[ie/bfmtv] Fix extractors (#11444)

Authored by: seproDev
pull/10843/head
sepro 2024-11-03 21:19:35 +01:00 zatwierdzone przez GitHub
rodzic beae2db127
commit 754940e9a5
Nie znaleziono w bazie danych klucza dla tego podpisu
ID klucza GPG: B5690EEEBB952194
1 zmienionych plików z 45 dodań i 21 usunięć

Wyświetl plik

@ -1,18 +1,33 @@
import re
from .common import InfoExtractor
from ..utils import extract_attributes
from ..utils import ExtractorError, extract_attributes
class BFMTVBaseIE(InfoExtractor):
_VALID_URL_BASE = r'https?://(?:www\.|rmc\.)?bfmtv\.com/'
_VALID_URL_TMPL = _VALID_URL_BASE + r'(?:[^/]+/)*[^/?&#]+_%s[A-Z]-(?P<id>\d{12})\.html'
_VIDEO_BLOCK_REGEX = r'(<div[^>]+class="video_block[^"]*"[^>]*>)'
_VIDEO_BLOCK_REGEX = r'(<div[^>]+class="video_block[^"]*"[^>]*>.*?</div>)'
_VIDEO_ELEMENT_REGEX = r'(<video-js[^>]+>)'
BRIGHTCOVE_URL_TEMPLATE = 'http://players.brightcove.net/%s/%s_default/index.html?videoId=%s'
def _brightcove_url_result(self, video_id, video_block):
account_id = video_block.get('accountid') or '876450612001'
player_id = video_block.get('playerid') or 'I2qBTln4u'
def _extract_video(self, video_block):
video_element = self._search_regex(
self._VIDEO_ELEMENT_REGEX, video_block, 'video element', default=None)
if video_element:
video_element_attrs = extract_attributes(video_element)
video_id = video_element_attrs.get('data-video-id')
if not video_id:
return
account_id = video_element_attrs.get('data-account') or '876450610001'
player_id = video_element_attrs.get('adjustplayer') or '19dszYXgm'
else:
video_block_attrs = extract_attributes(video_block)
video_id = video_block_attrs.get('videoid')
if not video_id:
return
account_id = video_block_attrs.get('accountid') or '876630703001'
player_id = video_block_attrs.get('playerid') or 'KbPwEbuHx'
return self.url_result(
self.BRIGHTCOVE_URL_TEMPLATE % (account_id, player_id, video_id),
'BrightcoveNew', video_id)
@ -40,23 +55,25 @@ class BFMTVIE(BFMTVBaseIE):
def _real_extract(self, url):
bfmtv_id = self._match_id(url)
webpage = self._download_webpage(url, bfmtv_id)
video_block = extract_attributes(self._search_regex(
video = self._extract_video(self._search_regex(
self._VIDEO_BLOCK_REGEX, webpage, 'video block'))
return self._brightcove_url_result(video_block['videoid'], video_block)
if not video:
raise ExtractorError('Failed to extract video')
return video
class BFMTVLiveIE(BFMTVIE): # XXX: Do not subclass from concrete IE
class BFMTVLiveIE(BFMTVBaseIE):
IE_NAME = 'bfmtv:live'
_VALID_URL = BFMTVBaseIE._VALID_URL_BASE + '(?P<id>(?:[^/]+/)?en-direct)'
_TESTS = [{
'url': 'https://www.bfmtv.com/en-direct/',
'info_dict': {
'id': '5615950982001',
'id': '6346069778112',
'ext': 'mp4',
'title': r're:^le direct BFMTV WEB \d{4}-\d{2}-\d{2} \d{2}:\d{2}$',
'title': r're:^Le Live BFM TV \d{4}-\d{2}-\d{2} \d{2}:\d{2}$',
'uploader_id': '876450610001',
'upload_date': '20220926',
'timestamp': 1664207191,
'upload_date': '20240202',
'timestamp': 1706887572,
'live_status': 'is_live',
'thumbnail': r're:https://.+/image\.jpg',
'tags': [],
@ -69,6 +86,15 @@ class BFMTVLiveIE(BFMTVIE): # XXX: Do not subclass from concrete IE
'only_matching': True,
}]
def _real_extract(self, url):
bfmtv_id = self._match_id(url)
webpage = self._download_webpage(url, bfmtv_id)
video = self._extract_video(self._search_regex(
self._VIDEO_BLOCK_REGEX, webpage, 'video block'))
if not video:
raise ExtractorError('Failed to extract video')
return video
class BFMTVArticleIE(BFMTVBaseIE):
IE_NAME = 'bfmtv:article'
@ -102,18 +128,16 @@ class BFMTVArticleIE(BFMTVBaseIE):
},
}]
def _entries(self, webpage):
for video_block_el in re.findall(self._VIDEO_BLOCK_REGEX, webpage):
video = self._extract_video(video_block_el)
if video:
yield video
def _real_extract(self, url):
bfmtv_id = self._match_id(url)
webpage = self._download_webpage(url, bfmtv_id)
entries = []
for video_block_el in re.findall(self._VIDEO_BLOCK_REGEX, webpage):
video_block = extract_attributes(video_block_el)
video_id = video_block.get('videoid')
if not video_id:
continue
entries.append(self._brightcove_url_result(video_id, video_block))
return self.playlist_result(
entries, bfmtv_id, self._og_search_title(webpage, fatal=False),
self._entries(webpage), bfmtv_id, self._og_search_title(webpage, fatal=False),
self._html_search_meta(['og:description', 'description'], webpage))