rss-bridge/bridges/ZeitBridge.php

137 lines
4.4 KiB
PHP

<?php
class ZeitBridge extends FeedExpander
{
const MAINTAINER = 'Mynacol';
const NAME = 'Zeit Online Bridge';
const URI = 'https://www.zeit.de/';
const CACHE_TIMEOUT = 1800; // 30min
const DESCRIPTION = 'Returns the full articles instead of only the intro';
const PARAMETERS = [[
'category' => [
'name' => 'Category',
'type' => 'list',
'values' => [
'Startseite'
=> 'https://newsfeed.zeit.de/index',
'Politik'
=> 'https://newsfeed.zeit.de/politik/index',
'Wirtschaft'
=> 'https://newsfeed.zeit.de/wirtschaft/index',
'Gesellschaft'
=> 'https://newsfeed.zeit.de/gesellschaft/index',
'Kultur'
=> 'https://newsfeed.zeit.de/kultur/index',
'Wissen'
=> 'https://newsfeed.zeit.de/wissen/index',
'Digital'
=> 'https://newsfeed.zeit.de/digital/index',
'ZEIT Campus ONLINE'
=> 'https://newsfeed.zeit.de/campus/index',
'ZEIT ONLINE Arbeit'
=> 'https://newsfeed.zeit.de/arbeit/index',
'ZEIT Magazin ONLINE'
=> 'https://newsfeed.zeit.de/zeit-magazin/index',
'Entdecken'
=> 'https://newsfeed.zeit.de/entdecken/index',
'Mobilität'
=> 'https://newsfeed.zeit.de/mobilitaet/index',
'Sport'
=> 'https://newsfeed.zeit.de/sport/index',
'Alle Inhalte'
=> 'https://newsfeed.zeit.de/all'
]
],
'limit' => [
'name' => 'Limit',
'type' => 'number',
'required' => false,
'title' => 'Specify number of full articles to return',
'defaultValue' => 5
]
]];
public function collectData()
{
$url = $this->getInput('category');
$limit = $this->getInput('limit') ?: 5;
$this->collectExpandableDatas($url, $limit);
}
protected function parseItem(array $item)
{
$item['enclosures'] = [];
$headers = [
'Cookie: zonconsent=' . date('Y-m-d\TH:i:s.v\Z'),
];
// one-page article
$article = getSimpleHTMLDOM($item['uri'], $headers);
if ($article->find('a[href="' . $item['uri'] . '/komplettansicht"]', 0)) {
$item['uri'] .= '/komplettansicht';
$article = getSimpleHTMLDOM($item['uri'], $headers);
}
$article = defaultLinkTo($article, $item['uri']);
$item = $this->parseArticle($item, $article);
return $item;
}
private function parseArticle($item, $article)
{
$article = $article->find('main', 0);
// remove known bad elements
foreach (
$article->find(
'aside, .visually-hidden, .carousel-container, #tickaroo-liveblog, .zplus-badge, .article-heading__container--podcast'
) as $bad
) {
$bad->remove();
}
// reload html, as remove() is buggy
$article = str_get_html($article->outertext);
// podcast audio, if available
$podcast_src = $article->find('.article-heading__podcast audio[src]', 0);
if ($podcast_src) {
$item['enclosures'][] = $podcast_src->src;
}
// full res images
foreach ($article->find('img[data-src]') as $img) {
$img->src = $img->getAttribute('data-src');
$item['enclosures'][] = $img->src;
}
// authors
$authors = $article->find('*[itemtype*="schema.org/Person"]');
if (!$authors) {
$authors = $article->find('.metadata__source');
}
if ($authors) {
$item['author'] = implode(', ', $authors);
}
// header image
$headerimg = $article->find('*[data-ct-row="headerimage"]', 0) ?? $article->find('header', 0);
if ($headerimg) {
$item['content'] .= implode('', $headerimg->find('img[src], figcaption'));
}
// article content
$pages = $article->find('.article-page');
if ($pages) {
foreach ($pages as $page) {
$elements = $page->find('p, h2, figcaption, img[src]');
$item['content'] .= implode('', $elements);
}
}
return $item;
}
}