mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-15 07:42:42 +00:00
117 lines
4 KiB
Python
117 lines
4 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import itertools
|
|
import json
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import (
|
|
compat_urllib_request,
|
|
compat_urlparse,
|
|
)
|
|
from ..utils import (
|
|
int_or_none,
|
|
js_to_json,
|
|
mimetype2ext,
|
|
unified_strdate,
|
|
)
|
|
|
|
|
|
class SandiaIE(InfoExtractor):
|
|
IE_DESC = 'Sandia National Laboratories'
|
|
_VALID_URL = r'https?://digitalops\.sandia\.gov/Mediasite/Play/(?P<id>[0-9a-f]+)'
|
|
_TEST = {
|
|
'url': 'http://digitalops.sandia.gov/Mediasite/Play/24aace4429fc450fb5b38cdbf424a66e1d',
|
|
'md5': '9422edc9b9a60151727e4b6d8bef393d',
|
|
'info_dict': {
|
|
'id': '24aace4429fc450fb5b38cdbf424a66e1d',
|
|
'ext': 'mp4',
|
|
'title': 'Xyce Software Training - Section 1',
|
|
'description': 're:(?s)SAND Number: SAND 2013-7800.{200,}',
|
|
'upload_date': '20120904',
|
|
'duration': 7794,
|
|
}
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
req = compat_urllib_request.Request(url)
|
|
req.add_header('Cookie', 'MediasitePlayerCaps=ClientPlugins=4')
|
|
webpage = self._download_webpage(req, video_id)
|
|
|
|
js_path = self._search_regex(
|
|
r'<script type="text/javascript" src="(/Mediasite/FileServer/Presentation/[^"]+)"',
|
|
webpage, 'JS code URL')
|
|
js_url = compat_urlparse.urljoin(url, js_path)
|
|
|
|
js_code = self._download_webpage(
|
|
js_url, video_id, note='Downloading player')
|
|
|
|
def extract_str(key, **args):
|
|
return self._search_regex(
|
|
r'Mediasite\.PlaybackManifest\.%s\s*=\s*(.+);\s*?\n' % re.escape(key),
|
|
js_code, key, **args)
|
|
|
|
def extract_data(key, **args):
|
|
data_json = extract_str(key, **args)
|
|
if data_json is None:
|
|
return data_json
|
|
return self._parse_json(
|
|
data_json, video_id, transform_source=js_to_json)
|
|
|
|
formats = []
|
|
for i in itertools.count():
|
|
fd = extract_data('VideoUrls[%d]' % i, default=None)
|
|
if fd is None:
|
|
break
|
|
formats.append({
|
|
'format_id': '%s' % i,
|
|
'format_note': fd['MimeType'].partition('/')[2],
|
|
'ext': mimetype2ext(fd['MimeType']),
|
|
'url': fd['Location'],
|
|
'protocol': 'f4m' if fd['MimeType'] == 'video/x-mp4-fragmented' else None,
|
|
})
|
|
self._sort_formats(formats)
|
|
|
|
slide_baseurl = compat_urlparse.urljoin(
|
|
url, extract_data('SlideBaseUrl'))
|
|
slide_template = slide_baseurl + re.sub(
|
|
r'\{0:D?([0-9+])\}', r'%0\1d', extract_data('SlideImageFileNameTemplate'))
|
|
slides = []
|
|
last_slide_time = 0
|
|
for i in itertools.count(1):
|
|
sd = extract_str('Slides[%d]' % i, default=None)
|
|
if sd is None:
|
|
break
|
|
timestamp = int_or_none(self._search_regex(
|
|
r'^Mediasite\.PlaybackManifest\.CreateSlide\("[^"]*"\s*,\s*([0-9]+),',
|
|
sd, 'slide %s timestamp' % i, fatal=False))
|
|
slides.append({
|
|
'url': slide_template % i,
|
|
'duration': timestamp - last_slide_time,
|
|
})
|
|
last_slide_time = timestamp
|
|
formats.append({
|
|
'format_id': 'slides',
|
|
'protocol': 'slideshow',
|
|
'url': json.dumps(slides),
|
|
'preference': -10000, # Downloader not yet written
|
|
})
|
|
self._sort_formats(formats)
|
|
|
|
title = extract_data('Title')
|
|
description = extract_data('Description', fatal=False)
|
|
duration = int_or_none(extract_data(
|
|
'Duration', fatal=False), scale=1000)
|
|
upload_date = unified_strdate(extract_data('AirDate', fatal=False))
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'description': description,
|
|
'formats': formats,
|
|
'upload_date': upload_date,
|
|
'duration': duration,
|
|
}
|