1
0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-12-22 16:06:49 +00:00

[Whyp] Add extractor back-ported from yt-dlp

* from https://github.com/yt-dlp/yt-dlp/pull/6803, thanks CoryTibbettsDev
This commit is contained in:
dirkf 2023-05-03 10:04:46 +01:00
parent 4566e6e53e
commit eaaf4c6736
2 changed files with 79 additions and 0 deletions

View File

@ -1576,6 +1576,7 @@ from .weibo import (
WeiboMobileIE
)
from .weiqitv import WeiqiTVIE
from .whyp import WhypIE
from .wistia import (
WistiaIE,
WistiaPlaylistIE,

View File

@ -0,0 +1,78 @@
# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
float_or_none,
merge_dicts,
str_or_none,
T,
traverse_obj,
url_or_none,
)
class WhypIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?whyp\.it/tracks/(?P<id>\d+)'
_TESTS = [{
'url': 'https://www.whyp.it/tracks/18337/home-page-example-track-b4kq7',
'md5': 'c1187b42ebf8605284e3dc92aeb33d16',
'info_dict': {
'url': 'https://cdn.whyp.it/50eb17cc-e9ff-4e18-b89b-dc9206a95cb1.mp3',
'id': '18337',
'title': 'Home Page Example Track',
'description': 'md5:bd758000fb93f3159339c852b5b9133c',
'ext': 'mp3',
'duration': 52.82,
'uploader': 'Brad',
'uploader_id': '1',
'thumbnail': 'https://cdn.whyp.it/a537bb36-3373-4c61-96c8-27fc1b2f427a.jpg',
},
}, {
'url': 'https://www.whyp.it/tracks/18337',
'only_matching': True,
}]
def _search_nuxt_data(self, webpage, video_id, context_name='__NUXT__', fatal=True, traverse=('data', 0)):
"""Parses Nuxt.js metadata. This works as long as the function __NUXT__ invokes is a pure function"""
import functools
import json
import re
from ..utils import (js_to_json, NO_DEFAULT)
re_ctx = re.escape(context_name)
FUNCTION_RE = r'\(function\((?P<arg_keys>.*?)\){return\s+(?P<js>{.*?})\s*;?\s*}\((?P<arg_vals>.*?)\)'
js, arg_keys, arg_vals = self._search_regex(
(p.format(re_ctx, FUNCTION_RE) for p in (r'<script>\s*window\.{0}={1}\s*\)\s*;?\s*</script>', r'{0}\(.*?{1}')),
webpage, context_name, group=('js', 'arg_keys', 'arg_vals'),
default=NO_DEFAULT if fatal else (None, None, None))
if js is None:
return {}
args = dict(zip(arg_keys.split(','), map(json.dumps, self._parse_json(
'[{0}]'.format(arg_vals), video_id, transform_source=js_to_json, fatal=fatal) or ())))
ret = self._parse_json(js, video_id, transform_source=functools.partial(js_to_json, vars=args), fatal=fatal)
return traverse_obj(ret, traverse) or {}
def _real_extract(self, url):
unique_id = self._match_id(url)
webpage = self._download_webpage(url, unique_id)
data = self._search_nuxt_data(webpage, unique_id)['rawTrack']
return merge_dicts({
'url': data['audio_url'],
'id': unique_id,
}, traverse_obj(data, {
'title': 'title',
'description': 'description',
'duration': ('duration', T(float_or_none)),
'uploader': ('user', 'username'),
'uploader_id': ('user', 'id', T(str_or_none)),
'thumbnail': ('artwork_url', T(url_or_none)),
}), {
'ext': 'mp3',
'vcodec': 'none',
'http_headers': {'Referer': 'https://whyp.it/'},
}, rev=True)