1
0
Fork 0
mirror of https://gitlab.com/dstftw/youtube-dl.git synced 2020-11-16 09:42:26 +00:00

[theonion] Add new extractor (closes #3928)

This commit is contained in:
Naglis Jonaitis 2014-10-12 15:42:35 +03:00
parent b4c3c8c172
commit ac20fc047a
2 changed files with 72 additions and 0 deletions

View file

@ -371,6 +371,7 @@ from .telemb import TeleMBIE
from .tenplay import TenPlayIE from .tenplay import TenPlayIE
from .testurl import TestURLIE from .testurl import TestURLIE
from .tf1 import TF1IE from .tf1 import TF1IE
from .theonion import TheOnionIE
from .theplatform import ThePlatformIE from .theplatform import ThePlatformIE
from .thesixtyone import TheSixtyOneIE from .thesixtyone import TheSixtyOneIE
from .thisav import ThisAVIE from .thisav import ThisAVIE

View file

@ -0,0 +1,71 @@
# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import ExtractorError
class TheOnionIE(InfoExtractor):
_VALID_URL = r'(?x)https?://(?:www\.)?theonion\.com/video/[^,]+,(?P<article_id>[0-9]+)/?'
_TEST = {
'url': 'http://www.theonion.com/video/man-wearing-mm-jacket-gods-image,36918/',
'md5': '19eaa9a39cf9b9804d982e654dc791ee',
'info_dict': {
'id': '2133',
'ext': 'mp4',
'title': 'Man Wearing M&M Jacket Apparently Made In God\'s Image',
'description': 'md5:cc12448686b5600baae9261d3e180910',
'thumbnail': 're:^https?://.*\.jpg\?\d+$',
}
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
article_id = mobj.group('article_id')
webpage = self._download_webpage(url, article_id)
video_id = self._search_regex(
r'"videoId":\s(\d+),', webpage, 'video ID')
title = self._og_search_title(webpage)
description = self._og_search_description(webpage)
thumbnail = self._og_search_thumbnail(webpage)
sources = re.findall(r'<source src="([^"]+)" type="([^"]+)"', webpage)
if not sources:
raise ExtractorError(
'No sources found for video %s' % (self.IE_NAME, video_id),
expected=True)
formats = []
for src, type_ in sources:
if type_ == 'video/mp4':
formats.append({
'format_id': 'mp4_sd',
'preference': 1,
'url': src,
})
elif type_ == 'video/webm':
formats.append({
'format_id': 'webm_sd',
'preference': 0,
'url': src,
})
elif type_ == 'application/x-mpegURL':
formats.extend(
self._extract_m3u8_formats(src, video_id, preference=-1))
else:
self.report_warning(
'Encountered unexpected format: %s' % type_)
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'formats': formats,
'thumbnail': thumbnail,
'description': description,
}