aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl/extractor')
-rw-r--r--youtube_dl/extractor/rts.py44
1 files changed, 36 insertions, 8 deletions
diff --git a/youtube_dl/extractor/rts.py b/youtube_dl/extractor/rts.py
index e8199b114..82d783078 100644
--- a/youtube_dl/extractor/rts.py
+++ b/youtube_dl/extractor/rts.py
@@ -15,7 +15,7 @@ from ..utils import (
class RTSIE(InfoExtractor):
IE_DESC = 'RTS.ch'
- _VALID_URL = r'^https?://(?:www\.)?rts\.ch/(?:[^/]+/){2,}(?P<id>[0-9]+)-.*?\.html'
+ _VALID_URL = r'^https?://(?:www\.)?rts\.ch/(?:(?:[^/]+/){2,}(?P<id>[0-9]+)-(?P<display_id>.+?)\.html|play/tv/-/video/(?P<display_id_new>.+?)\?id=(?P<id_new>[0-9]+))'
_TESTS = [
{
@@ -23,6 +23,7 @@ class RTSIE(InfoExtractor):
'md5': '753b877968ad8afaeddccc374d4256a5',
'info_dict': {
'id': '3449373',
+ 'display_id': 'les-enfants-terribles',
'ext': 'mp4',
'duration': 1488,
'title': 'Les Enfants Terribles',
@@ -30,7 +31,8 @@ class RTSIE(InfoExtractor):
'uploader': 'Divers',
'upload_date': '19680921',
'timestamp': -40280400,
- 'thumbnail': 're:^https?://.*\.image'
+ 'thumbnail': 're:^https?://.*\.image',
+ 'view_count': int,
},
},
{
@@ -38,6 +40,7 @@ class RTSIE(InfoExtractor):
'md5': 'c148457a27bdc9e5b1ffe081a7a8337b',
'info_dict': {
'id': '5624067',
+ 'display_id': 'entre-ciel-et-mer',
'ext': 'mp4',
'duration': 3720,
'title': 'Les yeux dans les cieux - Mon homard au Canada',
@@ -45,7 +48,8 @@ class RTSIE(InfoExtractor):
'uploader': 'Passe-moi les jumelles',
'upload_date': '20140404',
'timestamp': 1396635300,
- 'thumbnail': 're:^https?://.*\.image'
+ 'thumbnail': 're:^https?://.*\.image',
+ 'view_count': int,
},
},
{
@@ -53,6 +57,7 @@ class RTSIE(InfoExtractor):
'md5': 'b4326fecd3eb64a458ba73c73e91299d',
'info_dict': {
'id': '5745975',
+ 'display_id': '1-2-kloten-fribourg-5-2-second-but-pour-gotteron-par-kwiatowski',
'ext': 'mp4',
'duration': 48,
'title': '1/2, Kloten - Fribourg (5-2): second but pour Gottéron par Kwiatowski',
@@ -60,7 +65,8 @@ class RTSIE(InfoExtractor):
'uploader': 'Hockey',
'upload_date': '20140403',
'timestamp': 1396556882,
- 'thumbnail': 're:^https?://.*\.image'
+ 'thumbnail': 're:^https?://.*\.image',
+ 'view_count': int,
},
'skip': 'Blocked outside Switzerland',
},
@@ -69,6 +75,7 @@ class RTSIE(InfoExtractor):
'md5': '9bb06503773c07ce83d3cbd793cebb91',
'info_dict': {
'id': '5745356',
+ 'display_id': 'londres-cachee-par-un-epais-smog',
'ext': 'mp4',
'duration': 33,
'title': 'Londres cachée par un épais smog',
@@ -76,7 +83,8 @@ class RTSIE(InfoExtractor):
'uploader': 'Le Journal en continu',
'upload_date': '20140403',
'timestamp': 1396537322,
- 'thumbnail': 're:^https?://.*\.image'
+ 'thumbnail': 're:^https?://.*\.image',
+ 'view_count': int,
},
},
{
@@ -84,6 +92,7 @@ class RTSIE(InfoExtractor):
'md5': 'dd8ef6a22dff163d063e2a52bc8adcae',
'info_dict': {
'id': '5706148',
+ 'display_id': 'urban-hippie-de-damien-krisl-03-04-2014',
'ext': 'mp3',
'duration': 123,
'title': '"Urban Hippie", de Damien Krisl',
@@ -92,22 +101,40 @@ class RTSIE(InfoExtractor):
'timestamp': 1396551600,
},
},
+ {
+ 'url': 'http://www.rts.ch/play/tv/-/video/le-19h30?id=6348260',
+ 'md5': '968777c8779e5aa2434be96c54e19743',
+ 'info_dict': {
+ 'id': '6348260',
+ 'display_id': 'le-19h30',
+ 'ext': 'mp4',
+ 'duration': 1796,
+ 'title': 'Le 19h30',
+ 'description': '',
+ 'uploader': 'Le 19h30',
+ 'upload_date': '20141201',
+ 'timestamp': 1417458600,
+ 'thumbnail': 're:^https?://.*\.image',
+ 'view_count': int,
+ },
+ }
]
def _real_extract(self, url):
m = re.match(self._VALID_URL, url)
- video_id = m.group('id')
+ video_id = m.group('id') or m.group('id_new')
+ display_id = m.group('display_id') or m.group('display_id_new')
def download_json(internal_id):
return self._download_json(
'http://www.rts.ch/a/%s.html?f=json/article' % internal_id,
- video_id)
+ display_id)
all_info = download_json(video_id)
# video_id extracted out of URL is not always a real id
if 'video' not in all_info and 'audio' not in all_info:
- page = self._download_webpage(url, video_id)
+ page = self._download_webpage(url, display_id)
internal_id = self._html_search_regex(
r'<(?:video|audio) data-id="([0-9]+)"', page,
'internal video id')
@@ -143,6 +170,7 @@ class RTSIE(InfoExtractor):
return {
'id': video_id,
+ 'display_id': display_id,
'formats': formats,
'title': info['title'],
'description': info.get('intro'),