|
|
@ -98,6 +98,7 @@ class TVPEmbedIE(InfoExtractor): |
|
|
|
'title': 'Czas honoru, odc. 13 – Władek', |
|
|
|
}, |
|
|
|
}, { |
|
|
|
# not available |
|
|
|
'url': 'http://www.tvp.pl/sess/tvplayer.php?object_id=22670268', |
|
|
|
'md5': '8c9cd59d16edabf39331f93bf8a766c7', |
|
|
|
'info_dict': { |
|
|
@ -105,6 +106,7 @@ class TVPEmbedIE(InfoExtractor): |
|
|
|
'ext': 'mp4', |
|
|
|
'title': 'Panorama, 07.12.2015, 15:40', |
|
|
|
}, |
|
|
|
'skip': 'Transmisja została zakończona lub materiał niedostępny', |
|
|
|
}, { |
|
|
|
'url': 'tvp:22670268', |
|
|
|
'only_matching': True, |
|
|
@ -116,10 +118,13 @@ class TVPEmbedIE(InfoExtractor): |
|
|
|
webpage = self._download_webpage( |
|
|
|
'http://www.tvp.pl/sess/tvplayer.php?object_id=%s' % video_id, video_id) |
|
|
|
|
|
|
|
error_massage = get_element_by_attribute('class', 'msg error', webpage) |
|
|
|
if error_massage: |
|
|
|
error = self._html_search_regex( |
|
|
|
r'(?s)<p[^>]+\bclass=["\']notAvailable__text["\'][^>]*>(.+?)</p>', |
|
|
|
webpage, 'error', default=None) or clean_html( |
|
|
|
get_element_by_attribute('class', 'msg error', webpage)) |
|
|
|
if error: |
|
|
|
raise ExtractorError('%s said: %s' % ( |
|
|
|
self.IE_NAME, clean_html(error_massage)), expected=True) |
|
|
|
self.IE_NAME, clean_html(error)), expected=True) |
|
|
|
|
|
|
|
title = self._search_regex( |
|
|
|
r'name\s*:\s*([\'"])Title\1\s*,\s*value\s*:\s*\1(?P<title>.+?)\1', |
|
|
|