mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-02 11:17:30 +01:00
273762c8d0
YouTube has removed support for the old GET based live chat API, and it's now returning 404 Authored by siikamiika
124 lines
5.1 KiB
Python
124 lines
5.1 KiB
Python
from __future__ import division, unicode_literals
|
|
|
|
import json
|
|
|
|
from .fragment import FragmentFD
|
|
from ..compat import compat_urllib_error
|
|
from ..utils import (
|
|
try_get,
|
|
RegexNotFoundError,
|
|
)
|
|
from ..extractor.youtube import YoutubeBaseInfoExtractor as YT_BaseIE
|
|
|
|
|
|
class YoutubeLiveChatReplayFD(FragmentFD):
|
|
""" Downloads YouTube live chat replays fragment by fragment """
|
|
|
|
FD_NAME = 'youtube_live_chat_replay'
|
|
|
|
def real_download(self, filename, info_dict):
|
|
video_id = info_dict['video_id']
|
|
self.to_screen('[%s] Downloading live chat' % self.FD_NAME)
|
|
|
|
fragment_retries = self.params.get('fragment_retries', 0)
|
|
test = self.params.get('test', False)
|
|
|
|
ctx = {
|
|
'filename': filename,
|
|
'live': True,
|
|
'total_frags': None,
|
|
}
|
|
|
|
ie = YT_BaseIE(self.ydl)
|
|
|
|
def dl_fragment(url, data=None, headers=None):
|
|
http_headers = info_dict.get('http_headers', {})
|
|
if headers:
|
|
http_headers = http_headers.copy()
|
|
http_headers.update(headers)
|
|
return self._download_fragment(ctx, url, info_dict, http_headers, data)
|
|
|
|
def download_and_parse_fragment(url, frag_index, request_data):
|
|
count = 0
|
|
while count <= fragment_retries:
|
|
try:
|
|
success, raw_fragment = dl_fragment(url, request_data, {'content-type': 'application/json'})
|
|
if not success:
|
|
return False, None, None
|
|
try:
|
|
data = ie._extract_yt_initial_data(video_id, raw_fragment.decode('utf-8', 'replace'))
|
|
except RegexNotFoundError:
|
|
data = None
|
|
if not data:
|
|
data = json.loads(raw_fragment)
|
|
live_chat_continuation = try_get(
|
|
data,
|
|
lambda x: x['continuationContents']['liveChatContinuation'], dict) or {}
|
|
offset = continuation_id = None
|
|
processed_fragment = bytearray()
|
|
for action in live_chat_continuation.get('actions', []):
|
|
if 'replayChatItemAction' in action:
|
|
replay_chat_item_action = action['replayChatItemAction']
|
|
offset = int(replay_chat_item_action['videoOffsetTimeMsec'])
|
|
processed_fragment.extend(
|
|
json.dumps(action, ensure_ascii=False).encode('utf-8') + b'\n')
|
|
if offset is not None:
|
|
continuation_id = try_get(
|
|
live_chat_continuation,
|
|
lambda x: x['continuations'][0]['liveChatReplayContinuationData']['continuation'])
|
|
self._append_fragment(ctx, processed_fragment)
|
|
|
|
return True, continuation_id, offset
|
|
except compat_urllib_error.HTTPError as err:
|
|
count += 1
|
|
if count <= fragment_retries:
|
|
self.report_retry_fragment(err, frag_index, count, fragment_retries)
|
|
if count > fragment_retries:
|
|
self.report_error('giving up after %s fragment retries' % fragment_retries)
|
|
return False, None, None
|
|
|
|
self._prepare_and_start_frag_download(ctx)
|
|
|
|
success, raw_fragment = dl_fragment(
|
|
'https://www.youtube.com/watch?v={}'.format(video_id))
|
|
if not success:
|
|
return False
|
|
try:
|
|
data = ie._extract_yt_initial_data(video_id, raw_fragment.decode('utf-8', 'replace'))
|
|
except RegexNotFoundError:
|
|
return False
|
|
continuation_id = try_get(
|
|
data,
|
|
lambda x: x['contents']['twoColumnWatchNextResults']['conversationBar']['liveChatRenderer']['continuations'][0]['reloadContinuationData']['continuation'])
|
|
# no data yet but required to call _append_fragment
|
|
self._append_fragment(ctx, b'')
|
|
|
|
ytcfg = ie._extract_ytcfg(video_id, raw_fragment.decode('utf-8', 'replace'))
|
|
|
|
if not ytcfg:
|
|
return False
|
|
api_key = try_get(ytcfg, lambda x: x['INNERTUBE_API_KEY'])
|
|
innertube_context = try_get(ytcfg, lambda x: x['INNERTUBE_CONTEXT'])
|
|
if not api_key or not innertube_context:
|
|
return False
|
|
url = 'https://www.youtube.com/youtubei/v1/live_chat/get_live_chat_replay?key=' + api_key
|
|
|
|
frag_index = offset = 0
|
|
while continuation_id is not None:
|
|
frag_index += 1
|
|
request_data = {
|
|
'context': innertube_context,
|
|
'continuation': continuation_id,
|
|
}
|
|
if frag_index > 1:
|
|
request_data['currentPlayerState'] = {'playerOffsetMs': str(max(offset - 5000, 0))}
|
|
success, continuation_id, offset = download_and_parse_fragment(
|
|
url, frag_index, json.dumps(request_data, ensure_ascii=False).encode('utf-8') + b'\n')
|
|
if not success:
|
|
return False
|
|
if test:
|
|
break
|
|
|
|
self._finish_frag_download(ctx)
|
|
return True
|