1
0
mirror of https://codeberg.org/polarisfm/youtube-dl synced 2024-11-22 16:44:32 +01:00

Fixed formatting according to flake8

This commit is contained in:
Andy Bottom 2018-10-21 14:09:30 -05:00
parent 2d68da7781
commit d76916f79e

View File

@ -6,6 +6,7 @@ from ..utils import (
determine_ext,
)
class RightNowMediaIE(InfoExtractor):
IE_NAME = 'rightnowmedia'
_VALID_URL = r'https?://(?:www\.)?rightnowmedia\.org/Content/(?P<id>[0-9]+)/(?:downloadAndEmbed)'
@ -21,16 +22,15 @@ class RightNowMediaIE(InfoExtractor):
def _real_extract(self, url):
# Video Id
video_id = self._match_id(url)
# Download
video_info_dicts = self._download_json(
'https://www.rightnowmedia.org/Content/%s/downloadAndEmbed'
% video_id, video_id)
# Get All The Formats
formats = []
for video_info in video_info_dicts['downloadLinks']:
video_url = video_info.get('src')
quality = 'high' if 'HD 1080p' in video_info["QualityName"] else 'low'
formats.append({
'url': video_info["Link"],
@ -38,15 +38,15 @@ class RightNowMediaIE(InfoExtractor):
'format_note': quality,
'format_id': '%s-%s' % (quality, determine_ext(video_info["Link"])),
})
# Get the Title
title = compat_urllib_parse_unquote(re.findall(
r'.*?filename=(.*).*(?:.mp4)',
formats[0]['url'])[0])
r'.*?filename=(.*).*(?:.mp4)',
formats[0]['url'])[0])
# Sort Formats
self._sort_formats(formats)
# Return
return {
'id': video_id,
@ -54,7 +54,7 @@ class RightNowMediaIE(InfoExtractor):
'formats': formats,
}
class RightNowMediaPlaylistIE(InfoExtractor):
IE_NAME = 'rightnowmedia:playlist'
_VALID_URL = r'https?://(?:www\.)?rightnowmedia\.org/Content/(?:Series|KidsShow)/(?P<id>[0-9]+)'
@ -64,7 +64,7 @@ class RightNowMediaPlaylistIE(InfoExtractor):
'id': '265320'
},
'playlist_count': 9,
},{
}, {
'url': 'https://www.rightnowmedia.org/Content/KidsShow/298875',
'info_dict': {
'id': '298875'
@ -77,18 +77,18 @@ class RightNowMediaPlaylistIE(InfoExtractor):
playlist_id = self._match_id(url)
# Download Webpage
webpage = self._download_webpage(url, playlist_id)
webpage = self._download_webpage(url, playlist_id)
# Find The Correct Table
all_buckets = re.findall(
r'(?s)<table [^"]*"..*? id="primarySeriesTable" [^"]*"..*? \B.*\B<div class="row rowStripMargin nomobile">',
webpage)
# Find All The Video Elements
all_video_elements = re.findall(
r'.*?data-detail-content-id="(.*)">.*',
all_buckets[0])
# Finalize All The URLs
entries = []
for video_element in all_video_elements: