aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl
diff options
context:
space:
mode:
authorRemita Amine <remitamine@gmail.com>2021-02-01 22:35:28 +0100
committerRemita Amine <remitamine@gmail.com>2021-02-01 22:35:28 +0100
commitc11f7cf9bd6ef239f25e7fb9c54e092ae1490e2d (patch)
treeb66de43ecfe7e0daa1830401c7e1fd7533288f96 /youtube_dl
parent8fa7cc387d699899114f7430bcf61837d58557a8 (diff)
[vidzi] remove extractor(closes #12629)
Diffstat (limited to 'youtube_dl')
-rw-r--r--youtube_dl/extractor/extractors.py1
-rw-r--r--youtube_dl/extractor/vidzi.py68
2 files changed, 0 insertions, 69 deletions
diff --git a/youtube_dl/extractor/extractors.py b/youtube_dl/extractor/extractors.py
index ab8d6a5a5..97b0b4034 100644
--- a/youtube_dl/extractor/extractors.py
+++ b/youtube_dl/extractor/extractors.py
@@ -1399,7 +1399,6 @@ from .vidme import (
VidmeUserIE,
VidmeUserLikesIE,
)
-from .vidzi import VidziIE
from .vier import VierIE, VierVideosIE
from .viewlift import (
ViewLiftIE,
diff --git a/youtube_dl/extractor/vidzi.py b/youtube_dl/extractor/vidzi.py
deleted file mode 100644
index 42ea4952c..000000000
--- a/youtube_dl/extractor/vidzi.py
+++ /dev/null
@@ -1,68 +0,0 @@
-# coding: utf-8
-from __future__ import unicode_literals
-
-import re
-
-from .common import InfoExtractor
-from ..utils import (
- decode_packed_codes,
- js_to_json,
- NO_DEFAULT,
- PACKED_CODES_RE,
-)
-
-
-class VidziIE(InfoExtractor):
- _VALID_URL = r'https?://(?:www\.)?vidzi\.(?:tv|cc|si|nu)/(?:embed-)?(?P<id>[0-9a-zA-Z]+)'
- _TESTS = [{
- 'url': 'http://vidzi.tv/cghql9yq6emu.html',
- 'md5': '4f16c71ca0c8c8635ab6932b5f3f1660',
- 'info_dict': {
- 'id': 'cghql9yq6emu',
- 'ext': 'mp4',
- 'title': 'youtube-dl test video 1\\\\2\'3/4<5\\\\6ä7↭',
- },
- 'params': {
- # m3u8 download
- 'skip_download': True,
- },
- }, {
- 'url': 'http://vidzi.tv/embed-4z2yb0rzphe9-600x338.html',
- 'only_matching': True,
- }, {
- 'url': 'http://vidzi.cc/cghql9yq6emu.html',
- 'only_matching': True,
- }, {
- 'url': 'https://vidzi.si/rph9gztxj1et.html',
- 'only_matching': True,
- }, {
- 'url': 'http://vidzi.nu/cghql9yq6emu.html',
- 'only_matching': True,
- }]
-
- def _real_extract(self, url):
- video_id = self._match_id(url)
-
- webpage = self._download_webpage(
- 'http://vidzi.tv/%s' % video_id, video_id)
- title = self._html_search_regex(
- r'(?s)<h2 class="video-title">(.*?)</h2>', webpage, 'title')
-
- codes = [webpage]
- codes.extend([
- decode_packed_codes(mobj.group(0)).replace('\\\'', '\'')
- for mobj in re.finditer(PACKED_CODES_RE, webpage)])
- for num, code in enumerate(codes, 1):
- jwplayer_data = self._parse_json(
- self._search_regex(
- r'setup\(([^)]+)\)', code, 'jwplayer data',
- default=NO_DEFAULT if num == len(codes) else '{}'),
- video_id, transform_source=lambda s: js_to_json(
- re.sub(r'\s*\+\s*window\[.+?\]', '', s)))
- if jwplayer_data:
- break
-
- info_dict = self._parse_jwplayer_data(jwplayer_data, video_id, require_title=False)
- info_dict['title'] = title
-
- return info_dict