i < count($matches[0]); $i++) { if (empty($matches[1][$i])) continue; $content = strstr($content, $matches[0][$i]); if ($i + 1 !== count($matches[0])) { $dir_content = substr($content, 0, strpos($content, $matches[0][$i + 1])); } else { $dir_content = $content; } $items_to_create[$matches[1][$i]] = $dir_content; } $i = 0; foreach ($items_to_create as $item_title => $item_content) { $i++; $page_content = $this->format_page_content($header, $item_content); $html_file = $this->created_dir.'-'.$i.'.html'; $handle = fopen($this->base_work_dir.$this->created_dir.'/'.$html_file, 'w+'); fwrite($handle, $page_content); fclose($handle); $document_id = add_document( $_course, $this->created_dir.'/'.$html_file, 'file', filesize( $this->base_work_dir.$this->created_dir.'/'.$html_file ), $html_file ); if ($document_id) { // Put the document in item_property update. api_item_property_update( $_course, TOOL_DOCUMENT, $document_id, 'DocumentAdded', $_SESSION['_uid'], 0, 0, null, null, api_get_session_id() ); $infos = pathinfo($this->filepath); $slide_name = strip_tags(nl2br($item_title)); $slide_name = str_replace(array("\r\n", "\r", "\n"), '', $slide_name); $slide_name = html_entity_decode($slide_name); $previous = learnpath::add_item( 0, $previous, 'document', $document_id, $slide_name, '' ); if ($this->first_item == 0) { $this->first_item = $previous; } } } } /** * Manages page splitting * @param string Page header * @param string Page body * @return void */ public function dealPerPage($header, $body) { $_course = api_get_course_info(); // Split document to pages. $pages = explode('||page_break||', $body); $first_item = 0; foreach ($pages as $key => $page_content) { // For every pages, we create a new file. $key += 1; $page_content = $this->format_page_content($header, $page_content, $this->base_work_dir.$this->created_dir); $html_file = $this->created_dir.'-'.$key.'.html'; $handle = fopen($this->base_work_dir.$this->created_dir.'/'.$html_file, 'w+'); fwrite($handle, $page_content); fclose($handle); $document_id = add_document( $_course, $this->created_dir.$html_file, 'file', filesize($this->base_work_dir.$this->created_dir.$html_file), $html_file ); $slide_name = ''; if ($document_id) { // Put the document in item_property update. api_item_property_update( $_course, TOOL_DOCUMENT, $document_id, 'DocumentAdded', $_SESSION['_uid'], 0, 0, null, null, api_get_session_id() ); $infos = pathinfo($this->filepath); $slide_name = 'Page '.str_repeat('0', 2 - strlen($key)).$key; $previous = learnpath::add_item(0, $previous, 'document', $document_id, $slide_name, ''); if ($this->first_item == 0) { $this->first_item = $previous; } // Code for text indexing. if (api_get_setting('search_enabled') == 'true') { if (isset($_POST['index_document']) && $_POST['index_document']) { //Display::display_normal_message(print_r($_POST)); $di = new ChamiloIndexer(); isset($_POST['language']) ? $lang = Database::escape_string($_POST['language']) : $lang = 'english'; $di->connectDb(NULL, NULL, $lang); $ic_slide = new IndexableChunk(); $ic_slide->addValue('title', $slide_name); $specific_fields = get_specific_field_list(); $all_specific_terms = ''; foreach ($specific_fields as $specific_field) { if (isset($_REQUEST[$specific_field['code']])) { $sterms = trim($_REQUEST[$specific_field['code']]); $all_specific_terms .= ' '.$sterms; if (!empty($sterms)) { $sterms = explode(',', $sterms); foreach ($sterms as $sterm) { $ic_slide->addTerm(trim($sterm), $specific_field['code']); } } } } $page_content = $all_specific_terms.' '.$page_content; $ic_slide->addValue('content', $page_content); // Add a comment to say terms separated by commas. $courseid = api_get_course_id(); $ic_slide->addCourseId($courseid); $ic_slide->addToolId(TOOL_LEARNPATH); $lp_id = $this->lp_id; $xapian_data = array( SE_COURSE_ID => $courseid, SE_TOOL_ID => TOOL_LEARNPATH, SE_DATA => array('lp_id' => $lp_id, 'lp_item'=> $previous, 'document_id' => $document_id), SE_USER => (int) api_get_user_id(), ); $ic_slide->xapian_data = serialize($xapian_data); $di->addChunk($ic_slide); // Index and return search engine document id. $did = $di->index(); if ($did) { // Save it to db. $tbl_se_ref = Database::get_main_table(TABLE_MAIN_SEARCH_ENGINE_REF); $sql = 'INSERT INTO %s (id, course_code, tool_id, ref_id_high_level, ref_id_second_level, search_did) VALUES (NULL , \'%s\', \'%s\', %s, %s, %s)'; $sql = sprintf($sql, $tbl_se_ref, api_get_course_id(), TOOL_LEARNPATH, $lp_id, $previous, $did); Database::query($sql); } } } } } } /** * Returns additional Java command parameters * @return string The additional parameters to be used in the Java call */ public function add_command_parameters() { return ' -d woogie "'.$this->base_work_dir.'/'.$this->file_path.'" "'.$this->base_work_dir.$this->created_dir.'/'.$this->file_name.'.html"'; } /** * Formats a page content by reorganising the HTML code a little * @param string Page header * @param string Page content * @return string Formatted page content */ public function format_page_content($header, $content) { // Limit the width of the doc. list($max_width, $max_height) = explode('x', api_get_setting('service_ppt2lp', 'size')); $content = preg_replace("|]*>|i", "\\0\r\n
", $content, -1, $count); if ($count < 1) { $content = '
'.$content; } $content = preg_replace('||i', '
\\0', $content, -1, $count); if ($count < 1) { $content = $content.'
'; } // Add the headers. $content = $header.$content; // Resize all the picture to the max_width-10 preg_match_all("|]*>|i", $content, $images); foreach ($images[1] as $key => $image) { // Check if the base_work_dir.$this->created_dir.'/'.$image); $new_width = $max_width - 10; if ($img_width > $new_width) { $picture_resized = str_ireplace(' $max_width - 10) { $picture_resized = str_ireplace('width='.$img_width, 'width="'.($max_width - 10).'"', $images[0][$key]); $content = str_replace($images[0][$key], $picture_resized, $content); } } return $content; } /** * Add documents to the visioconference (to be implemented) */ public function add_docs_to_visio() { } }