1
0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2025-10-31 14:45:14 +00:00

[cleanup] Add more ruff rules (#10149)

Authored by: seproDev

Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com>
Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
This commit is contained in:
sepro
2024-06-12 01:09:58 +02:00
committed by GitHub
parent db50f19d76
commit add96eb9f8
915 changed files with 7027 additions and 7246 deletions

View File

@@ -59,14 +59,14 @@ class LinkedInLearningBaseIE(LinkedInBaseIE):
if video_slug:
query.update({
'videoSlug': video_slug,
'resolution': '_%s' % resolution,
'resolution': f'_{resolution}',
})
sub = ' %dp' % resolution
api_url = 'https://www.linkedin.com/learning-api/detailedCourses'
if not self._get_cookies(api_url).get('JSESSIONID'):
self.raise_login_required()
return self._download_json(
api_url, video_slug, 'Downloading%s JSON metadata' % sub, headers={
api_url, video_slug, f'Downloading{sub} JSON metadata', headers={
'Csrf-Token': self._get_cookies(api_url)['JSESSIONID'].value,
}, query=query)['elements'][0]
@@ -78,7 +78,7 @@ class LinkedInLearningBaseIE(LinkedInBaseIE):
return mobj.group(1)
def _get_video_id(self, video_data, course_slug, video_slug):
return self._get_urn_id(video_data) or '%s/%s' % (course_slug, video_slug)
return self._get_urn_id(video_data) or f'{course_slug}/{video_slug}'
class LinkedInIE(LinkedInBaseIE):
@@ -92,7 +92,7 @@ class LinkedInIE(LinkedInBaseIE):
'description': 'md5:2998a31f6f479376dd62831f53a80f71',
'uploader': 'Mishal K.',
'thumbnail': 're:^https?://media.licdn.com/dms/image/.*$',
'like_count': int
'like_count': int,
},
}, {
'url': 'https://www.linkedin.com/posts/the-mathworks_2_what-is-mathworks-cloud-center-activity-7151241570371948544-4Gu7',
@@ -104,7 +104,7 @@ class LinkedInIE(LinkedInBaseIE):
'uploader': 'MathWorks',
'thumbnail': 're:^https?://media.licdn.com/dms/image/.*$',
'like_count': int,
'subtitles': 'mincount:1'
'subtitles': 'mincount:1',
},
}]
@@ -159,9 +159,10 @@ class LinkedInLearningIE(LinkedInLearningBaseIE):
for line, (line_dict, next_dict) in enumerate(itertools.zip_longest(transcript_lines, transcript_lines[1:])):
start_time, caption = line_dict['transcriptStartAt'] / 1000, line_dict['caption']
end_time = next_dict['transcriptStartAt'] / 1000 if next_dict else duration or start_time + 1
srt_data += '%d\n%s --> %s\n%s\n\n' % (line + 1, srt_subtitles_timecode(start_time),
srt_subtitles_timecode(end_time),
caption)
srt_data += (
f'{line + 1}\n'
f'{srt_subtitles_timecode(start_time)} --> {srt_subtitles_timecode(end_time)}\n'
f'{caption}\n\n')
return srt_data
def _real_extract(self, url):
@@ -176,7 +177,7 @@ class LinkedInLearningIE(LinkedInLearningBaseIE):
progressive_url = video_url_data.get('progressiveUrl')
if progressive_url:
formats.append({
'format_id': 'progressive-%dp' % height,
'format_id': f'progressive-{height}p',
'url': progressive_url,
'ext': 'mp4',
'height': height,
@@ -208,7 +209,7 @@ class LinkedInLearningIE(LinkedInLearningBaseIE):
if transcript_lines:
subtitles['en'] = [{
'ext': 'srt',
'data': self.json2srt(transcript_lines, duration)
'data': self.json2srt(transcript_lines, duration),
}]
return {
@@ -222,7 +223,7 @@ class LinkedInLearningIE(LinkedInLearningBaseIE):
# It seems like this would be correctly handled by default
# However, unless someone can confirm this, the old
# behaviour is being kept as-is
'_format_sort_fields': ('res', 'source_preference')
'_format_sort_fields': ('res', 'source_preference'),
}
@@ -241,7 +242,7 @@ class LinkedInLearningCourseIE(LinkedInLearningBaseIE):
@classmethod
def suitable(cls, url):
return False if LinkedInLearningIE.suitable(url) else super(LinkedInLearningCourseIE, cls).suitable(url)
return False if LinkedInLearningIE.suitable(url) else super().suitable(url)
def _real_extract(self, url):
course_slug = self._match_id(url)
@@ -259,7 +260,7 @@ class LinkedInLearningCourseIE(LinkedInLearningBaseIE):
'_type': 'url_transparent',
'id': self._get_video_id(video, course_slug, video_slug),
'title': video.get('title'),
'url': 'https://www.linkedin.com/learning/%s/%s' % (course_slug, video_slug),
'url': f'https://www.linkedin.com/learning/{course_slug}/{video_slug}',
'chapter': chapter_title,
'chapter_number': chapter_number,
'chapter_id': chapter_id,