The url of the parsed page * string 'type' => Content type * string 'title' => The title of the content * string 'text' => The description for the content * string 'image' => A preview image of the content (only available * if $no_geuessing = false * array'images' = Array of preview pictures * string 'keywords' => The tags which belong to the content * * @see ParseUrl::getSiteinfo() for more information about scraping * embeddable content */ public static function getSiteinfoCached($url, $no_guessing = false, $do_oembed = true) { if ($url == "") { return false; } $parsed_url = DBA::selectFirst('parsed_url', ['content'], ['url' => normalise_link($url), 'guessing' => !$no_guessing, 'oembed' => $do_oembed] ); if (!empty($parsed_url['content'])) { $data = unserialize($parsed_url['content']); return $data; } $data = self::getSiteinfo($url, $no_guessing, $do_oembed); DBA::insert( 'parsed_url', [ 'url' => normalise_link($url), 'guessing' => !$no_guessing, 'oembed' => $do_oembed, 'content' => serialize($data), 'created' => DateTimeFormat::utcNow() ], true ); return $data; } /** * @brief Parse a page for embeddable content information * * This method parses to url for meta data which can be used to embed * the content. If available it prioritizes Open Graph meta tags. * If this is not available it uses the twitter cards meta tags. * As fallback it uses standard html elements with meta informations * like \Awesome Title\ or * \ * * @param string $url The url of the page which should be scraped * @param bool $no_guessing If true the parse doens't search for * preview pictures * @param bool $do_oembed The false option is used by the function fetch_oembed() * to avoid endless loops * @param int $count Internal counter to avoid endless loops * * @return array which contains needed data for embedding * string 'url' => The url of the parsed page * string 'type' => Content type * string 'title' => The title of the content * string 'text' => The description for the content * string 'image' => A preview image of the content (only available * if $no_geuessing = false * array'images' = Array of preview pictures * string 'keywords' => The tags which belong to the content * * @todo https://developers.google.com/+/plugins/snippet/ * @verbatim * * * * * *

Shiny Trinket

* *

Shiny trinkets are shiny.

* * @endverbatim */ public static function getSiteinfo($url, $no_guessing = false, $do_oembed = true, $count = 1) { $a = get_app(); $siteinfo = []; // Check if the URL does contain a scheme $scheme = parse_url($url, PHP_URL_SCHEME); if ($scheme == '') { $url = 'http://' . trim($url, '/'); } if ($count > 10) { logger('Endless loop detected for ' . $url, LOGGER_DEBUG); return $siteinfo; } $url = trim($url, "'"); $url = trim($url, '"'); $url = Network::stripTrackingQueryParams($url); $siteinfo['url'] = $url; $siteinfo['type'] = 'link'; $data = Network::curl($url); if (!$data['success']) { return $siteinfo; } // If the file is too large then exit if ($data['info']['download_content_length'] > 1000000) { return $siteinfo; } // If it isn't a HTML file then exit if (($data['info']['content_type'] != '') && !strstr(strtolower($data['info']['content_type']), 'html')) { return $siteinfo; } $header = $data['header']; $body = $data['body']; if ($do_oembed) { $oembed_data = OEmbed::fetchURL($url); if (!empty($oembed_data->type)) { if (!in_array($oembed_data->type, ['error', 'rich', ''])) { $siteinfo['type'] = $oembed_data->type; } if (($oembed_data->type == 'link') && ($siteinfo['type'] != 'photo')) { if (isset($oembed_data->title)) { $siteinfo['title'] = trim($oembed_data->title); } if (isset($oembed_data->description)) { $siteinfo['text'] = trim($oembed_data->description); } if (isset($oembed_data->thumbnail_url)) { $siteinfo['image'] = $oembed_data->thumbnail_url; } } } } // Fetch the first mentioned charset. Can be in body or header $charset = ''; if (preg_match('/charset=(.*?)[\'"\s\n]/', $header, $matches)) { $charset = trim(trim(trim(array_pop($matches)), ';,')); } if ($charset == '') { $charset = 'utf-8'; } if (($charset != '') && (strtoupper($charset) != 'UTF-8')) { logger('detected charset ' . $charset, LOGGER_DEBUG); $body = iconv($charset, 'UTF-8//TRANSLIT', $body); } $body = mb_convert_encoding($body, 'HTML-ENTITIES', 'UTF-8'); $doc = new DOMDocument(); @$doc->loadHTML($body); XML::deleteNode($doc, 'style'); XML::deleteNode($doc, 'script'); XML::deleteNode($doc, 'option'); XML::deleteNode($doc, 'h1'); XML::deleteNode($doc, 'h2'); XML::deleteNode($doc, 'h3'); XML::deleteNode($doc, 'h4'); XML::deleteNode($doc, 'h5'); XML::deleteNode($doc, 'h6'); XML::deleteNode($doc, 'ol'); XML::deleteNode($doc, 'ul'); $xpath = new DOMXPath($doc); $list = $xpath->query('//meta[@content]'); foreach ($list as $node) { $meta_tag = []; if ($node->attributes->length) { foreach ($node->attributes as $attribute) { $meta_tag[$attribute->name] = $attribute->value; } } if (@$meta_tag['http-equiv'] == 'refresh') { $path = $meta_tag['content']; $pathinfo = explode(';', $path); $content = ''; foreach ($pathinfo as $value) { if (substr(strtolower($value), 0, 4) == 'url=') { $content = substr($value, 4); } } if ($content != '') { $siteinfo = self::getSiteinfo($content, $no_guessing, $do_oembed, ++$count); return $siteinfo; } } } $list = $xpath->query('//title'); if ($list->length > 0) { $siteinfo['title'] = trim($list->item(0)->nodeValue); } $list = $xpath->query('//meta[@name]'); foreach ($list as $node) { $meta_tag = []; if ($node->attributes->length) { foreach ($node->attributes as $attribute) { $meta_tag[$attribute->name] = $attribute->value; } } if (empty($meta_tag['content'])) { continue; } $meta_tag['content'] = trim(html_entity_decode($meta_tag['content'], ENT_QUOTES, 'UTF-8')); switch (strtolower($meta_tag['name'])) { case 'fulltitle': $siteinfo['title'] = trim($meta_tag['content']); break; case 'description': $siteinfo['text'] = trim($meta_tag['content']); break; case 'thumbnail': $siteinfo['image'] = $meta_tag['content']; break; case 'twitter:image': $siteinfo['image'] = $meta_tag['content']; break; case 'twitter:image:src': $siteinfo['image'] = $meta_tag['content']; break; case 'twitter:card': // Convert Twitter types in our own if ($meta_tag['content'] == 'summary_large_image') { $siteinfo['type'] = 'photo'; } else { $siteinfo['type'] = $meta_tag['content']; } break; case 'twitter:description': $siteinfo['text'] = trim($meta_tag['content']); break; case 'twitter:title': $siteinfo['title'] = trim($meta_tag['content']); break; case 'dc.title': $siteinfo['title'] = trim($meta_tag['content']); break; case 'dc.description': $siteinfo['text'] = trim($meta_tag['content']); break; case 'keywords': $keywords = explode(',', $meta_tag['content']); break; case 'news_keywords': $keywords = explode(',', $meta_tag['content']); break; } } if (isset($keywords)) { $siteinfo['keywords'] = []; foreach ($keywords as $keyword) { if (!in_array(trim($keyword), $siteinfo['keywords'])) { $siteinfo['keywords'][] = trim($keyword); } } } $list = $xpath->query('//meta[@property]'); foreach ($list as $node) { $meta_tag = []; if ($node->attributes->length) { foreach ($node->attributes as $attribute) { $meta_tag[$attribute->name] = $attribute->value; } } if (!empty($meta_tag['content'])) { $meta_tag['content'] = trim(html_entity_decode($meta_tag['content'], ENT_QUOTES, 'UTF-8')); switch (strtolower($meta_tag['property'])) { case 'og:image': $siteinfo['image'] = $meta_tag['content']; break; case 'og:title': $siteinfo['title'] = trim($meta_tag['content']); break; case 'og:description': $siteinfo['text'] = trim($meta_tag['content']); break; } } } if ((@$siteinfo['image'] == '') && !$no_guessing) { $list = $xpath->query('//img[@src]'); foreach ($list as $node) { $img_tag = []; if ($node->attributes->length) { foreach ($node->attributes as $attribute) { $img_tag[$attribute->name] = $attribute->value; } } $src = self::completeUrl($img_tag['src'], $url); $photodata = Image::getInfoFromURL($src); if (($photodata) && ($photodata[0] > 150) && ($photodata[1] > 150)) { if ($photodata[0] > 300) { $photodata[1] = round($photodata[1] * (300 / $photodata[0])); $photodata[0] = 300; } if ($photodata[1] > 300) { $photodata[0] = round($photodata[0] * (300 / $photodata[1])); $photodata[1] = 300; } $siteinfo['images'][] = [ 'src' => $src, 'width' => $photodata[0], 'height' => $photodata[1] ]; } } } elseif (!empty($siteinfo['image'])) { $src = self::completeUrl($siteinfo['image'], $url); unset($siteinfo['image']); $photodata = Image::getInfoFromURL($src); if (($photodata) && ($photodata[0] > 10) && ($photodata[1] > 10)) { $siteinfo['images'][] = ['src' => $src, 'width' => $photodata[0], 'height' => $photodata[1]]; } } if ((@$siteinfo['text'] == '') && (@$siteinfo['title'] != '') && !$no_guessing) { $text = ''; $list = $xpath->query('//div[@class="article"]'); foreach ($list as $node) { if (strlen($node->nodeValue) > 40) { $text .= ' ' . trim($node->nodeValue); } } if ($text == '') { $list = $xpath->query('//div[@class="content"]'); foreach ($list as $node) { if (strlen($node->nodeValue) > 40) { $text .= ' ' . trim($node->nodeValue); } } } // If none text was found then take the paragraph content if ($text == '') { $list = $xpath->query('//p'); foreach ($list as $node) { if (strlen($node->nodeValue) > 40) { $text .= ' ' . trim($node->nodeValue); } } } if ($text != '') { $text = trim(str_replace(["\n", "\r"], [' ', ' '], $text)); while (strpos($text, ' ')) { $text = trim(str_replace(' ', ' ', $text)); } $siteinfo['text'] = trim(html_entity_decode(substr($text, 0, 350), ENT_QUOTES, 'UTF-8') . '...'); } } logger('Siteinfo for ' . $url . ' ' . print_r($siteinfo, true), LOGGER_DEBUG); Addon::callHooks('getsiteinfo', $siteinfo); return $siteinfo; } /** * @brief Convert tags from CSV to an array * * @param string $string Tags * @return array with formatted Hashtags */ public static function convertTagsToArray($string) { $arr_tags = str_getcsv($string); if (count($arr_tags)) { // add the # sign to every tag array_walk($arr_tags, ["self", "arrAddHashes"]); return $arr_tags; } } /** * @brief Add a hasht sign to a string * * This method is used as callback function * * @param string $tag The pure tag name * @param int $k Counter for internal use * @return void */ private static function arrAddHashes(&$tag, $k) { $tag = "#" . $tag; } /** * @brief Add a scheme to an url * * The src attribute of some html elements (e.g. images) * can miss the scheme so we need to add the correct * scheme * * @param string $url The url which possibly does have * a missing scheme (a link to an image) * @param string $scheme The url with a correct scheme * (e.g. the url from the webpage which does contain the image) * * @return string The url with a scheme */ private static function completeUrl($url, $scheme) { $urlarr = parse_url($url); // If the url does allready have an scheme // we can stop the process here if (isset($urlarr["scheme"])) { return($url); } $schemearr = parse_url($scheme); $complete = $schemearr["scheme"]."://".$schemearr["host"]; if (@$schemearr["port"] != "") { $complete .= ":".$schemearr["port"]; } if (strpos($urlarr["path"], "/") !== 0) { $complete .= "/"; } $complete .= $urlarr["path"]; if (@$urlarr["query"] != "") { $complete .= "?".$urlarr["query"]; } if (@$urlarr["fragment"] != "") { $complete .= "#".$urlarr["fragment"]; } return($complete); } }