mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2025-01-27 16:47:52 +00:00
[ie/piramidetv] Add extractors (#10777)
Closes #10706, Closes #10708 Authored by: kclauhk, HobbyistDev, seproDev Co-authored-by: HobbyistDev <tesutonihon4@gmail.com> Co-authored-by: sepro <sepro@sepr0.com>
This commit is contained in:
parent
1ef3ee7500
commit
af2c821d74
@ -1541,6 +1541,10 @@
|
||||
PinterestCollectionIE,
|
||||
PinterestIE,
|
||||
)
|
||||
from .piramidetv import (
|
||||
PiramideTVChannelIE,
|
||||
PiramideTVIE,
|
||||
)
|
||||
from .pixivsketch import (
|
||||
PixivSketchIE,
|
||||
PixivSketchUserIE,
|
||||
|
99
yt_dlp/extractor/piramidetv.py
Normal file
99
yt_dlp/extractor/piramidetv.py
Normal file
@ -0,0 +1,99 @@
|
||||
from .common import InfoExtractor
|
||||
from ..utils import parse_iso8601, smuggle_url, unsmuggle_url, url_or_none
|
||||
from ..utils.traversal import traverse_obj
|
||||
|
||||
|
||||
class PiramideTVIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://piramide\.tv/video/(?P<id>[\w-]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://piramide.tv/video/wWtBAORdJUTh',
|
||||
'info_dict': {
|
||||
'id': 'wWtBAORdJUTh',
|
||||
'ext': 'mp4',
|
||||
'title': 'md5:79f9c8183ea6a35c836923142cf0abcc',
|
||||
'description': '',
|
||||
'thumbnail': 'https://cdn.jwplayer.com/v2/media/W86PgQDn/thumbnails/B9gpIxkH.jpg',
|
||||
'channel': 'León Picarón',
|
||||
'channel_id': 'leonpicaron',
|
||||
'timestamp': 1696460362,
|
||||
'upload_date': '20231004',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://piramide.tv/video/wcYn6li79NgN',
|
||||
'info_dict': {
|
||||
'id': 'wcYn6li79NgN',
|
||||
'ext': 'mp4',
|
||||
'title': 'ACEPTO TENER UN BEBE CON MI NOVIA\u2026? | Parte 1',
|
||||
'description': '',
|
||||
'channel': 'ARTA GAME',
|
||||
'channel_id': 'arta_game',
|
||||
'thumbnail': 'https://cdn.jwplayer.com/v2/media/cnEdGp5X/thumbnails/rHAaWfP7.jpg',
|
||||
'timestamp': 1703434976,
|
||||
'upload_date': '20231224',
|
||||
},
|
||||
}]
|
||||
|
||||
def _extract_video(self, video_id):
|
||||
video_data = self._download_json(
|
||||
f'https://hermes.piramide.tv/video/data/{video_id}', video_id, fatal=False)
|
||||
formats, subtitles = self._extract_m3u8_formats_and_subtitles(
|
||||
f'https://cdn.piramide.tv/video/{video_id}/manifest.m3u8', video_id, fatal=False)
|
||||
next_video = traverse_obj(video_data, ('video', 'next_video', 'id', {str}))
|
||||
return next_video, {
|
||||
'id': video_id,
|
||||
'formats': formats,
|
||||
'subtitles': subtitles,
|
||||
**traverse_obj(video_data, ('video', {
|
||||
'id': ('id', {str}),
|
||||
'title': ('title', {str}),
|
||||
'description': ('description', {str}),
|
||||
'thumbnail': ('media', 'thumbnail', {url_or_none}),
|
||||
'channel': ('channel', 'name', {str}),
|
||||
'channel_id': ('channel', 'id', {str}),
|
||||
'timestamp': ('date', {parse_iso8601}),
|
||||
})),
|
||||
}
|
||||
|
||||
def _entries(self, video_id):
|
||||
visited = set()
|
||||
while True:
|
||||
visited.add(video_id)
|
||||
next_video, info = self._extract_video(video_id)
|
||||
yield info
|
||||
if not next_video or next_video in visited:
|
||||
break
|
||||
video_id = next_video
|
||||
|
||||
def _real_extract(self, url):
|
||||
url, smuggled_data = unsmuggle_url(url, {})
|
||||
video_id = self._match_id(url)
|
||||
if self._yes_playlist(video_id, video_id, smuggled_data):
|
||||
return self.playlist_result(self._entries(video_id), video_id)
|
||||
return self._extract_video(video_id)[1]
|
||||
|
||||
|
||||
class PiramideTVChannelIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://piramide\.tv/channel/(?P<id>[\w-]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://piramide.tv/channel/thekalo',
|
||||
'playlist_mincount': 10,
|
||||
'info_dict': {
|
||||
'id': 'thekalo',
|
||||
},
|
||||
}]
|
||||
|
||||
def _entries(self, channel_name):
|
||||
videos = self._download_json(
|
||||
f'https://hermes.piramide.tv/channel/list/{channel_name}/date/100000', channel_name)
|
||||
for video in traverse_obj(videos, ('videos', lambda _, v: v['id'])):
|
||||
yield self.url_result(smuggle_url(
|
||||
f'https://piramide.tv/video/{video["id"]}', {'force_noplaylist': True}),
|
||||
**traverse_obj(video, {
|
||||
'id': ('id', {str}),
|
||||
'title': ('title', {str}),
|
||||
'description': ('description', {str}),
|
||||
}))
|
||||
|
||||
def _real_extract(self, url):
|
||||
channel_name = self._match_id(url)
|
||||
return self.playlist_result(self._entries(channel_name), channel_name)
|
Loading…
Reference in New Issue
Block a user