mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-27 04:32:16 +00:00
172 lines
6.6 KiB
Python
172 lines
6.6 KiB
Python
# coding: utf-8
|
||
from __future__ import unicode_literals
|
||
|
||
from .common import InfoExtractor
|
||
from ..utils import parse_iso8601
|
||
|
||
|
||
class NextMediaIE(InfoExtractor):
|
||
IE_DESC = '蘋果日報'
|
||
_VALID_URL = r'http://hk.apple.nextmedia.com/[^/]+/[^/]+/(?P<date>\d+)/(?P<id>\d+)'
|
||
_TESTS = [{
|
||
'url': 'http://hk.apple.nextmedia.com/realtime/news/20141108/53109199',
|
||
'md5': 'dff9fad7009311c421176d1ac90bfe4f',
|
||
'info_dict': {
|
||
'id': '53109199',
|
||
'ext': 'mp4',
|
||
'title': '【佔領金鐘】50外國領事議員撐場 讚學生勇敢香港有希望',
|
||
'thumbnail': 're:^https?://.*\.jpg$',
|
||
'description': 'md5:28222b9912b6665a21011b034c70fcc7',
|
||
'timestamp': 1415456273,
|
||
'upload_date': '20141108',
|
||
}
|
||
}]
|
||
|
||
_URL_PATTERN = r'\{ url: \'(.+)\' \}'
|
||
|
||
def _real_extract(self, url):
|
||
news_id = self._match_id(url)
|
||
page = self._download_webpage(url, news_id)
|
||
return self._extract_from_nextmedia_page(news_id, url, page)
|
||
|
||
def _extract_from_nextmedia_page(self, news_id, url, page):
|
||
title = self._fetch_title(page)
|
||
video_url = self._search_regex(self._URL_PATTERN, page, 'video url')
|
||
|
||
attrs = {
|
||
'id': news_id,
|
||
'title': title,
|
||
'url': video_url, # ext can be inferred from url
|
||
'thumbnail': self._fetch_thumbnail(page),
|
||
'description': self._fetch_description(page),
|
||
}
|
||
|
||
timestamp = self._fetch_timestamp(page)
|
||
if timestamp:
|
||
attrs['timestamp'] = timestamp
|
||
else:
|
||
attrs['upload_date'] = self._fetch_upload_date(url)
|
||
|
||
return attrs
|
||
|
||
def _fetch_title(self, page):
|
||
return self._og_search_title(page)
|
||
|
||
def _fetch_thumbnail(self, page):
|
||
return self._og_search_thumbnail(page)
|
||
|
||
def _fetch_timestamp(self, page):
|
||
dateCreated = self._search_regex('"dateCreated":"([^"]+)"', page, 'created time')
|
||
return parse_iso8601(dateCreated)
|
||
|
||
def _fetch_upload_date(self, url):
|
||
return self._search_regex(self._VALID_URL, url, 'upload date', group='date')
|
||
|
||
def _fetch_description(self, page):
|
||
return self._og_search_property('description', page)
|
||
|
||
|
||
class NextMediaActionNewsIE(NextMediaIE):
|
||
IE_DESC = '蘋果日報 - 動新聞'
|
||
_VALID_URL = r'http://hk.dv.nextmedia.com/actionnews/[^/]+/(?P<date>\d+)/(?P<id>\d+)/\d+'
|
||
_TESTS = [{
|
||
'url': 'http://hk.dv.nextmedia.com/actionnews/hit/20150121/19009428/20061460',
|
||
'md5': '05fce8ffeed7a5e00665d4b7cf0f9201',
|
||
'info_dict': {
|
||
'id': '19009428',
|
||
'ext': 'mp4',
|
||
'title': '【壹週刊】細10年男友偷食 50歲邵美琪再失戀',
|
||
'thumbnail': 're:^https?://.*\.jpg$',
|
||
'description': 'md5:cd802fad1f40fd9ea178c1e2af02d659',
|
||
'timestamp': 1421791200,
|
||
'upload_date': '20150120',
|
||
}
|
||
}]
|
||
|
||
def _real_extract(self, url):
|
||
news_id = self._match_id(url)
|
||
actionnews_page = self._download_webpage(url, news_id)
|
||
article_url = self._og_search_url(actionnews_page)
|
||
article_page = self._download_webpage(article_url, news_id)
|
||
return self._extract_from_nextmedia_page(news_id, url, article_page)
|
||
|
||
|
||
class AppleDailyIE(NextMediaIE):
|
||
IE_DESC = '臺灣蘋果日報'
|
||
_VALID_URL = r'http://(www|ent).appledaily.com.tw/(?:animation|appledaily|enews|realtimenews)/[^/]+/[^/]+/(?P<date>\d+)/(?P<id>\d+)(/.*)?'
|
||
_TESTS = [{
|
||
'url': 'http://ent.appledaily.com.tw/enews/article/entertainment/20150128/36354694',
|
||
'md5': 'a843ab23d150977cc55ef94f1e2c1e4d',
|
||
'info_dict': {
|
||
'id': '36354694',
|
||
'ext': 'mp4',
|
||
'title': '周亭羽走過摩鐵陰霾2男陪吃 九把刀孤寒看醫生',
|
||
'thumbnail': 're:^https?://.*\.jpg$',
|
||
'description': 'md5:2acd430e59956dc47cd7f67cb3c003f4',
|
||
'upload_date': '20150128',
|
||
}
|
||
}, {
|
||
'url': 'http://www.appledaily.com.tw/realtimenews/article/strange/20150128/550549/%E4%B8%8D%E6%BB%BF%E8%A2%AB%E8%B8%A9%E8%85%B3%E3%80%80%E5%B1%B1%E6%9D%B1%E5%85%A9%E5%A4%A7%E5%AA%BD%E4%B8%80%E8%B7%AF%E6%89%93%E4%B8%8B%E8%BB%8A',
|
||
'md5': '86b4e9132d158279c7883822d94ccc49',
|
||
'info_dict': {
|
||
'id': '550549',
|
||
'ext': 'mp4',
|
||
'title': '不滿被踩腳 山東兩大媽一路打下車',
|
||
'thumbnail': 're:^https?://.*\.jpg$',
|
||
'description': 'md5:175b4260c1d7c085993474217e4ab1b4',
|
||
'upload_date': '20150128',
|
||
}
|
||
}, {
|
||
'url': 'http://www.appledaily.com.tw/animation/realtimenews/new/20150128/5003671',
|
||
'md5': '03df296d95dedc2d5886debbb80cb43f',
|
||
'info_dict': {
|
||
'id': '5003671',
|
||
'ext': 'mp4',
|
||
'title': '20正妹熱舞 《刀龍傳說Online》火辣上市',
|
||
'thumbnail': 're:^https?://.*\.jpg$',
|
||
'description': 'md5:23c0aac567dc08c9c16a3161a2c2e3cd',
|
||
'upload_date': '20150128',
|
||
},
|
||
'skip': 'redirect to http://www.appledaily.com.tw/animation/',
|
||
}, {
|
||
# No thumbnail
|
||
'url': 'http://www.appledaily.com.tw/animation/realtimenews/new/20150128/5003673/',
|
||
'md5': 'b06182cd386ea7bc6115ec7ff0f72aeb',
|
||
'info_dict': {
|
||
'id': '5003673',
|
||
'ext': 'mp4',
|
||
'title': '半夜尿尿 好像會看到___',
|
||
'description': 'md5:61d2da7fe117fede148706cdb85ac066',
|
||
'upload_date': '20150128',
|
||
},
|
||
'expected_warnings': [
|
||
'video thumbnail',
|
||
],
|
||
'skip': 'redirect to http://www.appledaily.com.tw/animation/',
|
||
}, {
|
||
'url': 'http://www.appledaily.com.tw/appledaily/article/supplement/20140417/35770334/',
|
||
'md5': 'eaa20e6b9df418c912d7f5dec2ba734d',
|
||
'info_dict': {
|
||
'id': '35770334',
|
||
'ext': 'mp4',
|
||
'title': '咖啡占卜測 XU裝熟指數',
|
||
'thumbnail': 're:^https?://.*\.jpg$',
|
||
'description': 'md5:7b859991a6a4fedbdf3dd3b66545c748',
|
||
'upload_date': '20140417',
|
||
},
|
||
}]
|
||
|
||
_URL_PATTERN = r'\{url: \'(.+)\'\}'
|
||
|
||
def _fetch_title(self, page):
|
||
return (self._html_search_regex(r'<h1 id="h1">([^<>]+)</h1>', page, 'news title', default=None) or
|
||
self._html_search_meta('description', page, 'news title'))
|
||
|
||
def _fetch_thumbnail(self, page):
|
||
return self._html_search_regex(r"setInitialImage\(\'([^']+)'\)", page, 'video thumbnail', fatal=False)
|
||
|
||
def _fetch_timestamp(self, page):
|
||
return None
|
||
|
||
def _fetch_description(self, page):
|
||
return self._html_search_meta('description', page, 'news description')
|