import logging logging.basicConfig( format='%(asctime)s - %(name)s - %(levelname)s - %(message)s', level=logging.INFO) import requests import time from feeds import hackernews, reddit, tildes OUTLINE_API = 'https://outlineapi.com/article' READ_API = 'http://127.0.0.1:33843' def list(): feed = [] feed += [(x, 'hackernews') for x in hackernews.feed()] feed += [(x, 'reddit') for x in reddit.feed()] feed += [(x, 'tildes') for x in tildes.feed()] return feed def get_article(url): try: params = {'source_url': url} headers = {'Referer': 'https://outline.com/'} r = requests.get(OUTLINE_API, params=params, headers=headers, timeout=20) if r.status_code == 429: logging.error('Rate limited by outline, sleeping 30s and skipping...') time.sleep(30) return '' if r.status_code != 200: raise Exception('Bad response code ' + str(r.status_code)) html = r.json()['data']['html'] if 'URL is not supported by Outline' in html: raise Exception('URL not supported by Outline') return html except BaseException as e: logging.error('Problem outlining article: {}'.format(str(e))) logging.info('Trying our server instead...') try: r = requests.post(READ_API, data=dict(url=url), timeout=10) if r.status_code != 200: raise Exception('Bad response code ' + str(r.status_code)) return r.text except BaseException as e: logging.error('Problem getting article: {}'.format(str(e))) return '' def update_story(story): res = {} logging.info('Updating story ' + str(story['ref'])) if story['source'] == 'hackernews': res = hackernews.story(story['ref']) elif story['source'] == 'reddit': res = reddit.story(story['ref']) elif story['source'] == 'tildes': res = tildes.story(story['ref']) else: return if res: story.update(res) if story.get('url', '') and not story.get('text', ''): if not story['url'].endswith('.pdf'): logging.info('Getting article ' + story['url']) story['text'] = get_article(story['url']) else: story['text'] = '
Unsupported article type.
'