0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-26 02:55:17 +00:00
yt-dlp/yt_dlp/extractor/blerp.py
pukkandan 45b2ee6f4f
Update to ytdl-commit-2dd6c6e
[YouTube] Avoid crash if uploader_id extraction fails
2dd6c6edd8

Except:
    * 295736c9cba714fb5de7d1c3dd31d86e50091cf8 [jsinterp] Improve parsing
    * 384f632e8a9b61e864a26678d85b2b39933b9bae [ITV] Overhaul ITV extractor
    * 33db85c571304bbd6863e3407ad8d08764c9e53b [feat]: Add support to external downloader aria2p
2023-02-17 17:52:23 +05:30

167 lines
4.6 KiB
Python

import json
from .common import InfoExtractor
from ..utils import strip_or_none, traverse_obj
class BlerpIE(InfoExtractor):
IE_NAME = 'blerp'
_VALID_URL = r'https?://(?:www\.)?blerp\.com/soundbites/(?P<id>[0-9a-zA-Z]+)'
_TESTS = [{
'url': 'https://blerp.com/soundbites/6320fe8745636cb4dd677a5a',
'info_dict': {
'id': '6320fe8745636cb4dd677a5a',
'title': 'Samsung Galaxy S8 Over the Horizon Ringtone 2016',
'uploader': 'luminousaj',
'uploader_id': '5fb81e51aa66ae000c395478',
'ext': 'mp3',
'tags': ['samsung', 'galaxy', 's8', 'over the horizon', '2016', 'ringtone'],
}
}, {
'url': 'https://blerp.com/soundbites/5bc94ef4796001000498429f',
'info_dict': {
'id': '5bc94ef4796001000498429f',
'title': 'Yee',
'uploader': '179617322678353920',
'uploader_id': '5ba99cf71386730004552c42',
'ext': 'mp3',
'tags': ['YEE', 'YEET', 'wo ha haah catchy tune yee', 'yee']
}
}]
_GRAPHQL_OPERATIONNAME = "webBitePageGetBite"
_GRAPHQL_QUERY = (
'''query webBitePageGetBite($_id: MongoID!) {
web {
biteById(_id: $_id) {
...bitePageFrag
__typename
}
__typename
}
}
fragment bitePageFrag on Bite {
_id
title
userKeywords
keywords
color
visibility
isPremium
owned
price
extraReview
isAudioExists
image {
filename
original {
url
__typename
}
__typename
}
userReactions {
_id
reactions
createdAt
__typename
}
topReactions
totalSaveCount
saved
blerpLibraryType
license
licenseMetaData
playCount
totalShareCount
totalFavoriteCount
totalAddedToBoardCount
userCategory
userAudioQuality
audioCreationState
transcription
userTranscription
description
createdAt
updatedAt
author
listingType
ownerObject {
_id
username
profileImage {
filename
original {
url
__typename
}
__typename
}
__typename
}
transcription
favorited
visibility
isCurated
sourceUrl
audienceRating
strictAudienceRating
ownerId
reportObject {
reportedContentStatus
__typename
}
giphy {
mp4
gif
__typename
}
audio {
filename
original {
url
__typename
}
mp3 {
url
__typename
}
__typename
}
__typename
}
''')
def _real_extract(self, url):
audio_id = self._match_id(url)
data = {
'operationName': self._GRAPHQL_OPERATIONNAME,
'query': self._GRAPHQL_QUERY,
'variables': {
'_id': audio_id
}
}
headers = {
'Content-Type': 'application/json'
}
json_result = self._download_json('https://api.blerp.com/graphql',
audio_id, data=json.dumps(data).encode('utf-8'), headers=headers)
bite_json = json_result['data']['web']['biteById']
info_dict = {
'id': bite_json['_id'],
'url': bite_json['audio']['mp3']['url'],
'title': bite_json['title'],
'uploader': traverse_obj(bite_json, ('ownerObject', 'username'), expected_type=strip_or_none),
'uploader_id': traverse_obj(bite_json, ('ownerObject', '_id'), expected_type=strip_or_none),
'ext': 'mp3',
'tags': list(filter(None, map(strip_or_none, (traverse_obj(bite_json, 'userKeywords', expected_type=list) or []))) or None)
}
return info_dict