@ -1,10 +1,14 @@
# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
int_or_none ,
ExtractorError ,
float_or_none ,
int_or_none ,
strip_or_none ,
)
@ -14,20 +18,19 @@ class RTBFIE(InfoExtractor):
( ? :
video / [ ^ ? ] + \? . * \bid = |
ouftivi / ( ? : [ ^ / ] + / ) * [ ^ ? ] + \? . * \bvideoId = |
auvio / [ ^ / ] + \? . * id =
auvio / [ ^ / ] + \? . * \b ( ? P < live > l ) ? id =
) ( ? P < id > \d + ) '''
_TESTS = [ {
' url ' : ' https://www.rtbf.be/video/detail_les-diables-au-coeur-episode-2?id=1921274 ' ,
' md5 ' : ' 799f334ddf2c0a582ba80c44655be570 ' ,
' md5 ' : ' 8c876a1cceeb6cf31b476461ade72384 ' ,
' info_dict ' : {
' id ' : ' 1921274 ' ,
' ext ' : ' mp4 ' ,
' title ' : ' Les Diables au coeur (épisode 2) ' ,
' description ' : ' Football - Diables Rouges ' ,
' duration ' : 3099 ,
' description ' : ' (du 25/04/2014) ' ,
' duration ' : 3099.54 ,
' upload_date ' : ' 20140425 ' ,
' timestamp ' : 1398456336 ,
' uploader ' : ' rtbfsport ' ,
' timestamp ' : 1398456300 ,
}
} , {
# geo restricted
@ -39,6 +42,18 @@ class RTBFIE(InfoExtractor):
} , {
' url ' : ' http://www.rtbf.be/auvio/detail_jeudi-en-prime-siegfried-bracke?id=2102996 ' ,
' only_matching ' : True ,
} , {
# Live
' url ' : ' https://www.rtbf.be/auvio/direct_pure-fm?lid=134775 ' ,
' only_matching ' : True ,
} , {
# Audio
' url ' : ' https://www.rtbf.be/auvio/detail_cinq-heures-cinema?id=2360811 ' ,
' only_matching ' : True ,
} , {
# With Subtitle
' url ' : ' https://www.rtbf.be/auvio/detail_les-carnets-du-bourlingueur?id=2361588 ' ,
' only_matching ' : True ,
} ]
_IMAGE_HOST = ' http://ds1.ds.static.rtbf.be '
_PROVIDERS = {
@ -53,46 +68,94 @@ class RTBFIE(InfoExtractor):
]
def _real_extract ( self , url ) :
video_id = self . _match_id ( url )
data = self . _download_json (
' http://www.rtbf.be/api/media/video?method=getVideoDetail&args[]= %s ' % video_id , video_id )
live , media_id = re . match ( self . _VALID_URL , url ) . groups ( )
embed_page = self . _download_webpage (
' https://www.rtbf.be/auvio/embed/ ' + ( ' direct ' if live else ' media ' ) ,
media_id , query = { ' id ' : media_id } )
data = self . _parse_json ( self . _html_search_regex (
r ' data-media= " ([^ " ]+) " ' , embed_page , ' media data ' ) , media_id )
error = data . get ( ' error ' )
if error :
raise ExtractorError ( ' %s said: %s ' % ( self . IE_NAME , error ) , expected = True )
data = data [ ' data ' ]
provider = data . get ( ' provider ' )
if provider in self . _PROVIDERS :
return self . url_result ( data [ ' url ' ] , self . _PROVIDERS [ provider ] )
title = data [ ' title ' ]
is_live = data . get ( ' isLive ' )
if is_live :
title = self . _live_title ( title )
height_re = r ' -( \ d+)p \ . '
formats = [ ]
for key , format_id in self . _QUALITIES :
format_url = data . get ( key + ' Url ' )
if format_url :
m3u8_url = data . get ( ' urlHlsAes128 ' ) or data . get ( ' urlHls ' )
if m3u8_url :
formats . extend ( self . _extract_m3u8_formats (
m3u8_url , media_id , ' mp4 ' , m3u8_id = ' hls ' , fatal = False ) )
fix_url = lambda x : x . replace ( ' //rtbf-vod. ' , ' //rtbf. ' ) if ' /geo/drm/ ' in x else x
http_url = data . get ( ' url ' )
if formats and http_url and re . search ( height_re , http_url ) :
http_url = fix_url ( http_url )
for m3u8_f in formats . copy ( ) :
height = m3u8_f . get ( ' height ' )
if not height :
continue
f = m3u8_f . copy ( )
del f [ ' protocol ' ]
f . update ( {
' format_id ' : m3u8_f [ ' format_id ' ] . replace ( ' hls- ' , ' http- ' ) ,
' url ' : re . sub ( height_re , ' - %d p. ' % height , http_url ) ,
} )
formats . append ( f )
else :
sources = data . get ( ' sources ' ) or { }
for key , format_id in self . _QUALITIES :
format_url = sources . get ( key )
if not format_url :
continue
height = int_or_none ( self . _search_regex (
height_re , format_url , ' height ' , default = None ) )
formats . append ( {
' format_id ' : format_id ,
' url ' : format_url ,
' url ' : fix_url ( format_url ) ,
' height ' : height ,
} )
thumbnails = [ ]
for thumbnail_id , thumbnail_url in data . get ( ' thumbnail ' , { } ) . items ( ) :
if thumbnail_id != ' default ' :
thumbnails . append ( {
' url ' : self . _IMAGE_HOST + thumbnail_url ,
' id ' : thumbnail_id ,
} )
mpd_url = data . get ( ' urlDash ' )
if not data . get ( ' drm ' ) and mpd_url :
formats . extend ( self . _extract_mpd_formats (
mpd_url , media_id , mpd_id = ' dash ' , fatal = False ) )
audio_url = data . get ( ' urlAudio ' )
if audio_url :
formats . append ( {
' format_id ' : ' audio ' ,
' url ' : audio_url ,
' vcodec ' : ' none ' ,
} )
self . _sort_formats ( formats )
subtitles = { }
for track in ( data . get ( ' tracks ' ) or { } ) . values ( ) :
sub_url = track . get ( ' url ' )
if not sub_url :
continue
subtitles . setdefault ( track . get ( ' lang ' ) or ' fr ' , [ ] ) . append ( {
' url ' : sub_url ,
} )
return {
' id ' : video_id ,
' id ' : media _id,
' formats ' : formats ,
' title ' : data [ ' title ' ] ,
' description ' : data . get ( ' description ' ) or data . get ( ' subtitle ' ) ,
' thumbnails ' : thumbnails ,
' duration ' : data . get ( ' duration ' ) or data . get ( ' realDuration ' ) ,
' timestamp ' : int_or_none ( data . get ( ' created ' ) ) ,
' view_count ' : int_or_none ( data . get ( ' viewCount ' ) ) ,
' uploader ' : data . get ( ' channel ' ) ,
' tags ' : data . get ( ' tags ' ) ,
' title ' : title ,
' description ' : strip_or_none ( data . get ( ' description ' ) ) ,
' thumbnail ' : data . get ( ' thumbnail' ) ,
' duration ' : float_or_none ( data . get ( ' realDuration ' ) ) ,
' timestamp ' : int_or_none ( data . get ( ' liveFrom ' ) ) ,
' series ' : data . get ( ' programLabel ' ) ,
' subtitles ' : subtitles ,
' is_live ' : is_live ,
}