+++ /dev/null
-# coding: utf-8
-from __future__ import unicode_literals
-
-from .common import InfoExtractor
-from ..compat import compat_str
-from ..utils import (
- int_or_none,
- float_or_none,
- try_get,
- unified_timestamp,
-)
-
-
-class FlipagramIE(InfoExtractor):
- _VALID_URL = r'https?://(?:www\.)?flipagram\.com/f/(?P<id>[^/?#&]+)'
- _TEST = {
- 'url': 'https://flipagram.com/f/nyvTSJMKId',
- 'md5': '888dcf08b7ea671381f00fab74692755',
- 'info_dict': {
- 'id': 'nyvTSJMKId',
- 'ext': 'mp4',
- 'title': 'Flipagram by sjuria101 featuring Midnight Memories by One Direction',
- 'description': 'md5:d55e32edc55261cae96a41fa85ff630e',
- 'duration': 35.571,
- 'timestamp': 1461244995,
- 'upload_date': '20160421',
- 'uploader': 'kitty juria',
- 'uploader_id': 'sjuria101',
- 'creator': 'kitty juria',
- 'view_count': int,
- 'like_count': int,
- 'repost_count': int,
- 'comment_count': int,
- 'comments': list,
- 'formats': 'mincount:2',
- },
- }
-
- def _real_extract(self, url):
- video_id = self._match_id(url)
- webpage = self._download_webpage(url, video_id)
-
- video_data = self._parse_json(
- self._search_regex(
- r'window\.reactH2O\s*=\s*({.+});', webpage, 'video data'),
- video_id)
-
- flipagram = video_data['flipagram']
- video = flipagram['video']
-
- json_ld = self._search_json_ld(webpage, video_id, default={})
- title = json_ld.get('title') or flipagram['captionText']
- description = json_ld.get('description') or flipagram.get('captionText')
-
- formats = [{
- 'url': video['url'],
- 'width': int_or_none(video.get('width')),
- 'height': int_or_none(video.get('height')),
- 'filesize': int_or_none(video_data.get('size')),
- }]
-
- preview_url = try_get(
- flipagram, lambda x: x['music']['track']['previewUrl'], compat_str)
- if preview_url:
- formats.append({
- 'url': preview_url,
- 'ext': 'm4a',
- 'vcodec': 'none',
- })
-
- self._sort_formats(formats)
-
- counts = flipagram.get('counts', {})
- user = flipagram.get('user', {})
- video_data = flipagram.get('video', {})
-
- thumbnails = [{
- 'url': self._proto_relative_url(cover['url']),
- 'width': int_or_none(cover.get('width')),
- 'height': int_or_none(cover.get('height')),
- 'filesize': int_or_none(cover.get('size')),
- } for cover in flipagram.get('covers', []) if cover.get('url')]
-
- # Note that this only retrieves comments that are initially loaded.
- # For videos with large amounts of comments, most won't be retrieved.
- comments = []
- for comment in video_data.get('comments', {}).get(video_id, {}).get('items', []):
- text = comment.get('comment')
- if not text or not isinstance(text, list):
- continue
- comments.append({
- 'author': comment.get('user', {}).get('name'),
- 'author_id': comment.get('user', {}).get('username'),
- 'id': comment.get('id'),
- 'text': text[0],
- 'timestamp': unified_timestamp(comment.get('created')),
- })
-
- return {
- 'id': video_id,
- 'title': title,
- 'description': description,
- 'duration': float_or_none(flipagram.get('duration'), 1000),
- 'thumbnails': thumbnails,
- 'timestamp': unified_timestamp(flipagram.get('iso8601Created')),
- 'uploader': user.get('name'),
- 'uploader_id': user.get('username'),
- 'creator': user.get('name'),
- 'view_count': int_or_none(counts.get('plays')),
- 'like_count': int_or_none(counts.get('likes')),
- 'repost_count': int_or_none(counts.get('reflips')),
- 'comment_count': int_or_none(counts.get('comments')),
- 'comments': comments,
- 'formats': formats,
- }