mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-27 10:52:34 +00:00
130 lines
4.5 KiB
Python
130 lines
4.5 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import compat_str
|
|
from ..utils import (
|
|
determine_ext,
|
|
int_or_none,
|
|
try_get,
|
|
unified_timestamp,
|
|
url_or_none,
|
|
)
|
|
|
|
|
|
class EggheadCourseIE(InfoExtractor):
|
|
IE_DESC = 'egghead.io course'
|
|
IE_NAME = 'egghead:course'
|
|
_VALID_URL = r'https://egghead\.io/courses/(?P<id>[^/?#&]+)'
|
|
_TEST = {
|
|
'url': 'https://egghead.io/courses/professor-frisby-introduces-composable-functional-javascript',
|
|
'playlist_count': 29,
|
|
'info_dict': {
|
|
'id': '72',
|
|
'title': 'Professor Frisby Introduces Composable Functional JavaScript',
|
|
'description': 're:(?s)^This course teaches the ubiquitous.*You\'ll start composing functionality before you know it.$',
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
playlist_id = self._match_id(url)
|
|
|
|
lessons = self._download_json(
|
|
'https://egghead.io/api/v1/series/%s/lessons' % playlist_id,
|
|
playlist_id, 'Downloading course lessons JSON')
|
|
|
|
entries = []
|
|
for lesson in lessons:
|
|
lesson_url = url_or_none(lesson.get('http_url'))
|
|
if not lesson_url:
|
|
continue
|
|
lesson_id = lesson.get('id')
|
|
if lesson_id:
|
|
lesson_id = compat_str(lesson_id)
|
|
entries.append(self.url_result(
|
|
lesson_url, ie=EggheadLessonIE.ie_key(), video_id=lesson_id))
|
|
|
|
course = self._download_json(
|
|
'https://egghead.io/api/v1/series/%s' % playlist_id,
|
|
playlist_id, 'Downloading course JSON', fatal=False) or {}
|
|
|
|
playlist_id = course.get('id')
|
|
if playlist_id:
|
|
playlist_id = compat_str(playlist_id)
|
|
|
|
return self.playlist_result(
|
|
entries, playlist_id, course.get('title'),
|
|
course.get('description'))
|
|
|
|
|
|
class EggheadLessonIE(InfoExtractor):
|
|
IE_DESC = 'egghead.io lesson'
|
|
IE_NAME = 'egghead:lesson'
|
|
_VALID_URL = r'https://egghead\.io/(?:api/v1/)?lessons/(?P<id>[^/?#&]+)'
|
|
_TESTS = [{
|
|
'url': 'https://egghead.io/lessons/javascript-linear-data-flow-with-container-style-types-box',
|
|
'info_dict': {
|
|
'id': '1196',
|
|
'display_id': 'javascript-linear-data-flow-with-container-style-types-box',
|
|
'ext': 'mp4',
|
|
'title': 'Create linear data flow with container style types (Box)',
|
|
'description': 'md5:9aa2cdb6f9878ed4c39ec09e85a8150e',
|
|
'thumbnail': r're:^https?:.*\.jpg$',
|
|
'timestamp': 1481296768,
|
|
'upload_date': '20161209',
|
|
'duration': 304,
|
|
'view_count': 0,
|
|
'tags': ['javascript', 'free'],
|
|
},
|
|
'params': {
|
|
'skip_download': True,
|
|
'format': 'bestvideo',
|
|
},
|
|
}, {
|
|
'url': 'https://egghead.io/api/v1/lessons/react-add-redux-to-a-react-application',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
display_id = self._match_id(url)
|
|
|
|
lesson = self._download_json(
|
|
'https://egghead.io/api/v1/lessons/%s' % display_id, display_id)
|
|
|
|
lesson_id = compat_str(lesson['id'])
|
|
title = lesson['title']
|
|
|
|
formats = []
|
|
for _, format_url in lesson['media_urls'].items():
|
|
format_url = url_or_none(format_url)
|
|
if not format_url:
|
|
continue
|
|
ext = determine_ext(format_url)
|
|
if ext == 'm3u8':
|
|
formats.extend(self._extract_m3u8_formats(
|
|
format_url, lesson_id, 'mp4', entry_protocol='m3u8',
|
|
m3u8_id='hls', fatal=False))
|
|
elif ext == 'mpd':
|
|
formats.extend(self._extract_mpd_formats(
|
|
format_url, lesson_id, mpd_id='dash', fatal=False))
|
|
else:
|
|
formats.append({
|
|
'url': format_url,
|
|
})
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': lesson_id,
|
|
'display_id': display_id,
|
|
'title': title,
|
|
'description': lesson.get('summary'),
|
|
'thumbnail': lesson.get('thumb_nail'),
|
|
'timestamp': unified_timestamp(lesson.get('published_at')),
|
|
'duration': int_or_none(lesson.get('duration')),
|
|
'view_count': int_or_none(lesson.get('plays_count')),
|
|
'tags': try_get(lesson, lambda x: x['tag_list'], list),
|
|
'series': try_get(
|
|
lesson, lambda x: x['series']['title'], compat_str),
|
|
'formats': formats,
|
|
}
|