diff options
Diffstat (limited to 'youtube_dl/InfoExtractors.py')
-rwxr-xr-x | youtube_dl/InfoExtractors.py | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/youtube_dl/InfoExtractors.py b/youtube_dl/InfoExtractors.py index 83cb32196..b3c3dbb43 100755 --- a/youtube_dl/InfoExtractors.py +++ b/youtube_dl/InfoExtractors.py @@ -2305,7 +2305,7 @@ class MyVideoIE(InfoExtractor): webpage = self._download_webpage(webpage_url, video_id) self.report_extraction(video_id) - mobj = re.search(r'<link rel=\'image_src\' href=\'(http://is[0-9].myvideo\.de/de/movie[0-9]+/[a-f0-9]+)/thumbs/.*?\.jpg\' />', + mobj = re.search(r'<link rel=\'image_src\' href=\'(http://is[0-9].myvideo\.de/de/movie[0-9]+/[a-f0-9]+)/thumbs/.*?\.jpg\'', webpage) if mobj is None: self._downloader.report_error(u'unable to extract media URL') @@ -3604,10 +3604,10 @@ class FunnyOrDieIE(InfoExtractor): self._downloader.report_error(u'unable to find video information') video_url = unescapeHTML(m.group('url')) - m = re.search(r"class='player_page_h1'>\s+<a.*?>(?P<title>.*?)</a>", webpage) + m = re.search(r"<h1 class='player_page_h1'.*?>(?P<title>.*?)</h1>", webpage, flags=re.DOTALL) if not m: self._downloader.trouble(u'Cannot find video title') - title = unescapeHTML(m.group('title')) + title = clean_html(m.group('title')) m = re.search(r'<meta property="og:description" content="(?P<desc>.*?)"', webpage) if m: @@ -4051,7 +4051,7 @@ class TEDIE(InfoExtractor): videoName=m.group('name') webpage=self._download_webpage(url, video_id, 'Downloading \"%s\" page' % videoName) # If the url includes the language we get the title translated - title_RE=r'<h1><span id="altHeadline" >(?P<title>.*)</span></h1>' + title_RE=r'<span id="altHeadline" >(?P<title>.*)</span>' title=re.search(title_RE, webpage).group('title') info_RE=r'''<script\ type="text/javascript">var\ talkDetails\ =(.*?) "id":(?P<videoID>[\d]+).*? |