transplant 0f6e296a8ec97bce781b722bbaf4cff95642d253
authorPhilipp Hagemeister <phihag@phihag.de>
Thu, 15 Mar 2012 06:56:32 +0000 (07:56 +0100)
committerPhilipp Hagemeister <phihag@phihag.de>
Thu, 15 Mar 2012 06:56:32 +0000 (07:56 +0100)
youtube_dl/__init__.py

index 789724041be58dc3494057bc8e132efe0e090aa4..77723155d3a65d586af8e6c31d5c3edeedcb3c04 100755 (executable)
@@ -2340,8 +2340,8 @@ class GoogleSearchIE(InfoExtractor):
        """Information Extractor for Google Video search queries."""
        _VALID_URL = r'gvsearch(\d+|all)?:[\s\S]+'
        _TEMPLATE_URL = 'http://video.google.com/videosearch?q=%s+site:video.google.com&start=%s&hl=en'
-       _VIDEO_INDICATOR = r'videoplay\?docid=([^\&>]+)\&'
-       _MORE_PAGES_INDICATOR = r'<span>Next</span>'
+       _VIDEO_INDICATOR = r'<a href="http://video\.google\.com/videoplay\?docid=([^"\&]+)'
+       _MORE_PAGES_INDICATOR = r'class="pn" id="pnnext"'
        _google_ie = None
        _max_google_results = 1000
        IE_NAME = u'video.google:search'
@@ -2392,12 +2392,11 @@ class GoogleSearchIE(InfoExtractor):
                """Downloads a specified number of results for a query"""
 
                video_ids = []
-               already_seen = set()
-               pagenum = 1
+               pagenum = 0
 
                while True:
                        self.report_download_page(query, pagenum)
-                       result_url = self._TEMPLATE_URL % (urllib.quote_plus(query), pagenum)
+                       result_url = self._TEMPLATE_URL % (urllib.quote_plus(query), pagenum*10)
                        request = urllib2.Request(result_url)
                        try:
                                page = urllib2.urlopen(request).read()
@@ -2408,9 +2407,8 @@ class GoogleSearchIE(InfoExtractor):
                        # Extract video identifiers
                        for mobj in re.finditer(self._VIDEO_INDICATOR, page):
                                video_id = mobj.group(1)
-                               if video_id not in already_seen:
+                               if video_id not in video_ids:
                                        video_ids.append(video_id)
-                                       already_seen.add(video_id)
                                        if len(video_ids) == n:
                                                # Specified n videos reached
                                                for id in video_ids: