summaryrefslogtreecommitdiffstats
path: root/yt_dlp/extractor/flickr.py
diff options
context:
space:
mode:
Diffstat (limited to 'yt_dlp/extractor/flickr.py')
-rw-r--r--yt_dlp/extractor/flickr.py114
1 files changed, 114 insertions, 0 deletions
diff --git a/yt_dlp/extractor/flickr.py b/yt_dlp/extractor/flickr.py
new file mode 100644
index 0000000..89a40d7
--- /dev/null
+++ b/yt_dlp/extractor/flickr.py
@@ -0,0 +1,114 @@
+from .common import InfoExtractor
+from ..compat import (
+ compat_str,
+ compat_urllib_parse_urlencode,
+)
+from ..utils import (
+ ExtractorError,
+ format_field,
+ int_or_none,
+ qualities,
+)
+
+
+class FlickrIE(InfoExtractor):
+ _VALID_URL = r'https?://(?:www\.|secure\.)?flickr\.com/photos/[\w\-_@]+/(?P<id>\d+)'
+ _TEST = {
+ 'url': 'http://www.flickr.com/photos/forestwander-nature-pictures/5645318632/in/photostream/',
+ 'md5': '164fe3fa6c22e18d448d4d5af2330f31',
+ 'info_dict': {
+ 'id': '5645318632',
+ 'ext': 'mpg',
+ 'description': 'Waterfalls in the Springtime at Dark Hollow Waterfalls. These are located just off of Skyline Drive in Virginia. They are only about 6/10 of a mile hike but it is a pretty steep hill and a good climb back up.',
+ 'title': 'Dark Hollow Waterfalls',
+ 'duration': 19,
+ 'timestamp': 1303528740,
+ 'upload_date': '20110423',
+ 'uploader_id': '10922353@N03',
+ 'uploader': 'Forest Wander',
+ 'uploader_url': 'https://www.flickr.com/photos/forestwander-nature-pictures/',
+ 'comment_count': int,
+ 'view_count': int,
+ 'tags': list,
+ 'license': 'Attribution-ShareAlike',
+ }
+ }
+ _API_BASE_URL = 'https://api.flickr.com/services/rest?'
+ # https://help.yahoo.com/kb/flickr/SLN25525.html
+ _LICENSES = {
+ '0': 'All Rights Reserved',
+ '1': 'Attribution-NonCommercial-ShareAlike',
+ '2': 'Attribution-NonCommercial',
+ '3': 'Attribution-NonCommercial-NoDerivs',
+ '4': 'Attribution',
+ '5': 'Attribution-ShareAlike',
+ '6': 'Attribution-NoDerivs',
+ '7': 'No known copyright restrictions',
+ '8': 'United States government work',
+ '9': 'Public Domain Dedication (CC0)',
+ '10': 'Public Domain Work',
+ }
+
+ def _call_api(self, method, video_id, api_key, note, secret=None):
+ query = {
+ 'photo_id': video_id,
+ 'method': 'flickr.%s' % method,
+ 'api_key': api_key,
+ 'format': 'json',
+ 'nojsoncallback': 1,
+ }
+ if secret:
+ query['secret'] = secret
+ data = self._download_json(self._API_BASE_URL + compat_urllib_parse_urlencode(query), video_id, note)
+ if data['stat'] != 'ok':
+ raise ExtractorError(data['message'])
+ return data
+
+ def _real_extract(self, url):
+ video_id = self._match_id(url)
+
+ api_key = self._download_json(
+ 'https://www.flickr.com/hermes_error_beacon.gne', video_id,
+ 'Downloading api key')['site_key']
+
+ video_info = self._call_api(
+ 'photos.getInfo', video_id, api_key, 'Downloading video info')['photo']
+ if video_info['media'] == 'video':
+ streams = self._call_api(
+ 'video.getStreamInfo', video_id, api_key,
+ 'Downloading streams info', video_info['secret'])['streams']
+
+ preference = qualities(
+ ['288p', 'iphone_wifi', '100', '300', '700', '360p', 'appletv', '720p', '1080p', 'orig'])
+
+ formats = []
+ for stream in streams['stream']:
+ stream_type = compat_str(stream.get('type'))
+ formats.append({
+ 'format_id': stream_type,
+ 'url': stream['_content'],
+ 'quality': preference(stream_type),
+ })
+
+ owner = video_info.get('owner', {})
+ uploader_id = owner.get('nsid')
+ uploader_path = owner.get('path_alias') or uploader_id
+ uploader_url = format_field(uploader_path, None, 'https://www.flickr.com/photos/%s/')
+
+ return {
+ 'id': video_id,
+ 'title': video_info['title']['_content'],
+ 'description': video_info.get('description', {}).get('_content'),
+ 'formats': formats,
+ 'timestamp': int_or_none(video_info.get('dateuploaded')),
+ 'duration': int_or_none(video_info.get('video', {}).get('duration')),
+ 'uploader_id': uploader_id,
+ 'uploader': owner.get('realname'),
+ 'uploader_url': uploader_url,
+ 'comment_count': int_or_none(video_info.get('comments', {}).get('_content')),
+ 'view_count': int_or_none(video_info.get('views')),
+ 'tags': [tag.get('_content') for tag in video_info.get('tags', {}).get('tag', [])],
+ 'license': self._LICENSES.get(video_info.get('license')),
+ }
+ else:
+ raise ExtractorError('not a video', expected=True)