mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-11-05 10:37:04 +01:00
82 lines
2.6 KiB
Python
82 lines
2.6 KiB
Python
|
from __future__ import unicode_literals
|
||
|
|
||
|
import re
|
||
|
import json
|
||
|
|
||
|
from .common import InfoExtractor
|
||
|
from ..utils import (
|
||
|
unescapeHTML,
|
||
|
qualities,
|
||
|
int_or_none,
|
||
|
)
|
||
|
|
||
|
|
||
|
class GiantBombIE(InfoExtractor):
|
||
|
_VALID_URL = r'https?://(?:www\.)?giantbomb\.com/videos/(?P<display_id>[^/]+)/(?P<id>\d+-\d+)'
|
||
|
_TEST = {
|
||
|
'url': 'http://www.giantbomb.com/videos/quick-look-destiny-the-dark-below/2300-9782/',
|
||
|
'md5': '57badeface303ecf6b98b812de1b9018',
|
||
|
'info_dict': {
|
||
|
'id': '2300-9782',
|
||
|
'display_id': 'quick-look-destiny-the-dark-below',
|
||
|
'ext': 'mp4',
|
||
|
'title': 'Quick Look: Destiny: The Dark Below',
|
||
|
'description': 'md5:0aa3aaf2772a41b91d44c63f30dfad24',
|
||
|
'duration': 2399,
|
||
|
'thumbnail': 're:^https?://.*\.jpg$',
|
||
|
}
|
||
|
}
|
||
|
|
||
|
def _real_extract(self, url):
|
||
|
mobj = re.match(self._VALID_URL, url)
|
||
|
video_id = mobj.group('id')
|
||
|
display_id = mobj.group('display_id')
|
||
|
|
||
|
webpage = self._download_webpage(url, display_id)
|
||
|
|
||
|
title = self._og_search_title(webpage)
|
||
|
description = self._og_search_description(webpage)
|
||
|
thumbnail = self._og_search_thumbnail(webpage)
|
||
|
|
||
|
video = json.loads(unescapeHTML(self._search_regex(
|
||
|
r'data-video="([^"]+)"', webpage, 'data-video')))
|
||
|
|
||
|
duration = int_or_none(video.get('lengthSeconds'))
|
||
|
|
||
|
quality = qualities([
|
||
|
'f4m_low', 'progressive_low', 'f4m_high',
|
||
|
'progressive_high', 'f4m_hd', 'progressive_hd'])
|
||
|
|
||
|
formats = []
|
||
|
for format_id, video_url in video['videoStreams'].items():
|
||
|
if format_id == 'f4m_stream':
|
||
|
continue
|
||
|
if video_url.endswith('.f4m'):
|
||
|
f4m_formats = self._extract_f4m_formats(video_url + '?hdcore=3.3.1', display_id)
|
||
|
if f4m_formats:
|
||
|
f4m_formats[0]['quality'] = quality(format_id)
|
||
|
formats.extend(f4m_formats)
|
||
|
else:
|
||
|
formats.append({
|
||
|
'url': video_url,
|
||
|
'format_id': format_id,
|
||
|
'quality': quality(format_id),
|
||
|
})
|
||
|
|
||
|
if not formats:
|
||
|
youtube_id = video.get('youtubeID')
|
||
|
if youtube_id:
|
||
|
return self.url_result(youtube_id, 'Youtube')
|
||
|
|
||
|
self._sort_formats(formats)
|
||
|
|
||
|
return {
|
||
|
'id': video_id,
|
||
|
'display_id': display_id,
|
||
|
'title': title,
|
||
|
'description': description,
|
||
|
'thumbnail': thumbnail,
|
||
|
'duration': duration,
|
||
|
'formats': formats,
|
||
|
}
|