mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-16 06:15:46 +00:00
[WDR] use single quotes for strings
This commit is contained in:
parent
14f7a2b8af
commit
3874e6ea66
1 changed files with 12 additions and 12 deletions
|
@ -17,7 +17,7 @@ from ..utils import (
|
||||||
class WDRIE(InfoExtractor):
|
class WDRIE(InfoExtractor):
|
||||||
_CURRENT_MAUS_URL = r'https?://www.wdrmaus.de/aktuelle-sendung/(wdr|index).php5'
|
_CURRENT_MAUS_URL = r'https?://www.wdrmaus.de/aktuelle-sendung/(wdr|index).php5'
|
||||||
_PAGE_REGEX = r'/mediathek/(?P<media_type>[^/]+)/(?P<type>[^/]+)/(?P<display_id>.+)\.html'
|
_PAGE_REGEX = r'/mediathek/(?P<media_type>[^/]+)/(?P<type>[^/]+)/(?P<display_id>.+)\.html'
|
||||||
_VALID_URL = r'(?P<page_url>https?://(?:www\d\.)?wdr\d?\.de)' + _PAGE_REGEX + "|" + _CURRENT_MAUS_URL
|
_VALID_URL = r'(?P<page_url>https?://(?:www\d\.)?wdr\d?\.de)' + _PAGE_REGEX + '|' + _CURRENT_MAUS_URL
|
||||||
|
|
||||||
_JS_URL_REGEX = r'(https?://deviceids-medp.wdr.de/ondemand/\d+/\d+\.js)'
|
_JS_URL_REGEX = r'(https?://deviceids-medp.wdr.de/ondemand/\d+/\d+\.js)'
|
||||||
|
|
||||||
|
@ -116,23 +116,23 @@ class WDRIE(InfoExtractor):
|
||||||
json_data = self._search_regex(r'\(({.*})\)', js_data, 'json')
|
json_data = self._search_regex(r'\(({.*})\)', js_data, 'json')
|
||||||
metadata = self._parse_json(json_data, display_id)
|
metadata = self._parse_json(json_data, display_id)
|
||||||
|
|
||||||
metadata_tracker_data = metadata["trackerData"]
|
metadata_tracker_data = metadata['trackerData']
|
||||||
metadata_media_resource = metadata["mediaResource"]
|
metadata_media_resource = metadata['mediaResource']
|
||||||
|
|
||||||
formats = []
|
formats = []
|
||||||
|
|
||||||
# check if the metadata contains a direct URL to a file
|
# check if the metadata contains a direct URL to a file
|
||||||
metadata_media_alt = metadata_media_resource.get("alt")
|
metadata_media_alt = metadata_media_resource.get('alt')
|
||||||
if metadata_media_alt:
|
if metadata_media_alt:
|
||||||
for tag_name in ["videoURL", 'audioURL']:
|
for tag_name in ['videoURL', 'audioURL']:
|
||||||
if tag_name in metadata_media_alt:
|
if tag_name in metadata_media_alt:
|
||||||
formats.append({
|
formats.append({
|
||||||
'url': metadata_media_alt[tag_name]
|
'url': metadata_media_alt[tag_name]
|
||||||
})
|
})
|
||||||
|
|
||||||
# check if there are flash-streams for this video
|
# check if there are flash-streams for this video
|
||||||
if "dflt" in metadata_media_resource and "videoURL" in metadata_media_resource["dflt"]:
|
if 'dflt' in metadata_media_resource and 'videoURL' in metadata_media_resource['dflt']:
|
||||||
video_url = metadata_media_resource["dflt"]["videoURL"]
|
video_url = metadata_media_resource['dflt']['videoURL']
|
||||||
if video_url.endswith('.f4m'):
|
if video_url.endswith('.f4m'):
|
||||||
full_video_url = video_url + '?hdcore=3.2.0&plugin=aasp-3.2.0.77.18'
|
full_video_url = video_url + '?hdcore=3.2.0&plugin=aasp-3.2.0.77.18'
|
||||||
formats.extend(self._extract_f4m_formats(full_video_url, display_id, f4m_id='hds', fatal=False))
|
formats.extend(self._extract_f4m_formats(full_video_url, display_id, f4m_id='hds', fatal=False))
|
||||||
|
@ -140,13 +140,13 @@ class WDRIE(InfoExtractor):
|
||||||
formats.extend(self._extract_smil_formats(video_url, 'stream', fatal=False))
|
formats.extend(self._extract_smil_formats(video_url, 'stream', fatal=False))
|
||||||
|
|
||||||
subtitles = {}
|
subtitles = {}
|
||||||
caption_url = metadata_media_resource.get("captionURL")
|
caption_url = metadata_media_resource.get('captionURL')
|
||||||
if caption_url:
|
if caption_url:
|
||||||
subtitles['de'] = [{
|
subtitles['de'] = [{
|
||||||
'url': caption_url
|
'url': caption_url
|
||||||
}]
|
}]
|
||||||
|
|
||||||
title = metadata_tracker_data.get("trackerClipTitle")
|
title = metadata_tracker_data.get('trackerClipTitle')
|
||||||
is_live = url_type == 'live'
|
is_live = url_type == 'live'
|
||||||
|
|
||||||
if is_live:
|
if is_live:
|
||||||
|
@ -163,13 +163,13 @@ class WDRIE(InfoExtractor):
|
||||||
self._sort_formats(formats)
|
self._sort_formats(formats)
|
||||||
|
|
||||||
return {
|
return {
|
||||||
'id': metadata_tracker_data.get("trackerClipId", display_id),
|
'id': metadata_tracker_data.get('trackerClipId', display_id),
|
||||||
'display_id': display_id,
|
'display_id': display_id,
|
||||||
'title': title,
|
'title': title,
|
||||||
'alt_title': metadata_tracker_data.get("trackerClipSubcategory"),
|
'alt_title': metadata_tracker_data.get('trackerClipSubcategory'),
|
||||||
'formats': formats,
|
'formats': formats,
|
||||||
'upload_date': upload_date,
|
'upload_date': upload_date,
|
||||||
'description': self._html_search_meta("Description", webpage),
|
'description': self._html_search_meta('Description', webpage),
|
||||||
'is_live': is_live,
|
'is_live': is_live,
|
||||||
'subtitles': subtitles,
|
'subtitles': subtitles,
|
||||||
}
|
}
|
||||||
|
|
Loading…
Reference in a new issue