[tvnet] Add extractor
This commit is contained in:
parent
93cffb1444
commit
b2df66aeca
2 changed files with 92 additions and 0 deletions
|
@ -1306,6 +1306,7 @@ from .vrv import (
|
|||
VRVSeriesIE,
|
||||
)
|
||||
from .vshare import VShareIE
|
||||
from .vtv import VTVIE
|
||||
from .medialaan import MedialaanIE
|
||||
from .vube import VubeIE
|
||||
from .vuclip import VuClipIE
|
||||
|
|
91
youtube_dl/extractor/vtv.py
Normal file
91
youtube_dl/extractor/vtv.py
Normal file
|
@ -0,0 +1,91 @@
|
|||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
from .common import InfoExtractor
|
||||
|
||||
import re
|
||||
|
||||
from ..utils import extract_attributes
|
||||
|
||||
class VTVIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(au|ca|cz|de|jp|kr|tw|us|vn)\.tvnet\.gov\.vn/[^/]*/(?P<id>[0-9]+)/?'
|
||||
_TESTS = [{
|
||||
# Livestream. Channel: VTV 1
|
||||
'url': 'http://us.tvnet.gov.vn/kenh-truyen-hinh/1011/vtv1',
|
||||
'info_dict': {
|
||||
'id': '1011',
|
||||
'ext': 'mp4',
|
||||
'title': r're:^VTV1 | LiveTV - TV Net [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
|
||||
'thumbnail': r're:https?://.*\.png$',
|
||||
}
|
||||
}, {
|
||||
# Downloading a video.
|
||||
'url': 'http://de.tvnet.gov.vn/video/109788/vtv1---bac-tuyet-tai-lao-cai-va-ha-giang/tin-nong-24h',
|
||||
'md5': '5263c63d738569ed507980f1e49ebc03',
|
||||
'info_dict': {
|
||||
'id': '109788',
|
||||
'ext': 'mp4',
|
||||
'title': 'VTV1 - Bắc tuyết tại Lào Cai và Hà Giang - TV Net',
|
||||
'thumbnail': r're:https?://.*\.JPG$',
|
||||
}
|
||||
}, {
|
||||
# Radio live stream. Channel: VOV 1
|
||||
'url': 'http://vn.tvnet.gov.vn/kenh-truyen-hinh/1014',
|
||||
'info_dict': {
|
||||
'id': '1014',
|
||||
'ext': 'm4a',
|
||||
'vcodec': 'none',
|
||||
'title': r're:VOV1 | LiveTV - TV Net [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
|
||||
'thumbnail': r're:https?://.*\.png$',
|
||||
}
|
||||
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
|
||||
title = self._html_search_regex(r'<title>(.+?)</title>', webpage, 'title', default=None, fatal=False)
|
||||
if title is None:
|
||||
title = self._og_search_title(webpage)
|
||||
title.strip()
|
||||
|
||||
mediaplayer_div = self._search_regex(r'(<div[^>]*id="mediaplayer"[^>]*>)', webpage, 'mediaplayer element')
|
||||
mediaplayer_div_attributes = extract_attributes(mediaplayer_div)
|
||||
|
||||
thumbnail = mediaplayer_div_attributes.get("data-image")
|
||||
|
||||
json_url = mediaplayer_div_attributes["data-file"]
|
||||
video_streams = self._download_json(json_url, video_id)
|
||||
|
||||
|
||||
# get any working playlist from streams. Currently there's 2 and the first always works,
|
||||
# but you never know in the future
|
||||
for stream in video_streams:
|
||||
formats = self._extract_m3u8_formats(stream.get("url"), video_id, ext="mp4", fatal=False)
|
||||
if formats:
|
||||
break
|
||||
|
||||
# better support radio streams
|
||||
if title.startswith("VOV"):
|
||||
for f in formats:
|
||||
f["ext"] = "m4a"
|
||||
f["vcodec"] = "none"
|
||||
|
||||
if "/video/" in url or "/radio/" in url:
|
||||
is_live = False
|
||||
elif "/kenh-truyen-hinh/" in url:
|
||||
is_live = True
|
||||
else:
|
||||
is_live = None
|
||||
|
||||
if is_live:
|
||||
title = self._live_title(title)
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': title,
|
||||
'thumbnail': thumbnail,
|
||||
'formats': formats,
|
||||
'is_live': is_live,
|
||||
}
|
Loading…
Reference in a new issue