aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorNaglis Jonaitis <njonaitis@gmail.com>2015-02-08 16:57:07 +0200
committerNaglis Jonaitis <njonaitis@gmail.com>2015-02-08 16:57:07 +0200
commita14292e8480f339bdbf9e52e93460aec5c31ef13 (patch)
tree5960652b212b3fdc17f45a3ec4567160a1c096b7
parentc0e46412e99ef87f5730326ab8fa4a56038c43d1 (diff)
[soulanime] Remove extractor (#4554)
Was supposed to be deleted by 67c2bcd
-rw-r--r--youtube_dl/extractor/soulanime.py80
1 files changed, 0 insertions, 80 deletions
diff --git a/youtube_dl/extractor/soulanime.py b/youtube_dl/extractor/soulanime.py
deleted file mode 100644
index feef33e27..000000000
--- a/youtube_dl/extractor/soulanime.py
+++ /dev/null
@@ -1,80 +0,0 @@
-from __future__ import unicode_literals
-
-import re
-
-from .common import InfoExtractor
-from ..utils import (
- HEADRequest,
- urlhandle_detect_ext,
-)
-
-
-class SoulAnimeWatchingIE(InfoExtractor):
- IE_NAME = "soulanime:watching"
- IE_DESC = "SoulAnime video"
- _TEST = {
- 'url': 'http://www.soul-anime.net/watching/seirei-tsukai-no-blade-dance-episode-9/',
- 'md5': '05fae04abf72298098b528e98abf4298',
- 'info_dict': {
- 'id': 'seirei-tsukai-no-blade-dance-episode-9',
- 'ext': 'mp4',
- 'title': 'seirei-tsukai-no-blade-dance-episode-9',
- 'description': 'seirei-tsukai-no-blade-dance-episode-9'
- }
- }
- _VALID_URL = r'http://[w.]*soul-anime\.(?P<domain>[^/]+)/watch[^/]*/(?P<id>[^/]+)'
-
- def _real_extract(self, url):
- mobj = re.match(self._VALID_URL, url)
- video_id = mobj.group('id')
- domain = mobj.group('domain')
-
- page = self._download_webpage(url, video_id)
-
- video_url_encoded = self._html_search_regex(
- r'<div id="download">[^<]*<a href="(?P<url>[^"]+)"', page, 'url')
- video_url = "http://www.soul-anime." + domain + video_url_encoded
-
- ext_req = HEADRequest(video_url)
- ext_handle = self._request_webpage(
- ext_req, video_id, note='Determining extension')
- ext = urlhandle_detect_ext(ext_handle)
-
- return {
- 'id': video_id,
- 'url': video_url,
- 'ext': ext,
- 'title': video_id,
- 'description': video_id
- }
-
-
-class SoulAnimeSeriesIE(InfoExtractor):
- IE_NAME = "soulanime:series"
- IE_DESC = "SoulAnime Series"
-
- _VALID_URL = r'http://[w.]*soul-anime\.(?P<domain>[^/]+)/anime./(?P<id>[^/]+)'
-
- _EPISODE_REGEX = r'<option value="(/watch[^/]*/[^"]+)">[^<]*</option>'
-
- _TEST = {
- 'url': 'http://www.soul-anime.net/anime1/black-rock-shooter-tv/',
- 'info_dict': {
- 'id': 'black-rock-shooter-tv'
- },
- 'playlist_count': 8
- }
-
- def _real_extract(self, url):
- mobj = re.match(self._VALID_URL, url)
- series_id = mobj.group('id')
- domain = mobj.group('domain')
-
- pattern = re.compile(self._EPISODE_REGEX)
-
- page = self._download_webpage(url, series_id, "Downloading series page")
- mobj = pattern.findall(page)
-
- entries = [self.url_result("http://www.soul-anime." + domain + obj) for obj in mobj]
-
- return self.playlist_result(entries, series_id)