import base64 import json import hashlib import random import re import time from .common import InfoExtractor from ..utils import ( clean_html, int_or_none, join_nonempty, strip_jsonp, str_or_none, traverse_obj, unescapeHTML, url_or_none, ExtractorError, UserNotLive, ) class QQMusicBaseIE(InfoExtractor): def _get_sign(self, payload: bytes): # This may not work for domains other than `y.qq.com` and browswer UA that contains `Headless` md5hex = hashlib.md5(payload).hexdigest().upper() hex_digits = [int(c, base=16) for c in md5hex] xor_digits = [] for i, xor in enumerate([212, 45, 80, 68, 195, 163, 163, 203, 157, 220, 254, 91, 204, 79, 104, 6]): xor_digits.append((hex_digits[i * 2] * 16 + hex_digits[i * 2 + 1]) ^ xor) char_indicies = [] for i in range(0, len(xor_digits) - 1, 3): char_indicies.extend([ xor_digits[i] >> 2, ((xor_digits[i] & 3) << 4) | (xor_digits[i + 1] >> 4), ((xor_digits[i + 1] & 15) << 2) | (xor_digits[i + 2] >> 6), xor_digits[i + 2] & 63, ]) char_indicies.extend([ xor_digits[15] >> 2, (xor_digits[15] & 3) << 4, ]) head = ''.join(md5hex[i] for i in [21, 4, 9, 26, 16, 20, 27, 30]) tail = ''.join(md5hex[i] for i in [18, 11, 3, 2, 1, 7, 6, 25]) body = ''.join('ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/='[i] for i in char_indicies) return re.sub(r'[\/+]', '', f'zzb{head}{body}{tail}'.lower()) def _get_g_tk(self): n = 5381 for chr in self._get_cookies('https://y.qq.com').get('qqmusic_key', ''): n += (n << 5) + ord(chr) return n & 2147483647 # Reference: m_r_GetRUin() in top_player.js # http://imgcache.gtimg.cn/music/portal_v3/y/top_player.js @staticmethod def m_r_get_ruin(): curMs = int(time.time() * 1000) % 1000 return int(round(random.random() * 2147483647) * curMs % 1E10) def download_init_data(self, url, mid): webpage = self._download_webpage(url, mid) return self._search_json( r'window\.__INITIAL_DATA__\s*=\s*', webpage.replace('undefined', 'null'), 'init data', mid) def make_fcu_req(self, mid, req_dict, **kwargs): payload = json.dumps({ 'comm': { 'cv': 4747474, 'ct': 24, 'format': 'json', 'inCharset': 'utf-8', 'outCharset': 'utf-8', 'notice': 0, 'platform': 'yqq.json', 'needNewCode': 1, 'uin': int_or_none(self._get_cookies('https://y.qq.com').get('o_cookie')) or 0, 'g_tk_new_20200303': self._get_g_tk(), 'g_tk': self._get_g_tk(), }, **req_dict }, separators=(',', ':')).encode('utf-8') return self._download_json( 'https://u.y.qq.com/cgi-bin/musics.fcg', mid, data=payload, query={'_': int(time.time()), 'sign': self._get_sign(payload)}, **kwargs) class QQMusicIE(QQMusicBaseIE): IE_NAME = 'qqmusic' IE_DESC = 'QQ音乐' _VALID_URL = r'https?://y\.qq\.com/n/ryqq/songDetail/(?P[0-9A-Za-z]+)' _TESTS = [{ 'url': 'https://y.qq.com/n/ryqq/songDetail/004Ti8rT003TaZ', 'md5': 'd7adc5c438d12e2cb648cca81593fd47', 'info_dict': { 'id': '004Ti8rT003TaZ', 'ext': 'mp3', 'title': '永夜のパレード (永夜的游行)', 'release_date': '20111230', 'duration': 281, 'creators': ['ケーキ姫', 'JUMA'], 'album': '幻想遊園郷 -Fantastic Park-', 'description': 'md5:b5261f3d595657ae561e9e6aee7eb7d9', 'size': 4501244, 'thumbnail': r're:^https?://.*\.jpg(?:$|[#?])', 'subtitles': 'count:1', }, 'params': {'listsubtitles': True}, }, { 'url': 'https://y.qq.com/n/ryqq/songDetail/004295Et37taLD', 'md5': '5f1e6cea39e182857da7ffc5ef5e6bb8', 'info_dict': { 'id': '004295Et37taLD', 'ext': 'mp3', 'title': '可惜没如果', 'album': '新地球 - 人 (Special Edition)', 'release_date': '20150129', 'duration': 298, 'creators': ['林俊杰'], 'description': 'md5:f568421ff618d2066e74b65a04149c4e', 'thumbnail': r're:^https?://.*\.jpg(?:$|[#?])', }, 'skip': 'premium member only', }, { 'note': 'There is no mp3-320 version of this song.', 'url': 'https://y.qq.com/n/ryqq/songDetail/004MsGEo3DdNxV', 'md5': '028aaef1ae13d8a9f4861a92614887f9', 'info_dict': { 'id': '004MsGEo3DdNxV', 'ext': 'mp3', 'title': '如果', 'album': '新传媒电视连续剧金曲系列II', 'release_date': '20050626', 'duration': 220, 'creators': ['李季美'], 'description': 'md5:fc711212aa623b28534954dc4bd67385', 'size': 3535730, 'thumbnail': r're:^https?://.*\.jpg(?:$|[#?])', }, }, { 'note': 'lyrics not in .lrc format', 'url': 'https://y.qq.com/n/ryqq/songDetail/001JyApY11tIp6', 'info_dict': { 'id': '001JyApY11tIp6', 'ext': 'mp3', 'title': 'Shadows Over Transylvania', 'release_date': '19970225', 'creator': 'Dark Funeral', 'description': 'md5:c9b20210587cbcd6836a1c597bab4525', 'thumbnail': r're:^https?://.*\.jpg(?:$|[#?])', }, 'params': {'skip_download': True}, 'skip': 'no longer available', }] _FORMATS = { 'F000': {'name': 'flac', 'prefix': 'F000', 'ext': 'flac', 'preference': 60}, 'A000': {'name': 'ape', 'prefix': 'A000', 'ext': 'ape', 'preference': 50}, 'M800': {'name': '320mp3', 'prefix': 'M800', 'ext': 'mp3', 'preference': 40, 'abr': 320}, 'M500': {'name': '128mp3', 'prefix': 'M500', 'ext': 'mp3', 'preference': 30, 'abr': 128}, 'C400': {'name': '96aac', 'prefix': 'C400', 'ext': 'm4a', 'preference': 20, 'abr': 96}, 'C200': {'name': '48aac', 'prefix': 'C200', 'ext': 'm4a', 'preference': 20, 'abr': 48}, } def _real_extract(self, url): mid = self._match_id(url) init_data = self.download_init_data(url, mid) media_id = traverse_obj(init_data, ( 'songList', lambda _, v: v['mid'] == mid, 'file', 'media_mid'), get_all=False) data = self.make_fcu_req(mid, { 'req_1': { 'module': 'vkey.GetVkeyServer', 'method': 'CgiGetVkey', 'param': { 'guid': str(self.m_r_get_ruin()), 'songmid': [mid], 'songtype': [0], 'uin': self._get_cookies('https://y.qq.com').get('o_cookie', '0'), 'loginflag': 1, 'platform': '20', **({ 'songmid': [mid] * len(self._FORMATS), 'songtype': [0] * len(self._FORMATS), 'filename': [f'{f["prefix"]}{media_id}.{f["ext"]}' for f in self._FORMATS.values()], } if media_id else {}), } }, 'req_2': { 'module': 'music.musichallSong.PlayLyricInfo', 'method': 'GetPlayLyricInfo', 'param': { 'songMID': mid, 'songID': traverse_obj(init_data, ('detail', 'id', {int})), } }}) formats = traverse_obj(data, ('req_1', 'data', 'midurlinfo', lambda _, v: v['songmid'] == mid and v['purl'], { 'url': ('purl', {str}, {lambda x: f'https://dl.stream.qqmusic.qq.com/{x}'}), 'format': ('filename', {lambda x: self._FORMATS[x[:4]]['name']}), 'format_id': ('filename', {lambda x: self._FORMATS[x[:4]]['name']}), 'size': ('filename', {lambda x: self._FORMATS[x[:4]]['name']}, {lambda x: traverse_obj(init_data, ('songList', ..., 'file', f'size_{x}'), get_all=False)}), 'quality': ('filename', {lambda x: self._FORMATS[x[:4]]['preference']}), 'abr': ('filename', {lambda x: self._FORMATS[x[:4]]['abr']}), })) lrc_content = traverse_obj(data, ('req_2', 'data', 'lyric', {lambda x: base64.b64decode(x).decode('utf-8')})) info_dict = { 'id': mid, 'formats': formats, **traverse_obj(init_data, ('detail', { 'title': ('title', {str}), 'album': ('albumName', {str}, {lambda x: x or None}), 'thumbnail': ('picurl', {url_or_none}), 'release_date': ('ctime', {lambda x: x.replace('-', '') or None}), 'description': ('info', 'intro', 'content', ..., 'value', {str}), }), get_all=False), **traverse_obj(init_data, ('songList', lambda _, v: v['mid'] == mid, { 'alt_title': ('subtitle', {str}, {lambda x: x or None}), 'duration': ('interval', {int}), }), get_all=False), 'creators': traverse_obj(init_data, ('detail', 'singer', ..., 'name')), } if lrc_content: info_dict['subtitles'] = {'origin': [{'ext': 'lrc', 'data': lrc_content}]} info_dict['description'] = join_nonempty(info_dict.get('description'), lrc_content, delim='\n') return info_dict class QQMusicSingerIE(QQMusicBaseIE): IE_NAME = 'qqmusic:singer' IE_DESC = 'QQ音乐 - 歌手' _VALID_URL = r'https?://y\.qq\.com/n/ryqq/singer/(?P[0-9A-Za-z]+)' _TESTS = [{ 'url': 'https://y.qq.com/n/ryqq/singer/001BLpXF2DyJe2', 'info_dict': { 'id': '001BLpXF2DyJe2', 'title': '林俊杰', 'description': 'md5:10624ce73b06fa400bc846f59b0305fa', 'thumbnail': r're:^https?://.*\.jpg(?:$|[#?])', }, 'playlist_mincount': 10, }, { 'url': 'https://y.qq.com/n/ryqq/singer/000Q00f213YzNV', 'info_dict': { 'id': '000Q00f213YzNV', 'title': '桃几OvO', 'description': '小破站小唱见~希望大家喜欢听我唱歌~!', 'thumbnail': r're:^https?://.*\.jpg(?:$|[#?])', }, 'playlist_count': 12, 'playlist': [{ 'info_dict': { 'id': '0016cvsy02mmCl', 'ext': 'mp3', 'title': '群青', 'release_date': '20210913', 'duration': 248, 'creators': ['桃几OvO'], 'album': '桃几2021年翻唱集', 'description': 'md5:4296005a04edcb5cdbe0889d5055a7ae', 'size': 3970822, 'thumbnail': r're:^https?://.*\.jpg(?:$|[#?])', }, }], }] def _entries(self, mid, init_data): size = 50 max_num = traverse_obj(init_data, ('singerDetail', 'songTotalNum')) for page in range(0, max_num // size + 1): data = self.make_fcu_req(mid, {'req_1': { 'module': 'music.web_singer_info_svr', 'method': 'get_singer_detail_info', 'param': { 'sort': 5, 'singermid': mid, 'sin': page * size, 'num': size, }}}, note=f'Downloading page {page}') yield from traverse_obj(data, ('req_1', 'data', 'songlist', ..., {lambda x: self.url_result( f'https://y.qq.com/n/ryqq/songDetail/{x["mid"]}', QQMusicIE, x['mid'], x.get('title'))})) def _real_extract(self, url): mid = self._match_id(url) init_data = self.download_init_data(url, mid) return self.playlist_result( self._entries(mid, init_data), mid, **traverse_obj(init_data, ('singerDetail', { 'title': ('basic_info', 'name', {str}), 'description': ('ex_info', 'desc', {str}), 'thumbnail': ('pic', 'pic', {url_or_none}), }))) class QQPlaylistBaseIE(InfoExtractor): def _extract_entries(self, info_json, path): return traverse_obj(info_json, (*path, {lambda song: self.url_result( f'https://y.qq.com/n/ryqq/songDetail/{song["songmid"]}', QQMusicIE, song['songmid'], song['songname'])})) class QQMusicAlbumIE(QQPlaylistBaseIE): IE_NAME = 'qqmusic:album' IE_DESC = 'QQ音乐 - 专辑' _VALID_URL = r'https?://y\.qq\.com/n/ryqq/albumDetail/(?P[0-9A-Za-z]+)' _TESTS = [{ 'url': 'https://y.qq.com/n/ryqq/albumDetail/000gXCTb2AhRR1', 'info_dict': { 'id': '000gXCTb2AhRR1', 'title': '我们都是这样长大的', 'description': 'md5:179c5dce203a5931970d306aa9607ea6', }, 'playlist_count': 4, }, { 'url': 'https://y.qq.com/n/ryqq/albumDetail/002Y5a3b3AlCu3', 'info_dict': { 'id': '002Y5a3b3AlCu3', 'title': '그리고…', 'description': 'md5:a48823755615508a95080e81b51ba729', }, 'playlist_count': 8, }] def _real_extract(self, url): mid = self._match_id(url) album_json = self._download_json( 'http://i.y.qq.com/v8/fcg-bin/fcg_v8_album_info_cp.fcg', mid, 'Download album page', query={'albummid': mid, 'format': 'json'})['data'] entries = self._extract_entries(album_json, ('list', ...)) return self.playlist_result(entries, mid, **traverse_obj(album_json, { 'title': ('name', {str}), 'description': ('desc', {str}, {str.strip}), })) class QQMusicToplistIE(QQPlaylistBaseIE): IE_NAME = 'qqmusic:toplist' IE_DESC = 'QQ音乐 - 排行榜' _VALID_URL = r'https?://y\.qq\.com/n/ryqq/toplist/(?P[0-9]+)' _TESTS = [{ 'url': 'https://y.qq.com/n/ryqq/toplist/123', 'info_dict': { 'id': '123', 'title': r're:美国热门音乐榜 \d{4}-\d{2}-\d{2}', 'description': '美国热门音乐榜,每周一更新。', }, 'playlist_count': 95, }, { 'url': 'https://y.qq.com/n/ryqq/toplist/3', 'info_dict': { 'id': '3', 'title': r're:巅峰榜·欧美 \d{4}-\d{2}-\d{2}', 'description': 'md5:4def03b60d3644be4c9a36f21fd33857', }, 'playlist_count': 100, }, { 'url': 'https://y.qq.com/n/ryqq/toplist/106', 'info_dict': { 'id': '106', 'title': r're:韩国Mnet榜 \d{4}-\d{2}-\d{2}', 'description': 'md5:cb84b325215e1d21708c615cac82a6e7', }, 'playlist_count': 50, }] def _real_extract(self, url): list_id = self._match_id(url) toplist_json = self._download_json( 'http://i.y.qq.com/v8/fcg-bin/fcg_v8_toplist_cp.fcg', list_id, note='Download toplist page', query={'type': 'toplist', 'topid': list_id, 'format': 'json'}) entries = self._extract_entries(toplist_json, ('songlist', ..., 'data')) list_name = join_nonempty(*traverse_obj(toplist_json, ((('topinfo', 'ListName'), 'update_time'),)), delim=' ') list_description = traverse_obj(toplist_json, ('topinfo', 'info')) return self.playlist_result(entries, list_id, list_name, list_description) class QQMusicPlaylistIE(QQPlaylistBaseIE): IE_NAME = 'qqmusic:playlist' IE_DESC = 'QQ音乐 - 歌单' _VALID_URL = r'https?://y\.qq\.com/n/ryqq/playlist/(?P[0-9]+)' _TESTS = [{ 'url': 'https://y.qq.com/n/ryqq/playlist/1374105607', 'info_dict': { 'id': '1374105607', 'title': '易入人心的华语民谣', 'description': '民谣的歌曲易于传唱、、歌词朗朗伤口、旋律简单温馨。属于那种才入耳孔。却上心头的感觉。没有太多的复杂情绪。简单而直接地表达乐者的情绪,就是这样的简单才易入人心。', }, 'playlist_count': 20, }] def _real_extract(self, url): list_id = self._match_id(url) list_json = self._download_json( 'http://i.y.qq.com/qzone-music/fcg-bin/fcg_ucc_getcdinfo_byids_cp.fcg', list_id, 'Download list page', query={'type': 1, 'json': 1, 'utf8': 1, 'onlysong': 0, 'disstid': list_id}, transform_source=strip_jsonp, headers={'Referer': url}) if not len(list_json.get('cdlist', [])): error_msg = '' if list_json.get('code') or list_json.get('subcode'): error_msg = f': Error {list_json.get("code")}-{list_json["subcode"]}: {list_json.get("msg", "")}' raise ExtractorError(f'Unable to get playlist info{error_msg}') entries = self._extract_entries(list_json, ('cdlist', 0, 'songlist', ...)) return self.playlist_result(entries, list_id, **traverse_obj(list_json, ('cdlist', 0, { 'title': ('dissname', {str}), 'description': ('desc', {lambda x: clean_html(unescapeHTML(x))}), }))) class QQMusicVideoIE(InfoExtractor): _VALID_URL = r'https?://y\.qq\.com/n/ryqq/mv/(?P[0-9A-Za-z]+)' def _parse_url_formats(self, url_data): formats = traverse_obj(url_data, ('mp4', lambda _, v: v.get('freeflow_url'), { 'url': ('freeflow_url', 0, {url_or_none}), 'filesize': ('fileSize', {int_or_none}), 'format_id': ('newFileType', {str_or_none}), })) return formats def _real_extract(self, url): video_id = self._match_id(url) payload = '{"comm":{"format":"json","platform":"yqq"},"mvInfo":{"module":"music.video.VideoData","method":"get_video_info_batch","param":{"vidlist":["%s"],"required":["vid","type","sid","cover_pic","duration","singers","new_switch_str","video_pay","hint","code","msg","name","desc","playcnt","pubdate","isfav","fileid","filesize_v2","switch_pay_type","pay","pay_info","uploader_headurl","uploader_nick","uploader_uin","uploader_encuin","play_forbid_reason"]}},"mvUrl":{"module":"music.stream.MvUrlProxy","method":"GetMvUrls","param":{"vids":["%s"]}}}' % (video_id, video_id) video_info = self._download_json('https://u.y.qq.com/cgi-bin/musicu.fcg', video_id, data=payload.encode()) if traverse_obj(video_info, ('mvInfo', 'data', video_id, 'play_forbid_reason')) == 3: self.raise_geo_restricted() return { 'id': video_id, 'formats': self._parse_url_formats(traverse_obj(video_info, ('mvUrl', 'data', video_id))), **traverse_obj(video_info, ('mvInfo', 'data', video_id, { 'title': ('name', {str_or_none}), 'description': ('desc', {str_or_none}), 'thumbnail': ('cover_pic', {url_or_none}), 'release_timestamp': ('pubdate', {int_or_none}), 'duration': ('duration', {int_or_none}), 'creator': ('singers', 0, 'name', {str_or_none}), })), } class QQMusicVideoLiveIE(InfoExtractor): _VALID_URL = r'https?://v\.y\.qq\.com/m/play\.html\?showid=(?P[0-9]+)' _HEIGHTS = { 112: 720, 110: 540, } def _parse_url_foramts(self, stream_info, stream_id): formats = [] for url in traverse_obj(stream_info, ('req_0', 'data', ('streamFLV', 'streamHLS'), {url_or_none})): formats.append({ 'url': url, 'quality': 10000, 'format': '原画', 'resolution': 'original', }) for stream in traverse_obj(stream_info, ('req_0', 'data', ('streamFlvList', 'streamHlsList'), ..., { 'url': ('stream', {url_or_none}), 'format': ('text', {str_or_none}), 'quality': ('resolution', {int_or_none}), 'height': ('resolution', {lambda i: self._HEIGHTS.get(i)}), })): if stream['url']: formats.append(stream) return formats def _real_extract(self, url): stream_id = self._match_id(url) payload = '{"comm":{"format":"json","inCharset":"utf-8","outCharset":"utf-8","notice":0,"platform":"h5","needNewCode":1},"req_0":{"module":"mlive.show.MliveShowMemberSvr","method":"GetNewestShow","param":{"showID":%s}}}' % stream_id show_info = self._download_json('https://vc.y.qq.com/cgi-bin/musicu.fcg', stream_id, data=payload.encode()) # import json # print(payload) # print(json.dumps(show_info, ensure_ascii=False)) show_id = traverse_obj(show_info, ('req_0', 'data', 'showID')) payload = '{"comm":{"format":"json","inCharset":"utf-8","outCharset":"utf-8","notice":0,"platform":"h5","needNewCode":1},"req_0":{"module":"mlive.show.MliveShowMemberSvr","method":"JoinBasicShow","param":{"showID":%s,"uinStr":"v%.0f"}}}' % (show_id, time.time() * 1000) stream_info = self._download_json('https://vc.y.qq.com/cgi-bin/musicu.fcg', stream_id, data=payload.encode()) # print(payload) # print(json.dumps(stream_info, ensure_ascii=False)) formats = self._parse_url_foramts(stream_info, stream_id) if not formats and traverse_obj(stream_info, ('req_0', 'data', 'showInfo', 'liveType')) == 0: raise UserNotLive(video_id=stream_id) return { 'id': stream_id, 'formats': self._parse_url_foramts(stream_info, stream_id), 'is_live': True, **traverse_obj(stream_info, ('req_0', 'data', 'showInfo', { 'title': ('title', {str_or_none}), 'thumbnail': ('cover_pic', {url_or_none}), 'uploader_id': ('uin', {str_or_none}), })), }