aboutsummaryrefslogtreecommitdiffstats
path: root/hypervideo_dl/extractor/lci.py
diff options
context:
space:
mode:
Diffstat (limited to 'hypervideo_dl/extractor/lci.py')
-rw-r--r--hypervideo_dl/extractor/lci.py32
1 files changed, 17 insertions, 15 deletions
diff --git a/hypervideo_dl/extractor/lci.py b/hypervideo_dl/extractor/lci.py
index 920872f..e7d2f8a 100644
--- a/hypervideo_dl/extractor/lci.py
+++ b/hypervideo_dl/extractor/lci.py
@@ -1,26 +1,28 @@
-# coding: utf-8
-from __future__ import unicode_literals
-
from .common import InfoExtractor
class LCIIE(InfoExtractor):
- _VALID_URL = r'https?://(?:www\.)?lci\.fr/[^/]+/[\w-]+-(?P<id>\d+)\.html'
- _TEST = {
- 'url': 'http://www.lci.fr/international/etats-unis-a-j-62-hillary-clinton-reste-sans-voix-2001679.html',
- 'md5': '2fdb2538b884d4d695f9bd2bde137e6c',
+ _VALID_URL = r'https?://(?:www\.)?(?:lci|tf1info)\.fr/[^/]+/[\w-]+-(?P<id>\d+)\.html'
+ _TESTS = [{
+ 'url': 'https://www.tf1info.fr/politique/election-presidentielle-2022-second-tour-j-2-marine-le-pen-et-emmanuel-macron-en-interview-de-lci-vendredi-soir-2217486.html',
'info_dict': {
- 'id': '13244802',
+ 'id': '13875948',
'ext': 'mp4',
- 'title': 'Hillary Clinton et sa quinte de toux, en plein meeting',
- 'description': 'md5:a4363e3a960860132f8124b62f4a01c9',
- }
- }
+ 'title': 'md5:660df5481fd418bc3bbb0d070e6fdb5a',
+ 'thumbnail': 'https://photos.tf1.fr/1280/720/presidentielle-2022-marine-le-pen-et-emmanuel-macron-invites-de-lci-ce-vendredi-9c0e73-e1a036-0@1x.jpg',
+ 'upload_date': '20220422',
+ 'duration': 33,
+ },
+ 'params': {
+ 'skip_download': True,
+ },
+ }, {
+ 'url': 'https://www.lci.fr/politique/election-presidentielle-2022-second-tour-j-2-marine-le-pen-et-emmanuel-macron-en-interview-de-lci-vendredi-soir-2217486.html',
+ 'only_matching': True,
+ }]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
- wat_id = self._search_regex(
- (r'data-watid=[\'"](\d+)', r'idwat["\']?\s*:\s*["\']?(\d+)'),
- webpage, 'wat id')
+ wat_id = self._search_regex(r'watId["\']?\s*:\s*["\']?(\d+)', webpage, 'wat id')
return self.url_result('wat:' + wat_id, 'Wat', wat_id)