aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--youtube_dl/extractor/cinemassacre.py65
1 files changed, 40 insertions, 25 deletions
diff --git a/youtube_dl/extractor/cinemassacre.py b/youtube_dl/extractor/cinemassacre.py
index 2301f61b6..496271be4 100644
--- a/youtube_dl/extractor/cinemassacre.py
+++ b/youtube_dl/extractor/cinemassacre.py
@@ -1,10 +1,12 @@
# encoding: utf-8
from __future__ import unicode_literals
+
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
+ int_or_none,
)
@@ -13,9 +15,10 @@ class CinemassacreIE(InfoExtractor):
_TESTS = [
{
'url': 'http://cinemassacre.com/2012/11/10/avgn-the-movie-trailer/',
- 'file': '19911.mp4',
- 'md5': '782f8504ca95a0eba8fc9177c373eec7',
+ 'md5': 'fde81fbafaee331785f58cd6c0d46190',
'info_dict': {
+ 'id': '19911',
+ 'ext': 'mp4',
'upload_date': '20121110',
'title': '“Angry Video Game Nerd: The Movie” – Trailer',
'description': 'md5:fb87405fcb42a331742a0dce2708560b',
@@ -23,9 +26,10 @@ class CinemassacreIE(InfoExtractor):
},
{
'url': 'http://cinemassacre.com/2013/10/02/the-mummys-hand-1940',
- 'file': '521be8ef82b16.mp4',
- 'md5': 'dec39ee5118f8d9cc067f45f9cbe3a35',
+ 'md5': 'd72f10cd39eac4215048f62ab477a511',
'info_dict': {
+ 'id': '521be8ef82b16',
+ 'ext': 'mp4',
'upload_date': '20131002',
'title': 'The Mummy’s Hand (1940)',
},
@@ -50,29 +54,40 @@ class CinemassacreIE(InfoExtractor):
r'<div class="entry-content">(?P<description>.+?)</div>',
webpage, 'description', flags=re.DOTALL, fatal=False)
- playerdata = self._download_webpage(playerdata_url, video_id)
+ playerdata = self._download_webpage(playerdata_url, video_id, 'Downloading player webpage')
+ video_thumbnail = self._search_regex(
+ r'image: \'(?P<thumbnail>[^\']+)\'', playerdata, 'thumbnail', fatal=False)
+ sd_url = self._search_regex(r'file: \'([^\']+)\', label: \'SD\'', playerdata, 'sd_file')
+ videolist_url = self._search_regex(r'file: \'([^\']+\.smil)\'}', playerdata, 'videolist_url')
- sd_url = self._html_search_regex(r'file: \'([^\']+)\', label: \'SD\'', playerdata, 'sd_file')
- hd_url = self._html_search_regex(
- r'file: \'([^\']+)\', label: \'HD\'', playerdata, 'hd_file',
- default=None)
- video_thumbnail = self._html_search_regex(r'image: \'(?P<thumbnail>[^\']+)\'', playerdata, 'thumbnail', fatal=False)
+ videolist = self._download_xml(videolist_url, video_id, 'Downloading videolist XML')
- formats = [{
- 'url': sd_url,
- 'ext': 'mp4',
- 'format': 'sd',
- 'format_id': 'sd',
- 'quality': 1,
- }]
- if hd_url:
- formats.append({
- 'url': hd_url,
- 'ext': 'mp4',
- 'format': 'hd',
- 'format_id': 'hd',
- 'quality': 2,
- })
+ formats = []
+ baseurl = sd_url[:sd_url.rfind('/')+1]
+ for video in videolist.findall('.//video'):
+ src = video.get('src')
+ if not src:
+ continue
+ file_ = src.partition(':')[-1]
+ width = int_or_none(video.get('width'))
+ height = int_or_none(video.get('height'))
+ bitrate = int_or_none(video.get('system-bitrate'))
+ format = {
+ 'url': baseurl + file_,
+ 'format_id': src.rpartition('.')[0].rpartition('_')[-1],
+ }
+ if width or height:
+ format.update({
+ 'tbr': bitrate // 1000 if bitrate else None,
+ 'width': width,
+ 'height': height,
+ })
+ else:
+ format.update({
+ 'abr': bitrate // 1000 if bitrate else None,
+ 'vcodec': 'none',
+ })
+ formats.append(format)
self._sort_formats(formats)
return {