[ora] Revert extraction to regexes

It's less fragile than using js_to_json with ora js
This commit is contained in:
Sergey M․ 2016-05-14 20:45:18 +06:00
parent 89ac4a19e6
commit ca950f49e9
No known key found for this signature in database
GPG Key ID: 2C393E0F18A9236D

View File

@ -6,15 +6,14 @@
from ..compat import compat_urlparse
from ..utils import (
get_element_by_attribute,
js_to_json,
qualities,
unescapeHTML,
)
class OraTVIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?(ora\.tv|unsafespeech\.com)/([^/]+/)*(?P<id>[^/\?#]+)'
_TEST = {
_VALID_URL = r'https?://(?:www\.)?(?:ora\.tv|unsafespeech\.com)/([^/]+/)*(?P<id>[^/\?#]+)'
_TESTS = [{
'url': 'https://www.ora.tv/larrykingnow/2015/12/16/vine-youtube-stars-zach-king-king-bach-on-their-viral-videos-0_36jupg6090pq',
'md5': 'fa33717591c631ec93b04b0e330df786',
'info_dict': {
@ -23,19 +22,19 @@ class OraTVIE(InfoExtractor):
'title': 'Vine & YouTube Stars Zach King & King Bach On Their Viral Videos!',
'description': 'md5:ebbc5b1424dd5dba7be7538148287ac1',
}
}
}, {
'url': 'http://www.unsafespeech.com/video/2016/5/10/student-self-censorship-and-the-thought-police-on-university-campuses-0_6622bnkppw4d',
'only_matching': True,
}]
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
ora_meta = self._parse_json(self._search_regex(
r'(?s);\s*ora_meta = ({.*?});</script>', webpage, 'ora_meta'), display_id,
transform_source=lambda data: js_to_json(re.sub('":(document|\().*?(:false|\(\)),', '":null,', data)))
video_data = ora_meta.get('video', ora_meta.get('current'))
m3u8_url = video_data['hls_stream']
video_data = self._search_regex(
r'"(?:video|current)"\s*:\s*({[^}]+?})', webpage, 'current video')
m3u8_url = self._search_regex(
r'hls_stream"?\s*:\s*"([^"]+)', video_data, 'm3u8 url', None)
if m3u8_url:
formats = self._extract_m3u8_formats(
m3u8_url, display_id, 'mp4', 'm3u8_native',
@ -64,11 +63,13 @@ def _real_extract(self, url):
r'"youtube_id"\s*:\s*"([^"]+)', webpage, 'youtube id'), 'Youtube')
return {
'id': video_data.get('id', display_id),
'id': self._search_regex(
r'"id"\s*:\s*(\d+)', video_data, 'video id', default=display_id),
'display_id': display_id,
'title': unescapeHTML(self._og_search_title(webpage)),
'description': get_element_by_attribute(
'class', 'video_txt_decription', webpage),
'thumbnail': self._proto_relative_url(video_data.get('thumb')),
'thumbnail': self._proto_relative_url(self._search_regex(
r'"thumb"\s*:\s*"([^"]+)', video_data, 'thumbnail', None)),
'formats': formats,
}