mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-13 20:03:17 +00:00
cefecac12c
to avoid using same folder when using pip install for example
95 lines
3.6 KiB
Python
95 lines
3.6 KiB
Python
from __future__ import division, unicode_literals
|
|
|
|
import re
|
|
import json
|
|
|
|
from .fragment import FragmentFD
|
|
|
|
|
|
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)
|
|
|
|
test = self.params.get('test', False)
|
|
|
|
ctx = {
|
|
'filename': filename,
|
|
'live': True,
|
|
'total_frags': None,
|
|
}
|
|
|
|
def dl_fragment(url):
|
|
headers = info_dict.get('http_headers', {})
|
|
return self._download_fragment(ctx, url, info_dict, headers)
|
|
|
|
def parse_yt_initial_data(data):
|
|
window_patt = b'window\\["ytInitialData"\\]\\s*=\\s*(.*?)(?<=});'
|
|
var_patt = b'var\\s+ytInitialData\\s*=\\s*(.*?)(?<=});'
|
|
for patt in window_patt, var_patt:
|
|
try:
|
|
raw_json = re.search(patt, data).group(1)
|
|
return json.loads(raw_json)
|
|
except AttributeError:
|
|
continue
|
|
|
|
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
|
|
data = parse_yt_initial_data(raw_fragment)
|
|
continuation_id = data['contents']['twoColumnWatchNextResults']['conversationBar']['liveChatRenderer']['continuations'][0]['reloadContinuationData']['continuation']
|
|
# no data yet but required to call _append_fragment
|
|
self._append_fragment(ctx, b'')
|
|
|
|
first = True
|
|
offset = None
|
|
while continuation_id is not None:
|
|
data = None
|
|
if first:
|
|
url = 'https://www.youtube.com/live_chat_replay?continuation={}'.format(continuation_id)
|
|
success, raw_fragment = dl_fragment(url)
|
|
if not success:
|
|
return False
|
|
data = parse_yt_initial_data(raw_fragment)
|
|
else:
|
|
url = ('https://www.youtube.com/live_chat_replay/get_live_chat_replay'
|
|
+ '?continuation={}'.format(continuation_id)
|
|
+ '&playerOffsetMs={}'.format(offset - 5000)
|
|
+ '&hidden=false'
|
|
+ '&pbj=1')
|
|
success, raw_fragment = dl_fragment(url)
|
|
if not success:
|
|
return False
|
|
data = json.loads(raw_fragment)['response']
|
|
|
|
first = False
|
|
continuation_id = None
|
|
|
|
live_chat_continuation = data['continuationContents']['liveChatContinuation']
|
|
offset = None
|
|
processed_fragment = bytearray()
|
|
if 'actions' in live_chat_continuation:
|
|
for action in live_chat_continuation['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')
|
|
continuation_id = live_chat_continuation['continuations'][0]['liveChatReplayContinuationData']['continuation']
|
|
|
|
self._append_fragment(ctx, processed_fragment)
|
|
|
|
if test or offset is None:
|
|
break
|
|
|
|
self._finish_frag_download(ctx)
|
|
|
|
return True
|