From fd8102820c4d14fdb1ff7e090553211717012f67 Mon Sep 17 00:00:00 2001 From: Yen Chi Hsuan Date: Wed, 4 Nov 2015 00:09:55 +0800 Subject: [PATCH] [democracynow] Rename js to json_data --- youtube_dl/extractor/democracynow.py | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/youtube_dl/extractor/democracynow.py b/youtube_dl/extractor/democracynow.py index 70c364e8b..72fc75d80 100644 --- a/youtube_dl/extractor/democracynow.py +++ b/youtube_dl/extractor/democracynow.py @@ -40,7 +40,7 @@ class DemocracynowIE(InfoExtractor): webpage = self._download_webpage(url, display_id) description = self._og_search_description(webpage) - js = self._parse_json(self._search_regex( + json_data = self._parse_json(self._search_regex( r']+type="text/json"[^>]*>\s*({[^>]+})', webpage, 'json'), display_id) video_id = None @@ -56,19 +56,19 @@ class DemocracynowIE(InfoExtractor): subtitles[lang].append(info_dict) # chapter_file are not subtitles - if 'caption_file' in js: + if 'caption_file' in json_data: add_subtitle_item(default_lang, { - 'url': compat_urlparse.urljoin(url, js['caption_file']), + 'url': compat_urlparse.urljoin(url, json_data['caption_file']), }) - for subtitle_item in js.get('captions', []): + for subtitle_item in json_data.get('captions', []): lang = subtitle_item.get('language', '').lower() or default_lang add_subtitle_item(lang, { 'url': compat_urlparse.urljoin(url, subtitle_item['url']), }) for key in ('file', 'audio', 'video'): - media_url = js.get(key, '') + media_url = json_data.get(key, '') if not media_url: continue media_url = re.sub(r'\?.*', '', compat_urlparse.urljoin(url, media_url)) @@ -81,7 +81,7 @@ class DemocracynowIE(InfoExtractor): return { 'id': video_id, - 'title': js.get('title'), + 'title': json_data.get('title'), 'description': description, 'subtitles': subtitles, 'formats': formats,