mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-23 17:02:30 +00:00
add96eb9f8
Authored by: seproDev Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com> Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
141 lines
5.6 KiB
Python
141 lines
5.6 KiB
Python
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
clean_html,
|
|
int_or_none,
|
|
parse_codecs,
|
|
parse_duration,
|
|
str_to_int,
|
|
unified_timestamp,
|
|
)
|
|
|
|
|
|
class GabTVIE(InfoExtractor):
|
|
_VALID_URL = r'https?://tv\.gab\.com/channel/[^/]+/view/(?P<id>[a-z0-9-]+)'
|
|
_TESTS = [{
|
|
'url': 'https://tv.gab.com/channel/wurzelroot/view/why-was-america-in-afghanistan-61217eacea5665de450d0488',
|
|
'info_dict': {
|
|
'id': '61217eacea5665de450d0488',
|
|
'ext': 'mp4',
|
|
'title': 'WHY WAS AMERICA IN AFGHANISTAN - AMERICA FIRST AGAINST AMERICAN OLIGARCHY',
|
|
'uploader': 'Wurzelroot',
|
|
'uploader_id': '608fb0a85738fd1974984f7d',
|
|
'thumbnail': 'https://tv.gab.com/image/61217eacea5665de450d0488',
|
|
},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url).split('-')[-1]
|
|
webpage = self._download_webpage(url, video_id)
|
|
channel_id = self._search_regex(r'data-channel-id=\"(?P<channel_id>[^\"]+)', webpage, 'channel_id')
|
|
channel_name = self._search_regex(r'data-channel-name=\"(?P<channel_id>[^\"]+)', webpage, 'channel_name')
|
|
title = self._search_regex(r'data-episode-title=\"(?P<channel_id>[^\"]+)', webpage, 'title')
|
|
view_key = self._search_regex(r'data-view-key=\"(?P<channel_id>[^\"]+)', webpage, 'view_key')
|
|
description = clean_html(
|
|
self._html_search_regex(self._meta_regex('description'), webpage, 'description', group='content')) or None
|
|
available_resolutions = re.findall(
|
|
rf'<a\ data-episode-id=\"{video_id}\"\ data-resolution=\"(?P<resolution>[^\"]+)', webpage)
|
|
|
|
formats = []
|
|
for resolution in available_resolutions:
|
|
frmt = {
|
|
'url': f'https://tv.gab.com/media/{video_id}?viewKey={view_key}&r={resolution}',
|
|
'format_id': resolution,
|
|
'vcodec': 'h264',
|
|
'acodec': 'aac',
|
|
'ext': 'mp4',
|
|
}
|
|
if 'audio-' in resolution:
|
|
frmt['abr'] = str_to_int(resolution.replace('audio-', ''))
|
|
frmt['height'] = 144
|
|
frmt['quality'] = -10
|
|
else:
|
|
frmt['height'] = str_to_int(resolution.replace('p', ''))
|
|
formats.append(frmt)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'formats': formats,
|
|
'description': description,
|
|
'uploader': channel_name,
|
|
'uploader_id': channel_id,
|
|
'thumbnail': f'https://tv.gab.com/image/{video_id}',
|
|
}
|
|
|
|
|
|
class GabIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?gab\.com/[^/]+/posts/(?P<id>\d+)'
|
|
_TESTS = [{
|
|
'url': 'https://gab.com/SomeBitchIKnow/posts/107163961867310434',
|
|
'md5': '8ca34fb00f1e1033b5c5988d79ec531d',
|
|
'info_dict': {
|
|
'id': '107163961867310434-0',
|
|
'ext': 'mp4',
|
|
'title': 'L on Gab',
|
|
'uploader_id': '946600',
|
|
'uploader': 'SomeBitchIKnow',
|
|
'description': 'md5:204055fafd5e1a519f5d6db953567ca3',
|
|
'timestamp': 1635192289,
|
|
'upload_date': '20211025',
|
|
},
|
|
}, {
|
|
'url': 'https://gab.com/TheLonelyProud/posts/107045884469287653',
|
|
'md5': 'f9cefcfdff6418e392611a828d47839d',
|
|
'info_dict': {
|
|
'id': '107045884469287653-0',
|
|
'ext': 'mp4',
|
|
'title': 'Jody Sadowski on Gab',
|
|
'uploader_id': '1390705',
|
|
'timestamp': 1633390571,
|
|
'upload_date': '20211004',
|
|
'uploader': 'TheLonelyProud',
|
|
},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
post_id = self._match_id(url)
|
|
json_data = self._download_json(f'https://gab.com/api/v1/statuses/{post_id}', post_id)
|
|
|
|
entries = []
|
|
for idx, media in enumerate(json_data['media_attachments']):
|
|
if media.get('type') not in ('video', 'gifv'):
|
|
continue
|
|
metadata = media['meta']
|
|
format_metadata = {
|
|
'acodec': parse_codecs(metadata.get('audio_encode')).get('acodec'),
|
|
'asr': int_or_none((metadata.get('audio_bitrate') or '').split(' ')[0]),
|
|
'fps': metadata.get('fps'),
|
|
}
|
|
|
|
formats = [{
|
|
'url': url,
|
|
'width': f.get('width'),
|
|
'height': f.get('height'),
|
|
'tbr': int_or_none(f.get('bitrate'), scale=1000),
|
|
**format_metadata,
|
|
} for url, f in ((media.get('url'), metadata.get('original') or {}),
|
|
(media.get('source_mp4'), metadata.get('playable') or {})) if url]
|
|
|
|
author = json_data.get('account') or {}
|
|
entries.append({
|
|
'id': f'{post_id}-{idx}',
|
|
'title': f'{json_data["account"]["display_name"]} on Gab',
|
|
'timestamp': unified_timestamp(json_data.get('created_at')),
|
|
'formats': formats,
|
|
'description': clean_html(json_data.get('content')),
|
|
'duration': metadata.get('duration') or parse_duration(metadata.get('length')),
|
|
'like_count': json_data.get('favourites_count'),
|
|
'comment_count': json_data.get('replies_count'),
|
|
'repost_count': json_data.get('reblogs_count'),
|
|
'uploader': author.get('username'),
|
|
'uploader_id': author.get('id'),
|
|
'uploader_url': author.get('url'),
|
|
})
|
|
|
|
if len(entries) > 1:
|
|
return self.playlist_result(entries, post_id)
|
|
|
|
return entries[0]
|