diff options
author | remitamine <remitamine@gmail.com> | 2016-03-18 21:17:45 +0100 |
---|---|---|
committer | remitamine <remitamine@gmail.com> | 2016-03-18 21:18:18 +0100 |
commit | 4cd70099ea79a4a82b26694937ca46d31f7436ca (patch) | |
tree | 1300022f93117d5f383698ff4dc99956dea92616 /youtube_dl/extractor/hbo.py | |
parent | 09fc33198a4cfc93a98ce1ba7d51d41c487e5f56 (diff) |
[hbo] Add new extractor
Diffstat (limited to 'youtube_dl/extractor/hbo.py')
-rw-r--r-- | youtube_dl/extractor/hbo.py | 122 |
1 files changed, 122 insertions, 0 deletions
diff --git a/youtube_dl/extractor/hbo.py b/youtube_dl/extractor/hbo.py new file mode 100644 index 000000000..dad0f3994 --- /dev/null +++ b/youtube_dl/extractor/hbo.py @@ -0,0 +1,122 @@ +# coding: utf-8 +from __future__ import unicode_literals + +import re + +from .common import InfoExtractor +from ..utils import ( + xpath_text, + xpath_element, + int_or_none, + parse_duration, +) + + +class HBOIE(InfoExtractor): + _VALID_URL = r'https?://(?:www\.)?hbo\.com/video/video\.html\?.*vid=(?P<id>[0-9]+)' + _TEST = { + 'url': 'http://www.hbo.com/video/video.html?autoplay=true&g=u&vid=1437839', + 'md5': '1c33253f0c7782142c993c0ba62a8753', + 'info_dict': { + 'id': '1437839', + 'ext': 'mp4', + 'title': 'Ep. 64 Clip: Encryption', + } + } + _FORMATS_INFO = { + '1920': { + 'width': 1280, + 'height': 720, + }, + '640': { + 'width': 768, + 'height': 432, + }, + 'highwifi': { + 'width': 640, + 'height': 360, + }, + 'high3g': { + 'width': 640, + 'height': 360, + }, + 'medwifi': { + 'width': 400, + 'height': 224, + }, + 'med3g': { + 'width': 400, + 'height': 224, + }, + } + + def _real_extract(self, url): + video_id = self._match_id(url) + video_data = self._download_xml( + 'http://render.lv3.hbo.com/data/content/global/videos/data/%s.xml' % video_id, video_id) + title = xpath_text(video_data, 'title', 'title', True) + + formats = [] + for source in xpath_element(video_data, 'videos', 'sources', True): + if source.tag == 'size': + path = xpath_text(source, './/path') + if not path: + continue + width = source.attrib.get('width') + format_info = self._FORMATS_INFO.get(width, {}) + height = format_info.get('height') + fmt = { + 'url': path, + 'format_id': 'http%s' % ('-%dp' % height if height else ''), + 'width': format_info.get('width'), + 'height': height, + } + rtmp = re.search(r'^(?P<url>rtmpe?://[^/]+/(?P<app>.+))/(?P<playpath>mp4:.+)$', path) + if rtmp: + fmt.update({ + 'url': rtmp.group('url'), + 'play_path': rtmp.group('playpath'), + 'app': rtmp.group('app'), + 'ext': 'flv', + 'format_id': fmt['format_id'].replace('http', 'rtmp'), + }) + formats.append(fmt) + else: + video_url = source.text + if not video_url: + continue + if source.tag == 'tarball': + formats.extend(self._extract_m3u8_formats( + video_url.replace('.tar', '/base_index_w8.m3u8'), + video_id, 'mp4', 'm3u8_native', m3u8_id='hls', fatal=False)) + else: + format_info = self._FORMATS_INFO.get(source.tag, {}) + formats.append({ + 'format_id': 'http-%s' % source.tag, + 'url': video_url, + 'width': format_info.get('width'), + 'height': format_info.get('height'), + }) + self._sort_formats(formats, ('width', 'height', 'tbr', 'format_id')) + + thumbnails = [] + card_sizes = xpath_element(video_data, 'titleCardSizes') + if card_sizes is not None: + for size in card_sizes: + path = xpath_text(size, 'path') + if not path: + continue + width = int_or_none(size.get('width')) + thumbnails.append({ + 'id': width, + 'url': path, + 'width': width, + }) + + return { + 'id': video_id, + 'title': title, + 'duration': parse_duration(xpath_element(video_data, 'duration/tv14')), + 'formats': formats, + 'thumbnails': thumbnails, + } |