mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-23 17:02:30 +00:00
3d2623a898
Authored by: coletdjnz, pukkandan
166 lines
6.3 KiB
Python
166 lines
6.3 KiB
Python
import base64
|
||
import time
|
||
import uuid
|
||
|
||
from .common import InfoExtractor
|
||
from ..networking.exceptions import HTTPError
|
||
from ..utils import (
|
||
ExtractorError,
|
||
int_or_none,
|
||
parse_resolution,
|
||
traverse_obj,
|
||
try_get,
|
||
url_or_none,
|
||
urljoin,
|
||
)
|
||
|
||
|
||
class MGTVIE(InfoExtractor):
|
||
_VALID_URL = r'https?://(?:w(?:ww)?\.)?mgtv\.com/(v|b)/(?:[^/]+/)*(?P<id>\d+)\.html'
|
||
IE_DESC = '芒果TV'
|
||
IE_NAME = 'MangoTV'
|
||
|
||
_TESTS = [{
|
||
'url': 'http://www.mgtv.com/v/1/290525/f/3116640.html',
|
||
'info_dict': {
|
||
'id': '3116640',
|
||
'ext': 'mp4',
|
||
'title': '我是歌手 第四季',
|
||
'description': '我是歌手第四季双年巅峰会',
|
||
'duration': 7461,
|
||
'thumbnail': r're:^https?://.*\.jpg$',
|
||
},
|
||
'params': {'skip_download': 'm3u8'},
|
||
}, {
|
||
'url': 'https://w.mgtv.com/b/427837/15588271.html',
|
||
'info_dict': {
|
||
'id': '15588271',
|
||
'ext': 'mp4',
|
||
'title': '春日迟迟再出发 沉浸版第1期:陆莹结婚半年查出肾炎被离婚 吴雅婷把一半票根退给前夫',
|
||
'description': 'md5:a7a05a05b1aa87bd50cae619b19bbca6',
|
||
'thumbnail': r're:^https?://.+\.jpg',
|
||
'duration': 4026,
|
||
},
|
||
'params': {'skip_download': 'm3u8'},
|
||
}, {
|
||
'url': 'https://w.mgtv.com/b/333652/7329822.html',
|
||
'info_dict': {
|
||
'id': '7329822',
|
||
'ext': 'mp4',
|
||
'title': '拜托,请你爱我',
|
||
'description': 'md5:cd81be6499bafe32e4d143abd822bf9c',
|
||
'thumbnail': r're:^https?://.+\.jpg',
|
||
'duration': 2656,
|
||
},
|
||
'params': {'skip_download': 'm3u8'},
|
||
}, {
|
||
'url': 'https://w.mgtv.com/b/427837/15591647.html',
|
||
'only_matching': True,
|
||
}, {
|
||
'url': 'https://w.mgtv.com/b/388252/15634192.html?fpa=33318&fpos=4&lastp=ch_home',
|
||
'only_matching': True,
|
||
}, {
|
||
'url': 'http://www.mgtv.com/b/301817/3826653.html',
|
||
'only_matching': True,
|
||
}, {
|
||
'url': 'https://w.mgtv.com/b/301817/3826653.html',
|
||
'only_matching': True,
|
||
}]
|
||
|
||
_RESOLUTIONS = {
|
||
'标清': ('480p', '854x480'),
|
||
'高清': ('540p', '960x540'),
|
||
'超清': ('720p', '1280x720'),
|
||
'蓝光': ('1080p', '1920x1080'),
|
||
}
|
||
|
||
def _real_extract(self, url):
|
||
video_id = self._match_id(url)
|
||
tk2 = base64.urlsafe_b64encode(
|
||
f'did={str(uuid.uuid4())}|pno=1030|ver=0.3.0301|clit={int(time.time())}'.encode())[::-1]
|
||
try:
|
||
api_data = self._download_json(
|
||
'https://pcweb.api.mgtv.com/player/video', video_id, query={
|
||
'tk2': tk2,
|
||
'video_id': video_id,
|
||
'type': 'pch5'
|
||
}, headers=self.geo_verification_headers())['data']
|
||
except ExtractorError as e:
|
||
if isinstance(e.cause, HTTPError) and e.cause.status == 401:
|
||
error = self._parse_json(e.cause.response.read().decode(), None)
|
||
if error.get('code') == 40005:
|
||
self.raise_geo_restricted(countries=self._GEO_COUNTRIES)
|
||
raise ExtractorError(error['msg'], expected=True)
|
||
raise
|
||
|
||
stream_data = self._download_json(
|
||
'https://pcweb.api.mgtv.com/player/getSource', video_id, query={
|
||
'tk2': tk2,
|
||
'pm2': api_data['atc']['pm2'],
|
||
'video_id': video_id,
|
||
'type': 'pch5',
|
||
'src': 'intelmgtv',
|
||
}, headers=self.geo_verification_headers())['data']
|
||
stream_domain = traverse_obj(stream_data, ('stream_domain', ..., {url_or_none}), get_all=False)
|
||
|
||
formats = []
|
||
for idx, stream in enumerate(traverse_obj(stream_data, ('stream', lambda _, v: v['url']))):
|
||
stream_name = traverse_obj(stream, 'name', 'standardName', 'barName', expected_type=str)
|
||
resolution = traverse_obj(
|
||
self._RESOLUTIONS, (stream_name, 1 if stream.get('scale') == '16:9' else 0))
|
||
format_url = traverse_obj(self._download_json(
|
||
urljoin(stream_domain, stream['url']), video_id, fatal=False,
|
||
note=f'Downloading video info for format {resolution or stream_name}'),
|
||
('info', {url_or_none}))
|
||
if not format_url:
|
||
continue
|
||
tbr = int_or_none(stream.get('filebitrate') or self._search_regex(
|
||
r'_(\d+)_mp4/', format_url, 'tbr', default=None))
|
||
formats.append({
|
||
'format_id': str(tbr or idx),
|
||
'url': format_url,
|
||
'ext': 'mp4',
|
||
'tbr': tbr,
|
||
'vcodec': stream.get('videoFormat'),
|
||
'acodec': stream.get('audioFormat'),
|
||
**parse_resolution(resolution),
|
||
'protocol': 'm3u8_native',
|
||
'http_headers': {
|
||
'Referer': url,
|
||
},
|
||
'format_note': stream_name,
|
||
})
|
||
|
||
return {
|
||
'id': video_id,
|
||
'formats': formats,
|
||
**traverse_obj(api_data, ('info', {
|
||
'title': ('title', {str.strip}),
|
||
'description': ('desc', {str}),
|
||
'duration': ('duration', {int_or_none}),
|
||
'thumbnail': ('thumb', {url_or_none}),
|
||
})),
|
||
'subtitles': self.extract_subtitles(video_id, stream_domain),
|
||
}
|
||
|
||
def _get_subtitles(self, video_id, domain):
|
||
info = self._download_json(f'https://pcweb.api.mgtv.com/video/title?videoId={video_id}',
|
||
video_id, fatal=False) or {}
|
||
subtitles = {}
|
||
for sub in try_get(info, lambda x: x['data']['title']) or []:
|
||
url_sub = sub.get('url')
|
||
if not url_sub:
|
||
continue
|
||
locale = sub.get('captionSimpleName') or 'en'
|
||
sub = self._download_json(f'{domain}{url_sub}', video_id, fatal=False,
|
||
note=f'Download subtitle for locale {sub.get("name")} ({locale})') or {}
|
||
sub_url = url_or_none(sub.get('info'))
|
||
if not sub_url:
|
||
continue
|
||
subtitles.setdefault(locale.lower(), []).append({
|
||
'url': sub_url,
|
||
'name': sub.get('name'),
|
||
'ext': 'srt'
|
||
})
|
||
return subtitles
|