logo

youtube-dl

[mirror] Download/Watch videos from video hosters
commit: 5e36b63486794750aca0ee6b9b83f27abf6332dc
parent 9249c50c18c40dab8dd78a93003d0daae569e27d
Author: Remita Amine <remitamine@gmail.com>
Date:   Sat,  2 Nov 2019 13:25:39 +0100

[iconosquare] remove extractor

Diffstat:

Myoutube_dl/extractor/extractors.py1-
Dyoutube_dl/extractor/iconosquare.py85-------------------------------------------------------------------------------
2 files changed, 0 insertions(+), 86 deletions(-)

diff --git a/youtube_dl/extractor/extractors.py b/youtube_dl/extractor/extractors.py @@ -455,7 +455,6 @@ from .hungama import ( HungamaSongIE, ) from .hypem import HypemIE -from .iconosquare import IconosquareIE from .ign import ( IGNIE, OneUPIE, diff --git a/youtube_dl/extractor/iconosquare.py b/youtube_dl/extractor/iconosquare.py @@ -1,85 +0,0 @@ -from __future__ import unicode_literals - -from .common import InfoExtractor -from ..utils import ( - int_or_none, - get_element_by_id, - remove_end, -) - - -class IconosquareIE(InfoExtractor): - _VALID_URL = r'https?://(?:www\.)?(?:iconosquare\.com|statigr\.am)/p/(?P<id>[^/]+)' - _TEST = { - 'url': 'http://statigr.am/p/522207370455279102_24101272', - 'md5': '6eb93b882a3ded7c378ee1d6884b1814', - 'info_dict': { - 'id': '522207370455279102_24101272', - 'ext': 'mp4', - 'title': 'Instagram photo by @aguynamedpatrick (Patrick Janelle)', - 'description': 'md5:644406a9ec27457ed7aa7a9ebcd4ce3d', - 'timestamp': 1376471991, - 'upload_date': '20130814', - 'uploader': 'aguynamedpatrick', - 'uploader_id': '24101272', - 'comment_count': int, - 'like_count': int, - }, - } - - def _real_extract(self, url): - video_id = self._match_id(url) - - webpage = self._download_webpage(url, video_id) - - media = self._parse_json( - get_element_by_id('mediaJson', webpage), - video_id) - - formats = [{ - 'url': f['url'], - 'format_id': format_id, - 'width': int_or_none(f.get('width')), - 'height': int_or_none(f.get('height')) - } for format_id, f in media['videos'].items()] - self._sort_formats(formats) - - title = remove_end(self._og_search_title(webpage), ' - via Iconosquare') - - timestamp = int_or_none(media.get('created_time') or media.get('caption', {}).get('created_time')) - description = media.get('caption', {}).get('text') - - uploader = media.get('user', {}).get('username') - uploader_id = media.get('user', {}).get('id') - - comment_count = int_or_none(media.get('comments', {}).get('count')) - like_count = int_or_none(media.get('likes', {}).get('count')) - - thumbnails = [{ - 'url': t['url'], - 'id': thumbnail_id, - 'width': int_or_none(t.get('width')), - 'height': int_or_none(t.get('height')) - } for thumbnail_id, t in media.get('images', {}).items()] - - comments = [{ - 'id': comment.get('id'), - 'text': comment['text'], - 'timestamp': int_or_none(comment.get('created_time')), - 'author': comment.get('from', {}).get('full_name'), - 'author_id': comment.get('from', {}).get('username'), - } for comment in media.get('comments', {}).get('data', []) if 'text' in comment] - - return { - 'id': video_id, - 'title': title, - 'description': description, - 'thumbnails': thumbnails, - 'timestamp': timestamp, - 'uploader': uploader, - 'uploader_id': uploader_id, - 'comment_count': comment_count, - 'like_count': like_count, - 'formats': formats, - 'comments': comments, - }