mirror of https://github.com/yt-dlp/yt-dlp.git
56 lines
1.8 KiB
Python
56 lines
1.8 KiB
Python
from __future__ import unicode_literals
|
|
|
|
import base64
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
compat_urllib_parse,
|
|
)
|
|
|
|
|
|
class InfoQIE(InfoExtractor):
|
|
_VALID_URL = r'^(?:https?://)?(?:www\.)?infoq\.com/[^/]+/[^/]+$'
|
|
_TEST = {
|
|
"name": "InfoQ",
|
|
"url": "http://www.infoq.com/presentations/A-Few-of-My-Favorite-Python-Things",
|
|
"file": "12-jan-pythonthings.mp4",
|
|
"info_dict": {
|
|
"description": "Mike Pirnat presents some tips and tricks, standard libraries and third party packages that make programming in Python a richer experience.",
|
|
"title": "A Few of My Favorite [Python] Things",
|
|
},
|
|
"params": {
|
|
"skip_download": True,
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
|
|
webpage = self._download_webpage(url, video_id=url)
|
|
self.report_extraction(url)
|
|
|
|
# Extract video URL
|
|
encoded_id = self._search_regex(r"jsclassref ?= ?'([^']*)'", webpage, 'encoded id')
|
|
real_id = compat_urllib_parse.unquote(base64.b64decode(encoded_id.encode('ascii')).decode('utf-8'))
|
|
video_url = 'rtmpe://video.infoq.com/cfx/st/' + real_id
|
|
|
|
# Extract title
|
|
video_title = self._search_regex(r'contentTitle = "(.*?)";',
|
|
webpage, 'title')
|
|
|
|
# Extract description
|
|
video_description = self._html_search_regex(r'<meta name="description" content="(.*)"(?:\s*/)?>',
|
|
webpage, 'description', fatal=False)
|
|
|
|
video_filename = video_url.split('/')[-1]
|
|
video_id, extension = video_filename.split('.')
|
|
|
|
return {
|
|
'id': video_id,
|
|
'url': video_url,
|
|
'title': video_title,
|
|
'ext': extension, # Extension is always(?) mp4, but seems to be flv
|
|
'description': video_description,
|
|
}
|