mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-27 01:31:25 +01:00
Compare commits
3 Commits
891c293b10
...
ecfed8d759
Author | SHA1 | Date | |
---|---|---|---|
|
ecfed8d759 | ||
|
2358117d39 | ||
|
9ef3e8aaa3 |
|
@ -446,6 +446,11 @@ from .cspan import (
|
||||||
CSpanCongressIE,
|
CSpanCongressIE,
|
||||||
CSpanIE,
|
CSpanIE,
|
||||||
)
|
)
|
||||||
|
from .ctc import (
|
||||||
|
CTCIE,
|
||||||
|
CTCSeasonIE,
|
||||||
|
CTCSeriesIE,
|
||||||
|
)
|
||||||
from .ctsnews import CtsNewsIE
|
from .ctsnews import CtsNewsIE
|
||||||
from .ctv import CTVIE
|
from .ctv import CTVIE
|
||||||
from .ctvnews import CTVNewsIE
|
from .ctvnews import CTVNewsIE
|
||||||
|
|
164
yt_dlp/extractor/ctc.py
Normal file
164
yt_dlp/extractor/ctc.py
Normal file
|
@ -0,0 +1,164 @@
|
||||||
|
from .common import InfoExtractor
|
||||||
|
from ..utils import (
|
||||||
|
traverse_obj,
|
||||||
|
int_or_none
|
||||||
|
)
|
||||||
|
|
||||||
|
class CTCIE(InfoExtractor):
|
||||||
|
IE_NAME = 'ctc'
|
||||||
|
_VALID_URL = (
|
||||||
|
r'https?://ctc\.ru/projects/filmi/(?P<project_name>[^/]+)/?(?:video/?)?$'
|
||||||
|
r'|https?://ctc\.ru/projects/(?P<category>show|multiki|serials)/(?P<project_name2>[^/]+)/video/'
|
||||||
|
r'(?:$'
|
||||||
|
r'|(?P<season_number>\d+)-sezon/(?P<episode_number>\d+)-(?:vypusk|serija)/?$'
|
||||||
|
r'|promo/[^/]+/?$'
|
||||||
|
r')'
|
||||||
|
)
|
||||||
|
_GEO_COUNTRIES = ['RU']
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
url_slug = url.split("https://ctc.ru/")[1]
|
||||||
|
|
||||||
|
item_response = self._download_json(
|
||||||
|
f'https://ctc.ru/api/page/v1/{url_slug}', url_slug,
|
||||||
|
note='Downloading item data', headers={
|
||||||
|
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:97.0) Gecko/20100101 Firefox/97.0"
|
||||||
|
}
|
||||||
|
)
|
||||||
|
|
||||||
|
track_hub_id = str(traverse_obj(item_response, ('content', 0, 'trackHubId'), expected_type=int))
|
||||||
|
if not track_hub_id:
|
||||||
|
self.raise_no_formats('trackHubId not found')
|
||||||
|
|
||||||
|
video_url = traverse_obj(
|
||||||
|
item_response, ('content', 0, 'videoUrl'), get_all=False
|
||||||
|
) or traverse_obj(
|
||||||
|
item_response, ('content', 0, 'trackUrl'), get_all=False
|
||||||
|
)
|
||||||
|
stream_response = self._download_json(
|
||||||
|
video_url.replace("/player/", "/playlist/"),
|
||||||
|
track_hub_id,
|
||||||
|
note='Downloading stream data', headers={
|
||||||
|
'X-Referer': 'https://ctc.ru',
|
||||||
|
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:97.0) Gecko/20100101 Firefox/97.0',
|
||||||
|
}
|
||||||
|
)
|
||||||
|
|
||||||
|
if traverse_obj(stream_response, ('playlist', 'items', 0, 'errors', 0, 'code')) == 102:
|
||||||
|
self.raise_geo_restricted(countries=self._GEO_COUNTRIES)
|
||||||
|
elif traverse_obj(stream_response, ('playlist', 'items', 0, 'errors', 0, 'code')) == 103:
|
||||||
|
self.raise_login_required(msg='This video is only available for registered users with the required subscription')
|
||||||
|
|
||||||
|
formats = []
|
||||||
|
for stream in traverse_obj(stream_response, ('playlist', 'items', 0, 'streams')):
|
||||||
|
protocol = stream.get('protocol')
|
||||||
|
video_id = traverse_obj(stream_response, ('playlist', 'items', 0, 'track_id'))
|
||||||
|
|
||||||
|
fmts = []
|
||||||
|
if protocol == 'HLS':
|
||||||
|
fmts, _ = self._extract_m3u8_formats_and_subtitles(
|
||||||
|
stream.get('url'), video_id, ext='mp4', preference=1, m3u8_id="video/hls", fatal=False)
|
||||||
|
formats.extend(fmts)
|
||||||
|
|
||||||
|
return {
|
||||||
|
'formats': formats,
|
||||||
|
'id': track_hub_id,
|
||||||
|
'title': ": ".join(filter(None, [
|
||||||
|
traverse_obj(stream_response, ('playlist', 'items', 0, 'project_name')),
|
||||||
|
traverse_obj(stream_response, ('playlist', 'items', 0, 'episode_name'))
|
||||||
|
])),
|
||||||
|
**traverse_obj(item_response, {
|
||||||
|
'description': ('content', 0, 'description'),
|
||||||
|
}),
|
||||||
|
**traverse_obj(stream_response, {
|
||||||
|
'episode': ('playlist', 'items', 0, 'episode_name'),
|
||||||
|
'duration': ('playlist', 'items', 0, 'duration'),
|
||||||
|
'thumbnail': ('playlist', 'items', 0, 'thumbnail_url'),
|
||||||
|
'season': ('playlist', 'items', 0, 'season_name'),
|
||||||
|
'age_limit': ('playlist', 'items', 0, 'min_age'),
|
||||||
|
}),
|
||||||
|
'season_number': int_or_none(self._match_valid_url(url).group('season_number')),
|
||||||
|
'episode_number': int_or_none(self._match_valid_url(url).group('episode_number')),
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
class CTCSeasonIE(InfoExtractor):
|
||||||
|
IE_NAME = 'ctc:season'
|
||||||
|
_VALID_URL = (
|
||||||
|
r'https?://ctc\.ru/projects/(?P<category>show|multiki|serials)/'
|
||||||
|
r'(?P<project_name>[^/]+)/video/(?P<season_number>\d+)-sezon/?$'
|
||||||
|
)
|
||||||
|
_GEO_COUNTRIES = ['RU']
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
url_slug = url.split("https://ctc.ru/")[1]
|
||||||
|
|
||||||
|
season_data = self._download_json(f'https://ctc.ru/api/page/v1/{url_slug}', url_slug)
|
||||||
|
|
||||||
|
entries = [{
|
||||||
|
'_type': 'url',
|
||||||
|
'title': episode.get('title'),
|
||||||
|
'url': f"https://ctc.ru{episode.get('popupUrl')}",
|
||||||
|
'ie_key': CTCIE.ie_key(),
|
||||||
|
'season_number': self._match_valid_url(url).group('season_number'),
|
||||||
|
'episode_number': self._search_regex(
|
||||||
|
r'/(?P<episode>\d+)-(vypusk|serija)/',
|
||||||
|
episode.get('popupUrl'),
|
||||||
|
'episode number',
|
||||||
|
)
|
||||||
|
} for episode in traverse_obj(season_data, ('content', 1, 'widgets')) if episode.get('popupUrl')]
|
||||||
|
|
||||||
|
return {
|
||||||
|
'_type': 'playlist',
|
||||||
|
'entries': entries,
|
||||||
|
**traverse_obj(season_data, {
|
||||||
|
'id': ('content', 9, 'entityId'),
|
||||||
|
'title': ('content', 0, 'widgets', 1, 'title'),
|
||||||
|
'season_number': self._match_valid_url(url).group('season_number'),
|
||||||
|
'series': ('content', 0, 'widgets', 1, 'title'),
|
||||||
|
'description': ('content', 0, 'widgets', 1, 'description'),
|
||||||
|
}),
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
class CTCSeriesIE(InfoExtractor):
|
||||||
|
IE_NAME = 'ctc:series'
|
||||||
|
_VALID_URL = (
|
||||||
|
r'https?://ctc\.ru/projects/(?P<category>show|multiki|serials)/'
|
||||||
|
r'(?P<slug>[^/]+)/?$'
|
||||||
|
)
|
||||||
|
_GEO_COUNTRIES = ['RU']
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
url_slug = url.split("https://ctc.ru/")[1]
|
||||||
|
|
||||||
|
series_data = self._download_json(f'https://ctc.ru/api/page/v1/{url_slug}', url_slug)
|
||||||
|
|
||||||
|
# cartoons doesn't indicate in the url what type it is, so
|
||||||
|
# if it's a movie, then redirect it to CTCIE
|
||||||
|
if traverse_obj(series_data, ('content', 5, 'tabs')) == []:
|
||||||
|
return self.url_result(f'https://ctc.ru/{url_slug}/video')
|
||||||
|
|
||||||
|
entries = [{
|
||||||
|
'_type': 'url',
|
||||||
|
'title': season.get('title'),
|
||||||
|
'url': f'https://ctc.ru{season.get("url")}',
|
||||||
|
'ie_key': CTCSeasonIE.ie_key(),
|
||||||
|
'season_number': self._search_regex(
|
||||||
|
r'/(?P<episode>\d+)-sezon/',
|
||||||
|
season.get("url"),
|
||||||
|
'season number',
|
||||||
|
),
|
||||||
|
'series': season.get('title'),
|
||||||
|
} for season in traverse_obj(series_data, ('content', 1, 'tabs')) if season.get("url").endswith("sezon/")]
|
||||||
|
|
||||||
|
return {
|
||||||
|
'_type': 'playlist',
|
||||||
|
'entries': entries,
|
||||||
|
**traverse_obj(series_data, {
|
||||||
|
'id': ('content', 0, 'projectId'),
|
||||||
|
'title': ('content', 0, 'widgets', 1, 'title'),
|
||||||
|
'series': ('content', 0, 'widgets', 1, 'title'),
|
||||||
|
'description': ('content', 0, 'widgets', 1, 'description')
|
||||||
|
}),
|
||||||
|
}
|
Loading…
Reference in New Issue
Block a user