@ -21,7 +21,7 @@ class YahooIE(InfoExtractor):
' url ' : ' http://screen.yahoo.com/julian-smith-travis-legg-watch-214727115.html ' ,
' md5 ' : ' 4962b075c08be8690a922ee026d05e69 ' ,
' info_dict ' : {
' id ' : ' 214727115 ' ,
' id ' : ' 2d25e626-2378-391f-ada0-ddaf1417e588 ' ,
' ext ' : ' mp4 ' ,
' title ' : ' Julian Smith & Travis Legg Watch Julian Smith ' ,
' description ' : ' Julian and Travis watch Julian Smith ' ,
@ -31,7 +31,7 @@ class YahooIE(InfoExtractor):
' url ' : ' http://screen.yahoo.com/wired/codefellas-s1-ep12-cougar-lies-103000935.html ' ,
' md5 ' : ' d6e6fc6e1313c608f316ddad7b82b306 ' ,
' info_dict ' : {
' id ' : ' 103000935 ' ,
' id ' : ' d1dedf8c-d58c-38c3-8963-e899929ae0a9 ' ,
' ext ' : ' mp4 ' ,
' title ' : ' Codefellas - The Cougar Lies with Spanish Moss ' ,
' description ' : ' Agent Topple \' s mustache does its dirty work, and Nicole brokers a deal for peace. But why is the NSA collecting millions of Instagram brunch photos? And if your waffles have nothing to hide, what are they so worried about? ' ,
@ -58,9 +58,11 @@ class YahooIE(InfoExtractor):
r ' mediaItems: ({.*?})$ ' , webpage , ' items ' , flags = re . MULTILINE ,
default = None )
if items_json is None :
long_id = self . _search_regex (
CONTENT_ID_REGEXES = [
r ' YUI \ .namespace \ ( " Media " \ ) \ .CONTENT_ID \ s*= \ s* " ([^ " ]+) " ' ,
webpage , ' content ID ' )
r ' root \ .App \ .Cache \ .context \ .videoCache \ .curVideo = \ { " ([^ " ]+) " '
]
long_id = self . _search_regex ( CONTENT_ID_REGEXES , webpage , ' content ID ' )
video_id = long_id
else :
items = json . loads ( items_json )