# coding: utf-8 from __future__ import unicode_literals import re from .common import InfoExtractor from ..utils import ( ExtractorError, float_or_none, int_or_none, ) class StreamableIE(InfoExtractor): _VALID_URL = r'https?://streamable\.com/(?:[es]/)?(?P<id>\w+)' _TESTS = [ { 'url': 'https://streamable.com/dnd1', 'md5': '3e3bc5ca088b48c2d436529b64397fef', 'info_dict': { 'id': 'dnd1', 'ext': 'mp4', 'title': 'Mikel Oiarzabal scores to make it 0-3 for La Real against Espanyol', 'thumbnail': r're:https?://.*\.jpg$', 'uploader': 'teabaker', 'timestamp': 1454964157.35115, 'upload_date': '20160208', 'duration': 61.516, 'view_count': int, } }, # older video without bitrate, width/height, etc. info { 'url': 'https://streamable.com/moo', 'md5': '2cf6923639b87fba3279ad0df3a64e73', 'info_dict': { 'id': 'moo', 'ext': 'mp4', 'title': '"Please don\'t eat me!"', 'thumbnail': r're:https?://.*\.jpg$', 'timestamp': 1426115495, 'upload_date': '20150311', 'duration': 12, 'view_count': int, } }, { 'url': 'https://streamable.com/e/dnd1', 'only_matching': True, }, { 'url': 'https://streamable.com/s/okkqk/drxjds', 'only_matching': True, } ] @staticmethod def _extract_url(webpage): mobj = re.search( r'<iframe[^>]+src=(?P<q1>[\'"])(?P<src>(?:https?:)?//streamable\.com/(?:(?!\1).+))(?P=q1)', webpage) if mobj: return mobj.group('src') def _real_extract(self, url): video_id = self._match_id(url) # Note: Using the ajax API, as the public Streamable API doesn't seem # to return video info like the title properly sometimes, and doesn't # include info like the video duration video = self._download_json( 'https://ajax.streamable.com/videos/%s' % video_id, video_id) # Format IDs: # 0 The video is being uploaded # 1 The video is being processed # 2 The video has at least one file ready # 3 The video is unavailable due to an error status = video.get('status') if status != 2: raise ExtractorError( 'This video is currently unavailable. It may still be uploading or processing.', expected=True) title = video.get('reddit_title') or video['title'] formats = [] for key, info in video['files'].items(): if not info.get('url'): continue formats.append({ 'format_id': key, 'url': self._proto_relative_url(info['url']), 'width': int_or_none(info.get('width')), 'height': int_or_none(info.get('height')), 'filesize': int_or_none(info.get('size')), 'fps': int_or_none(info.get('framerate')), 'vbr': float_or_none(info.get('bitrate'), 1000) }) self._sort_formats(formats) return { 'id': video_id, 'title': title, 'description': video.get('description'), 'thumbnail': self._proto_relative_url(video.get('thumbnail_url')), 'uploader': video.get('owner', {}).get('user_name'), 'timestamp': float_or_none(video.get('date_added')), 'duration': float_or_none(video.get('duration')), 'view_count': int_or_none(video.get('plays')), 'formats': formats }