1
1
mirror of https://github.com/ytdl-org/youtube-dl synced 2024-11-24 13:36:51 +01:00
youtube-dl/youtube_dl/extractor/screencastomatic.py

50 lines
1.7 KiB
Python
Raw Normal View History

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..compat import compat_urlparse
from ..utils import (
ExtractorError,
js_to_json,
)
class ScreencastOMaticIE(InfoExtractor):
_VALID_URL = r'https?://screencast-o-matic\.com/watch/(?P<id>[0-9a-zA-Z]+)'
_TEST = {
'url': 'http://screencast-o-matic.com/watch/c2lD3BeOPl',
'md5': '483583cb80d92588f15ccbedd90f0c18',
'info_dict': {
'id': 'c2lD3BeOPl',
'ext': 'mp4',
'title': 'Welcome to 3-4 Philosophy @ DECV!',
'thumbnail': 're:^https?://.*\.jpg$',
'description': 'as the title says! also: some general info re 1) VCE philosophy and 2) distance learning.',
}
}
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
setup_js = self._search_regex(
r"(?s)jwplayer\('mp4Player'\).setup\((\{.*?\})\);",
webpage, 'setup code')
data = self._parse_json(setup_js, video_id, transform_source=js_to_json)
try:
video_data = next(
m for m in data['modes'] if m.get('type') == 'html5')
except StopIteration:
raise ExtractorError('Could not find any video entries!')
video_url = compat_urlparse.urljoin(url, video_data['config']['file'])
thumbnail = data.get('image')
return {
'id': video_id,
'title': self._og_search_title(webpage),
'description': self._og_search_description(webpage),
'url': video_url,
'ext': 'mp4',
'thumbnail': thumbnail,
}