yt-dlp/yt_dlp/extractor/huya.py
hugepower 25c1cdaa26
[ie/huya:video] Add extractor (#10686)
Closes #10679
Authored by: hugepower
2024-09-13 23:12:38 +00:00

212 lines
7.5 KiB
Python

import base64
import hashlib
import random
import re
import urllib.parse
from .common import InfoExtractor
from ..utils import (
ExtractorError,
int_or_none,
parse_duration,
str_or_none,
try_get,
unescapeHTML,
unified_strdate,
update_url_query,
url_or_none,
)
from ..utils.traversal import traverse_obj
class HuyaLiveIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.|m\.)?huya\.com/(?!(?:video/play/))(?P<id>[^/#?&]+)(?:\D|$)'
IE_NAME = 'huya:live'
IE_DESC = 'huya.com'
TESTS = [{
'url': 'https://www.huya.com/572329',
'info_dict': {
'id': '572329',
'title': str,
'ext': 'flv',
'description': str,
'is_live': True,
'view_count': int,
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.huya.com/xiaoyugame',
'only_matching': True,
}]
_RESOLUTION = {
'蓝光': {
'width': 1920,
'height': 1080,
},
'超清': {
'width': 1280,
'height': 720,
},
'流畅': {
'width': 800,
'height': 480,
},
}
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id=video_id)
stream_data = self._search_json(r'stream:\s', webpage, 'stream', video_id=video_id, default=None)
room_info = try_get(stream_data, lambda x: x['data'][0]['gameLiveInfo'])
if not room_info:
raise ExtractorError('Can not extract the room info', expected=True)
title = room_info.get('roomName') or room_info.get('introduction') or self._html_extract_title(webpage)
screen_type = room_info.get('screenType')
live_source_type = room_info.get('liveSourceType')
stream_info_list = stream_data['data'][0]['gameStreamInfoList']
if not stream_info_list:
raise ExtractorError('Video is offline', expected=True)
formats = []
for stream_info in stream_info_list:
stream_url = stream_info.get('sFlvUrl')
if not stream_url:
continue
stream_name = stream_info.get('sStreamName')
re_secret = not screen_type and live_source_type in (0, 8, 13)
params = dict(urllib.parse.parse_qsl(unescapeHTML(stream_info['sFlvAntiCode'])))
fm, ss = '', ''
if re_secret:
fm, ss = self.encrypt(params, stream_info, stream_name)
for si in stream_data.get('vMultiStreamInfo'):
display_name, bitrate = re.fullmatch(
r'(.+?)(?:(\d+)M)?', si.get('sDisplayName')).groups()
rate = si.get('iBitRate')
if rate:
params['ratio'] = rate
else:
params.pop('ratio', None)
if bitrate:
rate = int(bitrate) * 1000
if re_secret:
params['wsSecret'] = hashlib.md5(
'_'.join([fm, params['u'], stream_name, ss, params['wsTime']]))
formats.append({
'ext': stream_info.get('sFlvUrlSuffix'),
'format_id': str_or_none(stream_info.get('iLineIndex')),
'tbr': rate,
'url': update_url_query(f'{stream_url}/{stream_name}.{stream_info.get("sFlvUrlSuffix")}',
query=params),
**self._RESOLUTION.get(display_name, {}),
})
return {
'id': video_id,
'title': title,
'formats': formats,
'view_count': room_info.get('totalCount'),
'thumbnail': room_info.get('screenshot'),
'description': room_info.get('contentIntro'),
'http_headers': {
'Origin': 'https://www.huya.com',
'Referer': 'https://www.huya.com/',
},
}
def encrypt(self, params, stream_info, stream_name):
ct = int_or_none(params.get('wsTime'), 16) + random.random()
presenter_uid = stream_info['lPresenterUid']
if not stream_name.startswith(str(presenter_uid)):
uid = presenter_uid
else:
uid = int_or_none(ct % 1e7 * 1e6 % 0xffffffff)
u1 = uid & 0xffffffff00000000
u2 = uid & 0xffffffff
u3 = uid & 0xffffff
u = u1 | u2 >> 24 | u3 << 8
params.update({
'u': str_or_none(u),
'seqid': str_or_none(int_or_none(ct * 1000) + uid),
'ver': '1',
'uuid': int_or_none(ct % 1e7 * 1e6 % 0xffffffff),
't': '100',
})
fm = base64.b64decode(params['fm']).decode().split('_', 1)[0]
ss = hashlib.md5('|'.join([params['seqid'], params['ctype'], params['t']]))
return fm, ss
class HuyaVideoIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?huya\.com/video/play/(?P<id>\d+)\.html'
IE_NAME = 'huya:video'
IE_DESC = '虎牙视频'
_TESTS = [{
'url': 'https://www.huya.com/video/play/1002412640.html',
'info_dict': {
'id': '1002412640',
'ext': 'mp4',
'title': '8月3日',
'thumbnail': r're:https?://.*\.jpg',
'duration': 14,
'uploader': '虎牙-ATS欧卡车队青木',
'uploader_id': '1564376151',
'upload_date': '20240803',
'view_count': int,
'comment_count': int,
'like_count': int,
},
},
{
'url': 'https://www.huya.com/video/play/556054543.html',
'info_dict': {
'id': '556054543',
'ext': 'mp4',
'title': '我不挑事 也不怕事',
'thumbnail': r're:https?://.*\.jpg',
'duration': 1864,
'uploader': '卡尔',
'uploader_id': '367138632',
'upload_date': '20210811',
'view_count': int,
'comment_count': int,
'like_count': int,
},
}]
def _real_extract(self, url: str):
video_id = self._match_id(url)
video_data = self._download_json(
'https://liveapi.huya.com/moment/getMomentContent', video_id,
query={'videoId': video_id})['data']['moment']['videoInfo']
formats = []
for definition in traverse_obj(video_data, ('definitions', lambda _, v: url_or_none(v['url']))):
formats.append({
'url': definition['url'],
**traverse_obj(definition, {
'format_id': ('defName', {str}),
'width': ('width', {int_or_none}),
'height': ('height', {int_or_none}),
'filesize': ('size', {int_or_none}),
}),
})
return {
'id': video_id,
'formats': formats,
**traverse_obj(video_data, {
'title': ('videoTitle', {str}),
'thumbnail': ('videoCover', {url_or_none}),
'duration': ('videoDuration', {parse_duration}),
'uploader': ('nickName', {str}),
'uploader_id': ('uid', {str_or_none}),
'upload_date': ('videoUploadTime', {unified_strdate}),
'view_count': ('videoPlayNum', {int_or_none}),
'comment_count': ('videoCommentNum', {int_or_none}),
'like_count': ('favorCount', {int_or_none}),
}),
}