mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-12-14 21:27:34 +01:00
[raiplay:playlist] Add extractor
This commit is contained in:
parent
a670b1ba26
commit
d21d0ba6c1
@ -857,6 +857,7 @@ from .rai import (
|
||||
RaiPlayIE,
|
||||
RaiPlayLiveIE,
|
||||
RaiIE,
|
||||
RaiPlaylistIE,
|
||||
)
|
||||
from .rbmaradio import RBMARadioIE
|
||||
from .rds import RDSIE
|
||||
|
@ -455,3 +455,29 @@ class RaiIE(RaiBaseIE):
|
||||
info.update(relinker_info)
|
||||
|
||||
return info
|
||||
|
||||
|
||||
class RaiPlaylistIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?raiplay\.it/programmi/(?P<id>[^/]+)'
|
||||
_TESTS = [{
|
||||
'url': 'http://www.raiplay.it/programmi/nondirloalmiocapo/',
|
||||
'info_dict': {
|
||||
'id': 'nondirloalmiocapo',
|
||||
'title': 'Non dirlo al mio capo',
|
||||
},
|
||||
'playlist_mincount': 12,
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
playlist_id = self._match_id(url)
|
||||
webpage = self._download_webpage(url, playlist_id)
|
||||
title = self._html_search_meta('programma', webpage, default=None)
|
||||
video_urls = re.findall(' href="(/raiplay/video.+)"', webpage)
|
||||
video_urls = [urljoin(url, video_url) for video_url in video_urls]
|
||||
entries = [
|
||||
self.url_result(
|
||||
video_url,
|
||||
RaiPlayIE.ie_key())
|
||||
for video_url in video_urls if RaiPlayIE.suitable(video_url)
|
||||
]
|
||||
return self.playlist_result(entries, playlist_id, title)
|
||||
|
Loading…
Reference in New Issue
Block a user