getSimpleHTMLDOM($url);
if(!$html)
$this->returnServerError('Could not acquire content from url: ' . $url . '!');
$content = $html->find('article', 0);
if(!$content)
$this->returnServerError('Could not find \'section\'!');
$text = preg_replace('##is', '', $content->innertext);
$text = strip_tags($text, '
');
return $text;
}
public function collectData(){
$html = $this->getSimpleHTMLDOM('http://www.nicematin.com/derniere-minute/rss') or $this->returnServerError('Could not request NiceMatin.');
$limit = 0;
foreach($html->find('item') as $element) {
if($limit < 10) {
// We need to fix the 'link' tag as simplehtmldom cannot parse it (just rename it and load back as dom)
$element_text = $element->outertext;
$element_text = str_replace('', '', $element_text);
$element_text = str_replace('', '', $element_text);
$element = str_get_html($element_text);
$item = array();
$item['title'] = $element->find('title', 0)->innertext;
$item['uri'] = $element->find('url', 0)->innertext;
$item['timestamp'] = strtotime($element->find('pubDate', 0)->plaintext);
$item['content'] = $this->NiceMatinExtractContent($item['uri']);
$this->items[] = $item;
$limit++;
}
}
}
}