mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-18 16:26:23 +00:00
[veejoy] applied suggested changes from pr
This commit is contained in:
parent
c191cda21d
commit
3b3f2dd6e1
2 changed files with 94 additions and 49 deletions
|
@ -1416,6 +1416,7 @@ from .ustudio import (
|
|||
from .varzesh3 import Varzesh3IE
|
||||
from .vbox7 import Vbox7IE
|
||||
from .veehd import VeeHDIE
|
||||
from .veejoy import VeejoyIE
|
||||
from .veoh import VeohIE
|
||||
from .vesti import VestiIE
|
||||
from .vevo import (
|
||||
|
@ -1660,5 +1661,4 @@ from .zingmp3 import (
|
|||
ZingMp3AlbumIE,
|
||||
)
|
||||
from .zoom import ZoomIE
|
||||
from .zype import ZypeIE
|
||||
from .veejoy import VeejoyIE
|
||||
from .zype import ZypeIE
|
|
@ -1,8 +1,19 @@
|
|||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
import json
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..compat import compat_str
|
||||
from ..utils import (
|
||||
ExtractorError,
|
||||
determine_ext,
|
||||
int_or_none,
|
||||
parse_iso8601,
|
||||
qualities,
|
||||
strip_or_none,
|
||||
try_get,
|
||||
update_url_query,
|
||||
url_or_none,
|
||||
)
|
||||
|
||||
|
||||
class VeejoyIE(InfoExtractor):
|
||||
|
@ -14,7 +25,9 @@ class VeejoyIE(InfoExtractor):
|
|||
'ext': 'mp4',
|
||||
'title': 'On-ride Tyrol Log Flume',
|
||||
'description': 'Through the ‘magical world of diamonds’ and straight into the cool water. Experience a different kind of water slide with the ‘Tyrol Log Flume’. One of the oldest and most popular attractions in the park!',
|
||||
'uploader': 'MACK Media'
|
||||
'uploader': 'MACK Media',
|
||||
'upload_date': '20210923',
|
||||
'timestamp': 1632388920
|
||||
}
|
||||
}, {
|
||||
'url': 'https://www.veejoy.de/en/movies/off-to-rulantica',
|
||||
|
@ -23,7 +36,9 @@ class VeejoyIE(InfoExtractor):
|
|||
'ext': 'mp4',
|
||||
'title': 'Off to Rulantica',
|
||||
'description': 'Rocking through the water on round boats, creating splashy fun with water cannons – and then, sliding down ‘Svalgurok’ on ten different slides: Soaking wet water fun is calling.',
|
||||
'uploader': 'Veejoy'
|
||||
'uploader': 'Veejoy',
|
||||
'upload_date': '20220811',
|
||||
'timestamp': 1660206600
|
||||
}
|
||||
}, {
|
||||
'url': 'https://www.veejoy.de/de/series/o-the-construction-documentary/the-building-site-grows',
|
||||
|
@ -32,83 +47,113 @@ class VeejoyIE(InfoExtractor):
|
|||
'ext': 'mp4',
|
||||
'title': 'Bau-„Leiter“',
|
||||
'description': 'Auf der Baustelle ist viel passiert. Patrick und Lukas bekommen ein Update vom Bauleiter, erklären technische Grundlagen am „lebenden Objekt“ und stellen sich einer Onride-Challenge.',
|
||||
'uploader': 'MACK Media'
|
||||
'uploader': 'MACK Media',
|
||||
'timestamp': 1658997000,
|
||||
'upload_date': '20220728'
|
||||
}
|
||||
}]
|
||||
|
||||
def get_video_id(self, url):
|
||||
return self._match_id(url)
|
||||
|
||||
def get_video_data(self, url, video_id):
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
next_data = self._html_search_regex(r'<script id="__NEXT_DATA__" type="application/json">([^<]+)</script>', webpage, 'next_data')
|
||||
return json.loads(next_data)["props"]["pageProps"]["media"]
|
||||
def _search_nextjs_data(self, webpage, video_id, transform_source=None, fatal=True, **kw):
|
||||
return self._parse_json(
|
||||
self._search_regex(
|
||||
r'(?s)<script[^>]+id=[\'"]__NEXT_DATA__[\'"][^>]*>([^<]+)</script>',
|
||||
webpage, 'next.js data', fatal=fatal, **kw),
|
||||
video_id, transform_source=transform_source, fatal=fatal)
|
||||
|
||||
def get_producer(self, video_data):
|
||||
if "item" in video_data["studioDetails"]:
|
||||
return video_data["studioDetails"]["item"]["title"]
|
||||
else:
|
||||
return "Veejoy"
|
||||
return (
|
||||
strip_or_none(
|
||||
try_get(video_data, lambda x: x['studioDetails']['item']['title'], compat_str))
|
||||
or 'Veejoy')
|
||||
|
||||
def get_thumbnails(self, video_data):
|
||||
thumbnails = []
|
||||
|
||||
thumb_3_4 = video_data["teaserImage"]["3_4"]
|
||||
if thumb_3_4:
|
||||
thumbnails.append({
|
||||
'url': thumb_3_4["srcSet"][1].split(" ")[0],
|
||||
})
|
||||
|
||||
thumb_16_9 = video_data["teaserImage"]["16_9"]
|
||||
if thumb_16_9:
|
||||
thumbnails.append({
|
||||
'url': thumb_16_9["srcSet"][1].split(" ")[0],
|
||||
})
|
||||
for res in ('3_4', '16_9'):
|
||||
thumb = try_get(video_data, lambda x: x['teaserImage'][res], dict)
|
||||
if not thumb:
|
||||
continue
|
||||
thumb = url_or_none(try_get(thumb, lambda x: x['srcSet'][1].split(' ')[0]))
|
||||
if thumb:
|
||||
thumbnails.append({
|
||||
'id': res,
|
||||
'url': thumb,
|
||||
})
|
||||
|
||||
return thumbnails
|
||||
|
||||
def get_asset_ref(self, video_data):
|
||||
for mediaAsset in video_data["mediaAssets"]:
|
||||
if mediaAsset["type"] == "SOURCE":
|
||||
return mediaAsset["assetReference"]
|
||||
for mediaAsset in video_data['mediaAssets']:
|
||||
if mediaAsset.get('type') == 'SOURCE':
|
||||
return mediaAsset.get('assetReference')
|
||||
|
||||
def get_asset_formats(self, video_data, video_id):
|
||||
asset = self._download_json("https://www.veejoy.de/api/service/get-media-summary?mediaIri=" + self.get_asset_ref(video_data) + "&locale=en", video_id)
|
||||
return asset["assetFormats"]
|
||||
return self._download_json(
|
||||
update_url_query('https://www.veejoy.de/api/service/get-media-summary', {
|
||||
'mediaIri': self.get_asset_ref(video_data),
|
||||
'locale': 'en'
|
||||
}),
|
||||
video_id).get('assetFormats')
|
||||
|
||||
def get_original_file_url(self, video_data, video_id):
|
||||
for asset_format in self.get_asset_formats(video_data, video_id):
|
||||
if asset_format["mimeType"] == "video/mp4":
|
||||
if asset_format.get('mimeType') == 'video/mp4':
|
||||
return asset_format
|
||||
|
||||
def get_video_formats(self, asset_formats):
|
||||
def get_video_formats(self, asset_formats, video_id):
|
||||
# This function is currently faulty and thus not used
|
||||
formats = []
|
||||
|
||||
q = qualities(['hq', 'mq', 'lq'])
|
||||
|
||||
for asset_format in asset_formats:
|
||||
if asset_format["mimeType"] == "application/vnd.apple.mpegurl":
|
||||
formats.append({
|
||||
'url': asset_format["contentUrl"],
|
||||
'width': asset_format["transcodingFormat"]["videoWidth"],
|
||||
'quality': asset_format["transcodingFormat"]["id"],
|
||||
'language': asset_format["language"],
|
||||
})
|
||||
f_url = url_or_none(asset_format.get('contentUrl'))
|
||||
if not f_url:
|
||||
continue
|
||||
ext = determine_ext(f_url)
|
||||
transcodingFormat = try_get(asset_format, lambda x: x['transcodingFormat'], dict) or {}
|
||||
|
||||
if transcodingFormat == {}:
|
||||
continue
|
||||
|
||||
label = strip_or_none(transcodingFormat.get('label') or '').split('-')
|
||||
extra = (
|
||||
('width', int_or_none(transcodingFormat.get('videoWidth'))),
|
||||
('quality', q(label[0])),
|
||||
('language', asset_format.get('language')),
|
||||
)
|
||||
if ext == 'm3u8':
|
||||
# expect 'mimeType': 'application/vnd.apple.mpegurl'
|
||||
fmts = self._extract_m3u8_formats(
|
||||
# if the yt-dl HLS downloader doesn't work: `entry_protocol='m3u8'`
|
||||
f_url, video_id, ext='mp4', entry_protocol='m3u8',
|
||||
m3u8_id=transcodingFormat.get('formatType'), fatal=False)
|
||||
for f in fmts:
|
||||
f.update((k, v) for k, v in extra if f.get(k) is None)
|
||||
formats.extend(fmts)
|
||||
else:
|
||||
# expect 'mimeType': 'video/mp4'
|
||||
fmt = {'url': f_url}
|
||||
fmt.update(extra)
|
||||
formats.append(fmt)
|
||||
|
||||
return formats
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self.get_video_id(url)
|
||||
video_data = self.get_video_data(url, video_id)
|
||||
video_id = self._match_id(url)
|
||||
video_data = self._search_nextjs_data(self._download_webpage(url, video_id), video_id).get('props').get('pageProps').get('media')
|
||||
title = video_data.get('title')
|
||||
final_url = self.get_original_file_url(video_data, video_id).get('contentUrl')
|
||||
producer = self.get_producer(video_data)
|
||||
thumbnails = self.get_thumbnails(video_data)
|
||||
final_asset = self.get_original_file_url(video_data, video_id)
|
||||
|
||||
return {
|
||||
'url': final_asset.get("contentUrl"),
|
||||
'url': final_url,
|
||||
'id': video_id,
|
||||
'title': video_data.get("title"),
|
||||
'description': video_data.get("shortDescription"),
|
||||
'duration': video_data.get("mediaDuration"),
|
||||
'title': title,
|
||||
'timestamp': parse_iso8601(video_data.get('liveDate')),
|
||||
'description': strip_or_none(video_data.get('shortDescription')),
|
||||
'duration': int_or_none(video_data.get('mediaDuration')),
|
||||
'uploader': producer,
|
||||
'creator': producer,
|
||||
'thumbnails': thumbnails,
|
||||
|
|
Loading…
Reference in a new issue