- formatting

pull/8/head
Hugo Alves De Azevedo 4 years ago
parent 2adf91f94f
commit 39672cdc79

@ -29,27 +29,24 @@ class AluraIE(InfoExtractor):
'id': '60095',
'ext': 'mp4',
'title': 'Referências, ref-set e alter'
},
'skip': 'Requires alura account credentials',
},
'skip': 'Requires alura account credentials'},
{
# URL without video
'url': 'https://cursos.alura.com.br/course/clojure-mutabilidade-com-atoms-e-refs/task/60098',
'only_matching': True,
},
'only_matching': True},
{
'url': 'https://cursos.alura.com.br/course/fundamentos-market-digital/task/55219',
'only_matching': True,
}
'only_matching': True}
]
def _real_extract(self, url):
video_id = self._match_id(url)
course = self._search_regex(self._VALID_URL, url, 'post url', group='course_name')
video_url = self._VIDEO_URL % (course,video_id)
video_url = self._VIDEO_URL % (course, video_id)
video_dict = self._download_json(video_url, None, 'Searching for videos', expected_status=[404,500])
video_dict = self._download_json(video_url, None, 'Searching for videos', expected_status=[404, 500])
if video_dict:
webpage = self._download_webpage(url, video_id)
@ -152,9 +149,21 @@ class AluraCourseIE(AluraIE):
page_url = urljoin(url, path)
section_path = self._download_webpage(page_url, course_path)
for path_video in re.findall(r'<a\b(?=[^>]* class="[^"]*(?<=[" ])task-menu-nav-item-link-VIDEO[" ])(?=[^>]* href="([^"]*))', section_path):
chapter = clean_html(self._search_regex(r'<h3[^>]+class=(["\'])task-menu-section-title-text\1[^>]*>(?P<chapter>[^<]+)',section_path, 'chapter', group='chapter'))
chapter_number = int_or_none(self._search_regex(r'<span[^>]+class=(["\'])task-menu-section-title-number[^>]*>(.*?)<strong>(?P<chapter_number>[^<]+)</strong>',section_path, 'chapter number', group='chapter_number'))
chapter = clean_html(
self._search_regex(
r'<h3[^>]+class=(["\'])task-menu-section-title-text\1[^>]*>(?P<chapter>[^<]+)',
section_path,
'chapter',
group='chapter'))
chapter_number = int_or_none(
self._search_regex(
r'<span[^>]+class=(["\'])task-menu-section-title-number[^>]*>(.*?)<strong>(?P<chapter_number>[^<]+)</strong>',
section_path,
'chapter number',
group='chapter_number'))
video_url = urljoin(url, path_video)
entry = {
'_type': 'url_transparent',
'id': self._match_id(video_url),
@ -164,4 +173,4 @@ class AluraCourseIE(AluraIE):
'chapter_number': chapter_number
}
entries.append(entry)
return self.playlist_result(entries, course_path, course_title)
return self.playlist_result(entries, course_path, course_title)

Loading…
Cancel
Save