|
@ -0,0 +1,130 @@ |
|
|
|
|
|
# encoding: utf-8 |
|
|
|
|
|
from __future__ import unicode_literals |
|
|
|
|
|
|
|
|
|
|
|
import re |
|
|
|
|
|
|
|
|
|
|
|
from .common import InfoExtractor |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class DramaFeverIE(InfoExtractor): |
|
|
|
|
|
IE_NAME = 'dramafever' |
|
|
|
|
|
_VALID_URL = r'^https?://(?:www\.)?dramafever\.com/drama/(?P<id>[0-9]+/[0-9]+)/' |
|
|
|
|
|
_TESTS = [{ |
|
|
|
|
|
'url': 'http://www.dramafever.com/drama/4512/1/Cooking_with_Shin/', |
|
|
|
|
|
'info_dict': { |
|
|
|
|
|
'id': '4512.1', |
|
|
|
|
|
'ext': 'flv', |
|
|
|
|
|
'title': 'Cooking with Shin 4512.1', |
|
|
|
|
|
'upload_date': '20140702', |
|
|
|
|
|
'description': 'Served at all special occasions and featured in the hit drama Heirs, Shin cooks Red Bean Rice.', |
|
|
|
|
|
} |
|
|
|
|
|
}] |
|
|
|
|
|
|
|
|
|
|
|
def _real_extract(self, url): |
|
|
|
|
|
video_id = self._match_id(url).replace("/", ".") |
|
|
|
|
|
|
|
|
|
|
|
consumer_secret = self._get_consumer_secret(video_id) |
|
|
|
|
|
|
|
|
|
|
|
ep_json = self._download_json( |
|
|
|
|
|
"http://www.dramafever.com/amp/episode/feed.json?guid=%s" % video_id, |
|
|
|
|
|
video_id, note='Downloading episode metadata', |
|
|
|
|
|
errnote="Video may not be available for your location")["channel"]["item"] |
|
|
|
|
|
|
|
|
|
|
|
title = ep_json["media-group"]["media-title"] |
|
|
|
|
|
description = ep_json["media-group"]["media-description"] |
|
|
|
|
|
thumbnail = ep_json["media-group"]["media-thumbnail"]["@attributes"]["url"] |
|
|
|
|
|
duration = int(ep_json["media-group"]["media-content"][0]["@attributes"]["duration"]) |
|
|
|
|
|
mobj = re.match(r"([0-9]{4})-([0-9]{2})-([0-9]{2})", ep_json["pubDate"]) |
|
|
|
|
|
upload_date = mobj.group(1) + mobj.group(2) + mobj.group(3) if mobj is not None else None |
|
|
|
|
|
|
|
|
|
|
|
formats = [] |
|
|
|
|
|
for vid_format in ep_json["media-group"]["media-content"]: |
|
|
|
|
|
src = vid_format["@attributes"]["url"] |
|
|
|
|
|
if '.f4m' in src: |
|
|
|
|
|
formats.extend(self._extract_f4m_formats(src, video_id)) |
|
|
|
|
|
|
|
|
|
|
|
self._sort_formats(formats) |
|
|
|
|
|
video_subtitles = self.extract_subtitles(video_id, consumer_secret) |
|
|
|
|
|
|
|
|
|
|
|
return { |
|
|
|
|
|
'id': video_id, |
|
|
|
|
|
'title': title, |
|
|
|
|
|
'description': description, |
|
|
|
|
|
'thumbnail': thumbnail, |
|
|
|
|
|
'upload_date': upload_date, |
|
|
|
|
|
'duration': duration, |
|
|
|
|
|
'formats': formats, |
|
|
|
|
|
'subtitles': video_subtitles, |
|
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
|
|
def _get_consumer_secret(self, video_id): |
|
|
|
|
|
df_js = self._download_webpage( |
|
|
|
|
|
"http://www.dramafever.com/static/126960d/v2/js/plugins/jquery.threadedcomments.js", video_id) |
|
|
|
|
|
return self._search_regex(r"'cs': '([0-9a-zA-Z]+)'", df_js, "cs") |
|
|
|
|
|
|
|
|
|
|
|
def _get_episodes(self, series_id, consumer_secret, episode_filter=None): |
|
|
|
|
|
_PAGE_SIZE = 60 |
|
|
|
|
|
|
|
|
|
|
|
curr_page = 1 |
|
|
|
|
|
max_pages = curr_page + 1 |
|
|
|
|
|
results = [] |
|
|
|
|
|
while max_pages >= curr_page: |
|
|
|
|
|
page_url = "http://www.dramafever.com/api/4/episode/series/?cs=%s&series_id=%s&page_size=%d&page_number=%d" % \ |
|
|
|
|
|
(consumer_secret, series_id, _PAGE_SIZE, curr_page) |
|
|
|
|
|
series = self._download_json( |
|
|
|
|
|
page_url, series_id, note="Downloading series json page #%d" % curr_page) |
|
|
|
|
|
max_pages = series['num_pages'] |
|
|
|
|
|
results.extend([ep for ep in series['value'] if episode_filter is None or episode_filter(ep)]) |
|
|
|
|
|
curr_page += 1 |
|
|
|
|
|
return results |
|
|
|
|
|
|
|
|
|
|
|
def _get_subtitles(self, video_id, consumer_secret): |
|
|
|
|
|
|
|
|
|
|
|
res = None |
|
|
|
|
|
info = self._get_episodes( |
|
|
|
|
|
video_id.split(".")[0], consumer_secret, |
|
|
|
|
|
episode_filter=lambda x: x['guid'] == video_id) |
|
|
|
|
|
|
|
|
|
|
|
if len(info) == 1 and info[0]['subfile'] != '': |
|
|
|
|
|
res = {'en': [{'url': info[0]['subfile'], 'ext': 'srt'}]} |
|
|
|
|
|
return res |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class DramaFeverSeriesIE(DramaFeverIE): |
|
|
|
|
|
IE_NAME = 'dramafever:series' |
|
|
|
|
|
_VALID_URL = r'^https?://(?:www\.)?dramafever\.com/drama/(?P<id>[0-9]+)/\d*[a-zA-Z_][a-zA-Z0-9_]*/' |
|
|
|
|
|
_TESTS = [{ |
|
|
|
|
|
'url': 'http://www.dramafever.com/drama/4512/Cooking_with_Shin/', |
|
|
|
|
|
'info_dict': { |
|
|
|
|
|
'id': '4512', |
|
|
|
|
|
'title': 'Cooking with Shin', |
|
|
|
|
|
'description': 'Professional chef and cooking instructor Shin Kim takes some of the delicious dishes featured in your favorite dramas and shows you how to make them right at home.', |
|
|
|
|
|
}, |
|
|
|
|
|
'playlist_count': 4, |
|
|
|
|
|
}, { |
|
|
|
|
|
'url': 'http://www.dramafever.com/drama/124/IRIS/', |
|
|
|
|
|
'info_dict': { |
|
|
|
|
|
'id': '124', |
|
|
|
|
|
'title': 'IRIS', |
|
|
|
|
|
'description': 'Lee Byung Hun and Kim Tae Hee star in this powerhouse drama and ratings megahit of action, intrigue and romance.', |
|
|
|
|
|
}, |
|
|
|
|
|
'playlist_count': 20, |
|
|
|
|
|
}] |
|
|
|
|
|
|
|
|
|
|
|
def _real_extract(self, url): |
|
|
|
|
|
series_id = self._match_id(url) |
|
|
|
|
|
consumer_secret = self._get_consumer_secret(series_id) |
|
|
|
|
|
|
|
|
|
|
|
series_json = self._download_json( |
|
|
|
|
|
"http://www.dramafever.com/api/4/series/query/?cs=%s&series_id=%s" % (consumer_secret, series_id), |
|
|
|
|
|
series_id, note='Downloading series metadata')["series"][series_id] |
|
|
|
|
|
|
|
|
|
|
|
title = series_json["name"] |
|
|
|
|
|
description = series_json["description_short"] |
|
|
|
|
|
|
|
|
|
|
|
episodes = self._get_episodes(series_id, consumer_secret) |
|
|
|
|
|
entries = [] |
|
|
|
|
|
for ep in episodes: |
|
|
|
|
|
entries.append(self.url_result( |
|
|
|
|
|
'http://www.dramafever.com%s' % ep['episode_url'], 'DramaFever', ep['guid'])) |
|
|
|
|
|
return self.playlist_result(entries, series_id, title, description) |