aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor/ccc.py
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl/extractor/ccc.py')
-rw-r--r--youtube_dl/extractor/ccc.py35
1 files changed, 24 insertions, 11 deletions
diff --git a/youtube_dl/extractor/ccc.py b/youtube_dl/extractor/ccc.py
index 6924eac70..e94b1e35b 100644
--- a/youtube_dl/extractor/ccc.py
+++ b/youtube_dl/extractor/ccc.py
@@ -5,6 +5,7 @@ import re
from .common import InfoExtractor
from ..utils import (
int_or_none,
+ parse_duration,
qualities,
unified_strdate,
)
@@ -12,21 +13,25 @@ from ..utils import (
class CCCIE(InfoExtractor):
IE_NAME = 'media.ccc.de'
- _VALID_URL = r'https?://(?:www\.)?media\.ccc\.de/[^?#]+/[^?#/]*?_(?P<id>[0-9]{8,})._[^?#/]*\.html'
+ _VALID_URL = r'https?://(?:www\.)?media\.ccc\.de/v/(?P<id>[^/?#&]+)'
- _TEST = {
- 'url': 'http://media.ccc.de/browse/congress/2013/30C3_-_5443_-_en_-_saal_g_-_201312281830_-_introduction_to_processor_design_-_byterazor.html#video',
+ _TESTS = [{
+ 'url': 'https://media.ccc.de/v/30C3_-_5443_-_en_-_saal_g_-_201312281830_-_introduction_to_processor_design_-_byterazor#video',
'md5': '3a1eda8f3a29515d27f5adb967d7e740',
'info_dict': {
- 'id': '20131228183',
+ 'id': '30C3_-_5443_-_en_-_saal_g_-_201312281830_-_introduction_to_processor_design_-_byterazor',
'ext': 'mp4',
'title': 'Introduction to Processor Design',
- 'description': 'md5:5ddbf8c734800267f2cee4eab187bc1b',
+ 'description': 'md5:80be298773966f66d56cb11260b879af',
'thumbnail': 're:^https?://.*\.jpg$',
'view_count': int,
- 'upload_date': '20131229',
+ 'upload_date': '20131228',
+ 'duration': 3660,
}
- }
+ }, {
+ 'url': 'https://media.ccc.de/v/32c3-7368-shopshifting#download',
+ 'only_matching': True,
+ }]
def _real_extract(self, url):
video_id = self._match_id(url)
@@ -40,21 +45,25 @@ class CCCIE(InfoExtractor):
title = self._html_search_regex(
r'(?s)<h1>(.*?)</h1>', webpage, 'title')
description = self._html_search_regex(
- r"(?s)<p class='description'>(.*?)</p>",
+ r"(?s)<h3>About</h3>(.+?)<h3>",
webpage, 'description', fatal=False)
upload_date = unified_strdate(self._html_search_regex(
- r"(?s)<span class='[^']*fa-calendar-o'></span>(.*?)</li>",
+ r"(?s)<span[^>]+class='[^']*fa-calendar-o'[^>]*>(.+?)</span>",
webpage, 'upload date', fatal=False))
view_count = int_or_none(self._html_search_regex(
r"(?s)<span class='[^']*fa-eye'></span>(.*?)</li>",
webpage, 'view count', fatal=False))
+ duration = parse_duration(self._html_search_regex(
+ r'(?s)<span[^>]+class=(["\']).*?fa-clock-o.*?\1[^>]*></span>(?P<duration>.+?)</li',
+ webpage, 'duration', fatal=False, group='duration'))
matches = re.finditer(r'''(?xs)
- <(?:span|div)\s+class='label\s+filetype'>(?P<format>.*?)</(?:span|div)>\s*
+ <(?:span|div)\s+class='label\s+filetype'>(?P<format>[^<]*)</(?:span|div)>\s*
+ <(?:span|div)\s+class='label\s+filetype'>(?P<lang>[^<]*)</(?:span|div)>\s*
<a\s+download\s+href='(?P<http_url>[^']+)'>\s*
(?:
.*?
- <a\s+href='(?P<torrent_url>[^']+\.torrent)'
+ <a\s+(?:download\s+)?href='(?P<torrent_url>[^']+\.torrent)'
)?''', webpage)
formats = []
for m in matches:
@@ -62,12 +71,15 @@ class CCCIE(InfoExtractor):
format_id = self._search_regex(
r'.*/([a-z0-9_-]+)/[^/]*$',
m.group('http_url'), 'format id', default=None)
+ if format_id:
+ format_id = m.group('lang') + '-' + format_id
vcodec = 'h264' if 'h264' in format_id else (
'none' if format_id in ('mp3', 'opus') else None
)
formats.append({
'format_id': format_id,
'format': format,
+ 'language': m.group('lang'),
'url': m.group('http_url'),
'vcodec': vcodec,
'preference': preference(format_id),
@@ -95,5 +107,6 @@ class CCCIE(InfoExtractor):
'thumbnail': thumbnail,
'view_count': view_count,
'upload_date': upload_date,
+ 'duration': duration,
'formats': formats,
}