# coding: utf-8 from __future__ import unicode_literals from .common import InfoExtractor import re from ..utils import ( clean_html, determine_ext, get_element_by_class, urljoin, compat_parse_qs, ExtractorError) class WikimediaIE(InfoExtractor): IE_NAME = 'wikimedia.org' _NETRC_MACHINE = 'wikimediaorg' _API_BASE_URL = 'https://commons.wikimedia.org/' _VALID_URL = 'https://commons.wikimedia.org/wiki/File:(?P[^/]+)' _TEST = { 'url': 'https://commons.wikimedia.org/wiki/File:Die_Temperaturkurve_der_Erde_(ZDF,_Terra_X)_720p_HD_50FPS.webm', 'info_dict': { 'description': 'md5:7cd84f76e7081f1be033d0b155b4a460', 'ext': 'webm', 'id': 'Die_Temperaturkurve_der_Erde_(ZDF,_Terra_X)_720p_HD_50FPS', 'title': 'Die Temperaturkurve der Erde (ZDF, Terra X) 720p HD 50FPS.webm - Wikimedia Commons', 'license': 'md5:62907cddf705a9f7ae7076c15407a977', 'author': None, 'subtitles': {'de': [{'ext': 'srt', 'url': 'https://commons.wikimedia.org/w/api.php?action=timedtext&title=File%3ADie_Temperaturkurve_der_Erde_%28ZDF%2C_Terra_X%29_720p_HD_50FPS.webm&lang=de&trackformat=vtt '}], 'en-gb': [{'ext': 'srt', 'url': 'https://commons.wikimedia.org/w/api.php?action=timedtext&title=File%3ADie_Temperaturkurve_der_Erde_%28ZDF%2C_Terra_X%29_720p_HD_50FPS.webm&lang=en-gb&trackformat=vtt '}], 'nl': [{'ext': 'srt', 'url': 'https://commons.wikimedia.org/w/api.php?action=timedtext&title=File%3ADie_Temperaturkurve_der_Erde_%28ZDF%2C_Terra_X%29_720p_HD_50FPS.webm&lang=nl&trackformat=vtt '}], 'en': [{'ext': 'srt', 'url': 'https://commons.wikimedia.org/w/api.php?action=timedtext&title=File%3ADie_Temperaturkurve_der_Erde_%28ZDF%2C_Terra_X%29_720p_HD_50FPS_-_redub_NL.webm&lang=en&trackformat=vtt '}]}} } def _real_extract(self, url): video_id = self._match_id(url) ext = determine_ext(url, None) if ext is None: raise ExtractorError('invalid video url', expected=True) webpage = self._download_webpage(url, video_id) self.report_extraction(video_id) video_url = self._html_search_regex(']*src="([^"]+)"', webpage, 'video URL') license = get_element_by_class('layouttemplate licensetpl mw-content-ltr', webpage) license = clean_html(license) description = get_element_by_class('description', webpage) author = self._html_search_regex(']*>[\n\s]([^<]+?)\s*', webpage, 'video author', default=None) info = {'url': video_url, 'description': clean_html(description), 'ext': ext, 'id': video_id.replace('.' + ext, ''), 'title': self._og_search_title(webpage).replace('File:', ''), 'license': license, 'author': author} subtitles = {} for sub in re.findall(r'''\bsrc\s*=\s*[\"\'](\/w\/api(.*?)[\s\"])\b''', webpage): sub = sub[0].replace('"', '''''') sub = urljoin('https://commons.wikimedia.org', sub) qs = compat_parse_qs(sub) lang = qs.get('lang', [None])[-1] if not lang: continue subtitles[lang] = [{'ext': 'srt', 'url': sub}] info['subtitles'] = subtitles return info