0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-27 03:03:01 +00:00
yt-dlp/yt_dlp/extractor/openrec.py
Lesmiscore (Naoya Ozaki) 6e6beffd04
[openrec] Refactor extractors (#2941)
Authored by: Lesmiscore
2022-03-09 21:08:09 +09:00

137 lines
5.2 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
ExtractorError,
get_first,
int_or_none,
traverse_obj,
unified_strdate,
unified_timestamp,
)
from ..compat import compat_str
class OpenRecBaseIE(InfoExtractor):
def _extract_pagestore(self, webpage, video_id):
return self._parse_json(
self._search_regex(r'(?m)window\.pageStore\s*=\s*(\{.+?\});$', webpage, 'window.pageStore'), video_id)
def _extract_movie(self, webpage, video_id, name, is_live):
window_stores = self._extract_pagestore(webpage, video_id)
movie_stores = [
# extract all three important data (most of data are duplicated each other, but slightly different!)
traverse_obj(window_stores, ('v8', 'state', 'movie'), expected_type=dict),
traverse_obj(window_stores, ('v8', 'movie'), expected_type=dict),
traverse_obj(window_stores, 'movieStore', expected_type=dict),
]
if not any(movie_stores):
raise ExtractorError(f'Failed to extract {name} info')
m3u8_playlists = get_first(movie_stores, 'media') or {}
formats = []
for name, m3u8_url in m3u8_playlists.items():
if not m3u8_url:
continue
formats.extend(self._extract_m3u8_formats(
m3u8_url, video_id, ext='mp4', live=is_live, m3u8_id=name))
self._sort_formats(formats)
return {
'id': video_id,
'title': get_first(movie_stores, 'title'),
'description': get_first(movie_stores, 'introduction'),
'thumbnail': get_first(movie_stores, 'thumbnailUrl'),
'formats': formats,
'uploader': get_first(movie_stores, ('channel', 'user', 'name')),
'uploader_id': get_first(movie_stores, ('channel', 'user', 'id')),
'timestamp': int_or_none(get_first(movie_stores, ['publishedAt', 'time']), scale=1000) or unified_timestamp(get_first(movie_stores, 'publishedAt')),
'is_live': is_live,
}
class OpenRecIE(OpenRecBaseIE):
IE_NAME = 'openrec'
_VALID_URL = r'https?://(?:www\.)?openrec\.tv/live/(?P<id>[^/]+)'
_TESTS = [{
'url': 'https://www.openrec.tv/live/2p8v31qe4zy',
'only_matching': True,
}, {
'url': 'https://www.openrec.tv/live/wez93eqvjzl',
'only_matching': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(f'https://www.openrec.tv/live/{video_id}', video_id)
return self._extract_movie(webpage, video_id, 'live', True)
class OpenRecCaptureIE(OpenRecBaseIE):
IE_NAME = 'openrec:capture'
_VALID_URL = r'https?://(?:www\.)?openrec\.tv/capture/(?P<id>[^/]+)'
_TESTS = [{
'url': 'https://www.openrec.tv/capture/l9nk2x4gn14',
'only_matching': True,
}, {
'url': 'https://www.openrec.tv/capture/mldjr82p7qk',
'info_dict': {
'id': 'mldjr82p7qk',
'title': 'たいじの恥ずかしい英語力',
'uploader': 'たいちゃんねる',
'uploader_id': 'Yaritaiji',
'upload_date': '20210803',
},
}]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(f'https://www.openrec.tv/capture/{video_id}', video_id)
window_stores = self._extract_pagestore(webpage, video_id)
movie_store = window_stores.get('movie')
capture_data = window_stores.get('capture')
if not capture_data:
raise ExtractorError('Cannot extract title')
formats = self._extract_m3u8_formats(
capture_data.get('source'), video_id, ext='mp4')
self._sort_formats(formats)
return {
'id': video_id,
'title': capture_data.get('title'),
'thumbnail': capture_data.get('thumbnailUrl'),
'formats': formats,
'timestamp': unified_timestamp(traverse_obj(movie_store, 'createdAt', expected_type=compat_str)),
'uploader': traverse_obj(movie_store, ('channel', 'name'), expected_type=compat_str),
'uploader_id': traverse_obj(movie_store, ('channel', 'id'), expected_type=compat_str),
'upload_date': unified_strdate(capture_data.get('createdAt')),
}
class OpenRecMovieIE(OpenRecBaseIE):
IE_NAME = 'openrec:movie'
_VALID_URL = r'https?://(?:www\.)?openrec\.tv/movie/(?P<id>[^/]+)'
_TESTS = [{
'url': 'https://www.openrec.tv/movie/nqz5xl5km8v',
'info_dict': {
'id': 'nqz5xl5km8v',
'title': '限定コミュニティ(Discord)参加方法ご説明動画',
'description': 'md5:ebd563e5f5b060cda2f02bf26b14d87f',
'thumbnail': r're:https://.+',
'uploader': 'タイキとカズヒロ',
'uploader_id': 'taiki_to_kazuhiro',
'timestamp': 1638856800,
},
}]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(f'https://www.openrec.tv/movie/{video_id}', video_id)
return self._extract_movie(webpage, video_id, 'movie', False)