mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-16 14:25:47 +00:00
152 lines
5.1 KiB
Python
152 lines
5.1 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
import itertools
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
determine_ext,
|
|
int_or_none,
|
|
unified_timestamp,
|
|
)
|
|
|
|
|
|
class VineIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?vine\.co/(?:v|oembed)/(?P<id>\w+)'
|
|
_TESTS = [{
|
|
'url': 'https://vine.co/v/b9KOOWX7HUx',
|
|
'md5': '2f36fed6235b16da96ce9b4dc890940d',
|
|
'info_dict': {
|
|
'id': 'b9KOOWX7HUx',
|
|
'ext': 'mp4',
|
|
'title': 'Chicken.',
|
|
'alt_title': 'Vine by Jack',
|
|
'timestamp': 1368997951,
|
|
'upload_date': '20130519',
|
|
'uploader': 'Jack',
|
|
'uploader_id': '76',
|
|
'view_count': int,
|
|
'like_count': int,
|
|
'comment_count': int,
|
|
'repost_count': int,
|
|
},
|
|
}, {
|
|
'url': 'https://vine.co/v/e192BnZnZ9V',
|
|
'info_dict': {
|
|
'id': 'e192BnZnZ9V',
|
|
'ext': 'mp4',
|
|
'title': 'ยิ้ม~ เขิน~ อาย~ น่าร้ากอ้ะ >//< @n_whitewo @orlameena #lovesicktheseries #lovesickseason2',
|
|
'alt_title': 'Vine by Pimry_zaa',
|
|
'timestamp': 1436057405,
|
|
'upload_date': '20150705',
|
|
'uploader': 'Pimry_zaa',
|
|
'uploader_id': '1135760698325307392',
|
|
'view_count': int,
|
|
'like_count': int,
|
|
'comment_count': int,
|
|
'repost_count': int,
|
|
},
|
|
'params': {
|
|
'skip_download': True,
|
|
},
|
|
}, {
|
|
'url': 'https://vine.co/v/MYxVapFvz2z',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://vine.co/v/bxVjBbZlPUH',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://vine.co/oembed/MYxVapFvz2z.json',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
data = self._download_json(
|
|
'https://archive.vine.co/posts/%s.json' % video_id, video_id)
|
|
|
|
def video_url(kind):
|
|
for url_suffix in ('Url', 'URL'):
|
|
format_url = data.get('video%s%s' % (kind, url_suffix))
|
|
if format_url:
|
|
return format_url
|
|
|
|
formats = []
|
|
for quality, format_id in enumerate(('low', '', 'dash')):
|
|
format_url = video_url(format_id.capitalize())
|
|
if not format_url:
|
|
continue
|
|
# DASH link returns plain mp4
|
|
if format_id == 'dash' and determine_ext(format_url) == 'mpd':
|
|
formats.extend(self._extract_mpd_formats(
|
|
format_url, video_id, mpd_id='dash', fatal=False))
|
|
else:
|
|
formats.append({
|
|
'url': format_url,
|
|
'format_id': format_id or 'standard',
|
|
'quality': quality,
|
|
})
|
|
self._sort_formats(formats)
|
|
|
|
username = data.get('username')
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': data.get('description'),
|
|
'alt_title': 'Vine by %s' % username if username else None,
|
|
'thumbnail': data.get('thumbnailUrl'),
|
|
'timestamp': unified_timestamp(data.get('created')),
|
|
'uploader': username,
|
|
'uploader_id': data.get('userIdStr'),
|
|
'view_count': int_or_none(data.get('loops')),
|
|
'like_count': int_or_none(data.get('likes')),
|
|
'comment_count': int_or_none(data.get('comments')),
|
|
'repost_count': int_or_none(data.get('reposts')),
|
|
'formats': formats,
|
|
}
|
|
|
|
|
|
class VineUserIE(InfoExtractor):
|
|
IE_NAME = 'vine:user'
|
|
_VALID_URL = r'(?:https?://)?vine\.co/(?P<u>u/)?(?P<user>[^/]+)/?(\?.*)?$'
|
|
_VINE_BASE_URL = 'https://vine.co/'
|
|
_TESTS = [
|
|
{
|
|
'url': 'https://vine.co/Visa',
|
|
'info_dict': {
|
|
'id': 'Visa',
|
|
},
|
|
'playlist_mincount': 46,
|
|
},
|
|
{
|
|
'url': 'https://vine.co/u/941705360593584128',
|
|
'only_matching': True,
|
|
},
|
|
]
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
user = mobj.group('user')
|
|
u = mobj.group('u')
|
|
|
|
profile_url = '%sapi/users/profiles/%s%s' % (
|
|
self._VINE_BASE_URL, 'vanity/' if not u else '', user)
|
|
profile_data = self._download_json(
|
|
profile_url, user, note='Downloading user profile data')
|
|
|
|
user_id = profile_data['data']['userId']
|
|
timeline_data = []
|
|
for pagenum in itertools.count(1):
|
|
timeline_url = '%sapi/timelines/users/%s?page=%s&size=100' % (
|
|
self._VINE_BASE_URL, user_id, pagenum)
|
|
timeline_page = self._download_json(
|
|
timeline_url, user, note='Downloading page %d' % pagenum)
|
|
timeline_data.extend(timeline_page['data']['records'])
|
|
if timeline_page['data']['nextPage'] is None:
|
|
break
|
|
|
|
entries = [
|
|
self.url_result(e['permalinkUrl'], 'Vine') for e in timeline_data]
|
|
return self.playlist_result(entries, user)
|