diff options
author | Sergey M․ <dstftw@gmail.com> | 2015-08-23 22:32:20 +0600 |
---|---|---|
committer | Sergey M․ <dstftw@gmail.com> | 2015-08-23 22:32:20 +0600 |
commit | 551c7837ace81190ce9141551dceec24dfdae1bc (patch) | |
tree | cb6c212aa634db2d928cb2a9853a948583f94ba6 /youtube_dl/extractor/spankwire.py | |
parent | 59e6acc757a9df85ca78e519623e84072ffd9c01 (diff) |
[spankwire] Simplify and properly format
Diffstat (limited to 'youtube_dl/extractor/spankwire.py')
-rw-r--r-- | youtube_dl/extractor/spankwire.py | 105 |
1 files changed, 45 insertions, 60 deletions
diff --git a/youtube_dl/extractor/spankwire.py b/youtube_dl/extractor/spankwire.py index 0f2d8d0de..0a47441b1 100644 --- a/youtube_dl/extractor/spankwire.py +++ b/youtube_dl/extractor/spankwire.py @@ -18,33 +18,34 @@ from ..aes import aes_decrypt_text class SpankwireIE(InfoExtractor): _VALID_URL = r'https?://(?:www\.)?(?P<url>spankwire\.com/[^/]*/video(?P<videoid>[0-9]+)/?)' _TESTS = [{ - 'url': 'http://www.spankwire.com/Buckcherry-s-X-Rated-Music-Video-Crazy-Bitch/video103545/', - 'md5': '8bbfde12b101204b39e4b9fe7eb67095', - 'info_dict': { - 'id': '103545', - 'ext': 'mp4', - 'title': 'Buckcherry`s X Rated Music Video Crazy Bitch', - 'description': 'Crazy Bitch X rated music video.', - 'uploader': 'oreusz', - 'uploader_id': '124697', - 'upload_date': '20070507', - 'age_limit': 18, - } - }, - { - 'url': 'http://www.spankwire.com/Titcums-Compiloation-I/video1921551/', - 'md5': '09b3c20833308b736ae8902db2f8d7e6', - 'info_dict': { - 'id': '1921551', - 'ext': 'mp4', - 'title': 'Titcums Compiloation I', - 'description': 'cum on tits', - 'uploader': 'dannyh78999', - 'uploader_id': '3056053', - 'upload_date': '20150822', - 'age_limit': 18, - } - }] + # download URL pattern: */<height>P_<tbr>K_<video_id>.mp4 + 'url': 'http://www.spankwire.com/Buckcherry-s-X-Rated-Music-Video-Crazy-Bitch/video103545/', + 'md5': '8bbfde12b101204b39e4b9fe7eb67095', + 'info_dict': { + 'id': '103545', + 'ext': 'mp4', + 'title': 'Buckcherry`s X Rated Music Video Crazy Bitch', + 'description': 'Crazy Bitch X rated music video.', + 'uploader': 'oreusz', + 'uploader_id': '124697', + 'upload_date': '20070507', + 'age_limit': 18, + } + }, { + # download URL pattern: */mp4_<format_id>_<video_id>.mp4 + 'url': 'http://www.spankwire.com/Titcums-Compiloation-I/video1921551/', + 'md5': '09b3c20833308b736ae8902db2f8d7e6', + 'info_dict': { + 'id': '1921551', + 'ext': 'mp4', + 'title': 'Titcums Compiloation I', + 'description': 'cum on tits', + 'uploader': 'dannyh78999', + 'uploader_id': '3056053', + 'upload_date': '20150822', + 'age_limit': 18, + }, + }] def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) @@ -81,9 +82,10 @@ class SpankwireIE(InfoExtractor): r'<span\s+id="spCommentCount"[^>]*>([\d,\.]+)</span>', webpage, 'comment count', fatal=False)) - video_urls = list(map( - compat_urllib_parse_unquote, - re.findall(r'playerData\.cdnPath[0-9]{3,}\s*=\s*(?:encodeURIComponent\()?["\']([^"\']+)["\']', webpage))) + videos = re.findall( + r'playerData\.cdnPath([0-9]{3,})\s*=\s*(?:encodeURIComponent\()?["\']([^"\']+)["\']', webpage) + heights = [int(video[0]) for video in videos] + video_urls = list(map(compat_urllib_parse_unquote, [video[1] for video in videos])) if webpage.find('flashvars\.encrypted = "true"') != -1: password = self._search_regex( r'flashvars\.video_title = "([^"]+)', @@ -93,39 +95,22 @@ class SpankwireIE(InfoExtractor): video_urls)) formats = [] - for video_url in video_urls: + for height, video_url in zip(heights, video_urls): path = compat_urllib_parse_urlparse(video_url).path - format = path.split('/')[4].split('_')[:2] - if format[0] == 'mp4': - format_id, quality = format - format = "-".join(format) - if quality == 'normal': - height = 180 - elif quality == 'high': - height = 240 - elif quality == 'ultra': - height = 480 - elif quality == '720p': - height = 720 - formats.append({ - 'url': video_url, - 'format': format, - 'height': height, - 'format_id': format, + _, quality = path.split('/')[4].split('_')[:2] + f = { + 'url': video_url, + 'height': height, + } + tbr = self._search_regex(r'^(\d+)[Kk]$', quality, 'tbr', default=None) + if tbr: + f.update({ + 'tbr': int(tbr), + 'format_id': '%dp' % height, }) else: - resolution, bitrate_str = format - format = "-".join(format) - height = int(resolution.rstrip('Pp')) - tbr = int(bitrate_str.rstrip('Kk')) - formats.append({ - 'url': video_url, - 'resolution': resolution, - 'format': format, - 'tbr': tbr, - 'height': height, - 'format_id': format, - }) + f['format_id'] = quality + formats.append(f) self._sort_formats(formats) age_limit = self._rta_search(webpage) |