diff options
author | Yen Chi Hsuan <yan12125@gmail.com> | 2015-02-25 00:45:11 +0800 |
---|---|---|
committer | Yen Chi Hsuan <yan12125@gmail.com> | 2015-02-25 01:26:44 +0800 |
commit | 570311610e33eb67d5a65a86705d54daaee17361 (patch) | |
tree | 6113bc3b2e821455e5522fac2cce530bc23b484b | |
parent | 7f09a662a0db2ad0a338645e4f2dbd43056a6fa6 (diff) |
[Letv] Add playlist support
-rw-r--r-- | youtube_dl/extractor/__init__.py | 6 | ||||
-rw-r--r-- | youtube_dl/extractor/letv.py | 54 |
2 files changed, 58 insertions, 2 deletions
diff --git a/youtube_dl/extractor/__init__.py b/youtube_dl/extractor/__init__.py index 829ab18a9..7b7d41adf 100644 --- a/youtube_dl/extractor/__init__.py +++ b/youtube_dl/extractor/__init__.py @@ -237,7 +237,11 @@ from .krasview import KrasViewIE from .ku6 import Ku6IE from .la7 import LA7IE from .laola1tv import Laola1TvIE -from .letv import LetvIE +from .letv import ( + LetvIE, + LetvTvIE, + LetvPlaylistIE +) from .lifenews import LifeNewsIE from .liveleak import LiveLeakIE from .livestream import ( diff --git a/youtube_dl/extractor/letv.py b/youtube_dl/extractor/letv.py index 3eb20678e..d7d315634 100644 --- a/youtube_dl/extractor/letv.py +++ b/youtube_dl/extractor/letv.py @@ -2,6 +2,7 @@ from __future__ import unicode_literals import os.path +import re import time import datetime @@ -11,7 +12,7 @@ from ..utils import (ExtractorError, parse_iso8601) class LetvIE(InfoExtractor): - _VALID_URL = r'http://www.letv.com/ptv/vplay/(?P<id>\d+).html' + _VALID_URL = r'http://www\.letv\.com/ptv/vplay/(?P<id>\d+).html' _TESTS = [{ 'url': 'http://www.letv.com/ptv/vplay/22005890.html', @@ -118,3 +119,54 @@ class LetvIE(InfoExtractor): 'thumbnail': playurl['pic'], 'timestamp': publish_time, } + + +class LetvTvIE(InfoExtractor): + _VALID_URL = r'http://www.letv.com/tv/(?P<id>\d+).html' + _TESTS = [{ + 'url': 'http://www.letv.com/tv/46177.html', + 'info_dict': { + 'id': '46177', + 'title': '美人天下', + 'description': 'md5:395666ff41b44080396e59570dbac01c' + }, + 'playlist_count': 35 + }] + + def _real_extract(self, url): + playlist_id = self._match_id(url) + page = self._download_webpage(url, playlist_id) + + media_urls = list(set(re.findall( + r'http://www.letv.com/ptv/vplay/\d+.html', page))) + entries = [self.url_result(media_url, ie='Letv') + for media_url in media_urls] + + title = self._html_search_meta('keywords', page, fatal=False).split(',')[0] + description = self._html_search_meta('description', page, fatal=False) + + return self.playlist_result(entries, playlist_id, playlist_title=title, + playlist_description=description) + + +class LetvPlaylistIE(LetvTvIE): + _VALID_URL = r'http://tv.letv.com/[a-z]+/(?P<id>[a-z]+)/index.s?html' + _TESTS = [{ + 'url': 'http://tv.letv.com/izt/wuzetian/index.html', + 'info_dict': { + 'id': 'wuzetian', + 'title': '武媚娘传奇', + 'description': 'md5:e12499475ab3d50219e5bba00b3cb248' + }, + 'playlist_count': 96 + }, { + 'url': 'http://tv.letv.com/pzt/lswjzzjc/index.shtml', + 'info_dict': { + 'id': 'lswjzzjc', + # should be "劲舞青春", but I can't find a simple way to determine + # the playlist title + 'title': '乐视午间自制剧场', + 'description': 'md5:b1eef244f45589a7b5b1af9ff25a4489' + }, + 'playlist_mincount': 7 + }] |