2014-02-24 14:00:31 +01:00
from __future__ import unicode_literals
2018-02-09 02:05:12 +01:00
import base64
2018-05-24 03:51:47 +02:00
import json
import re
2013-08-27 12:38:30 +02:00
from . common import InfoExtractor
2016-03-17 16:07:36 +01:00
from . theplatform import ThePlatformIE
2017-02-08 16:23:42 +01:00
from . adobepass import AdobePassIE
2018-09-08 09:44:06 +02:00
from . . compat import compat_urllib_parse_unquote
2014-12-13 12:24:42 +01:00
from . . utils import (
2019-10-28 19:31:20 +01:00
int_or_none ,
2021-03-21 16:56:31 +01:00
parse_age_limit ,
2019-10-28 19:31:20 +01:00
parse_duration ,
2021-08-04 06:19:44 +02:00
RegexNotFoundError ,
2015-12-08 16:16:14 +01:00
smuggle_url ,
2019-10-28 19:31:20 +01:00
try_get ,
unified_timestamp ,
2016-12-17 18:11:13 +01:00
update_url_query ,
2014-07-23 01:47:18 +02:00
)
2013-08-27 12:38:30 +02:00
2021-03-21 16:56:31 +01:00
class NBCIE ( ThePlatformIE ) :
2017-10-25 18:23:27 +02:00
_VALID_URL = r ' https?(?P<permalink>://(?:www \ .)?nbc \ .com/(?:classic-tv/)?[^/]+/video/[^/]+/(?P<id>n? \ d+)) '
2014-12-12 17:10:32 +01:00
_TESTS = [
{
2017-02-08 16:23:42 +01:00
' url ' : ' http://www.nbc.com/the-tonight-show/video/jimmy-fallon-surprises-fans-at-ben-jerrys/2848237 ' ,
2014-12-12 17:10:32 +01:00
' info_dict ' : {
2017-02-08 16:23:42 +01:00
' id ' : ' 2848237 ' ,
2016-01-30 21:34:46 +01:00
' ext ' : ' mp4 ' ,
2015-02-19 15:00:39 +01:00
' title ' : ' Jimmy Fallon Surprises Fans at Ben & Jerry \' s ' ,
' description ' : ' Jimmy gives out free scoops of his new " Tonight Dough " ice cream flavor by surprising customers at the Ben & Jerry \' s scoop shop. ' ,
2016-04-01 19:06:11 +02:00
' timestamp ' : 1424246400 ,
' upload_date ' : ' 20150218 ' ,
' uploader ' : ' NBCU-COM ' ,
2014-12-12 17:10:32 +01:00
} ,
2016-01-30 21:34:46 +01:00
' params ' : {
# m3u8 download
' skip_download ' : True ,
} ,
2014-02-25 23:57:54 +01:00
} ,
2015-05-04 13:09:18 +02:00
{
' url ' : ' http://www.nbc.com/saturday-night-live/video/star-wars-teaser/2832821 ' ,
' info_dict ' : {
2016-01-30 21:34:46 +01:00
' id ' : ' 2832821 ' ,
' ext ' : ' mp4 ' ,
2015-05-04 13:09:18 +02:00
' title ' : ' Star Wars Teaser ' ,
' description ' : ' md5:0b40f9cbde5b671a7ff62fceccc4f442 ' ,
2016-04-01 19:06:11 +02:00
' timestamp ' : 1417852800 ,
' upload_date ' : ' 20141206 ' ,
' uploader ' : ' NBCU-COM ' ,
2015-05-04 13:09:18 +02:00
} ,
2016-01-30 21:34:46 +01:00
' params ' : {
# m3u8 download
' skip_download ' : True ,
} ,
2015-05-04 13:09:18 +02:00
' skip ' : ' Only works from US ' ,
2015-05-04 15:53:05 +02:00
} ,
2016-06-08 14:50:01 +02:00
{
# HLS streams requires the 'hdnea3' cookie
' url ' : ' http://www.nbc.com/Kings/video/goliath/n1806 ' ,
' info_dict ' : {
2017-02-08 16:23:42 +01:00
' id ' : ' 101528f5a9e8127b107e98c5e6ce4638 ' ,
2016-06-08 14:50:01 +02:00
' ext ' : ' mp4 ' ,
' title ' : ' Goliath ' ,
' description ' : ' When an unknown soldier saves the life of the King \' s son in battle, he \' s thrust into the limelight and politics of the kingdom. ' ,
' timestamp ' : 1237100400 ,
' upload_date ' : ' 20090315 ' ,
' uploader ' : ' NBCU-COM ' ,
} ,
' params ' : {
' skip_download ' : True ,
} ,
' skip ' : ' Only works from US ' ,
2017-10-25 18:23:27 +02:00
} ,
{
' url ' : ' https://www.nbc.com/classic-tv/charles-in-charge/video/charles-in-charge-pilot/n3310 ' ,
' only_matching ' : True ,
} ,
2018-09-08 09:44:06 +02:00
{
# Percent escaped url
' url ' : ' https://www.nbc.com/up-all-night/video/day-after-valentine %27s -day/n2189 ' ,
' only_matching ' : True ,
}
2014-12-12 17:10:32 +01:00
]
2014-02-25 23:57:54 +01:00
def _real_extract ( self , url ) :
2017-05-07 10:31:14 +02:00
permalink , video_id = re . match ( self . _VALID_URL , url ) . groups ( )
2018-09-08 09:44:06 +02:00
permalink = ' http ' + compat_urllib_parse_unquote ( permalink )
2020-01-16 15:37:16 +01:00
video_data = self . _download_json (
2019-10-14 19:36:25 +02:00
' https://friendship.nbc.co/v2/graphql ' , video_id , query = {
2020-01-16 15:37:16 +01:00
' query ' : ''' query bonanzaPage(
$ app : NBCUBrands ! = nbc
$ name : String !
$ oneApp : Boolean
$ platform : SupportedPlatforms ! = web
$ type : EntityPageType ! = VIDEO
$ userId : String !
) {
bonanzaPage (
app : $ app
name : $ name
oneApp : $ oneApp
platform : $ platform
type : $ type
userId : $ userId
) {
metadata {
2019-10-14 19:36:25 +02:00
. . . on VideoPageData {
description
episodeNumber
keywords
locked
mpxAccountId
mpxGuid
rating
2020-01-16 15:37:16 +01:00
resourceId
2019-10-14 19:36:25 +02:00
seasonNumber
secondaryTitle
seriesShortTitle
}
}
}
2020-01-16 15:37:16 +01:00
} ''' ,
' variables ' : json . dumps ( {
' name ' : permalink ,
' oneApp ' : True ,
' userId ' : ' 0 ' ,
} ) ,
} ) [ ' data ' ] [ ' bonanzaPage ' ] [ ' metadata ' ]
2017-05-07 09:58:34 +02:00
query = {
' mbr ' : ' true ' ,
' manifest ' : ' m3u ' ,
}
2019-10-14 19:36:25 +02:00
video_id = video_data [ ' mpxGuid ' ]
2021-03-21 16:56:31 +01:00
tp_path = ' NnzsPC/media/guid/ %s / %s ' % ( video_data . get ( ' mpxAccountId ' ) or ' 2410887629 ' , video_id )
tpm = self . _download_theplatform_metadata ( tp_path , video_id )
title = tpm . get ( ' title ' ) or video_data . get ( ' secondaryTitle ' )
2019-10-14 19:36:25 +02:00
if video_data . get ( ' locked ' ) :
2017-05-07 09:58:34 +02:00
resource = self . _get_mvpd_resource (
2020-01-16 15:37:16 +01:00
video_data . get ( ' resourceId ' ) or ' nbcentertainment ' ,
title , video_id , video_data . get ( ' rating ' ) )
2017-05-07 09:58:34 +02:00
query [ ' auth ' ] = self . _extract_mvpd_auth (
url , video_id , ' nbcentertainment ' , resource )
theplatform_url = smuggle_url ( update_url_query (
2019-10-14 19:36:25 +02:00
' http://link.theplatform.com/s/NnzsPC/media/guid/ %s / %s ' % ( video_data . get ( ' mpxAccountId ' ) or ' 2410887629 ' , video_id ) ,
2017-05-07 09:58:34 +02:00
query ) , { ' force_smil_url ' : True } )
2021-03-21 16:56:31 +01:00
# Empty string or 0 can be valid values for these. So the check must be `is None`
description = video_data . get ( ' description ' )
if description is None :
description = tpm . get ( ' description ' )
episode_number = int_or_none ( video_data . get ( ' episodeNumber ' ) )
if episode_number is None :
episode_number = int_or_none ( tpm . get ( ' nbcu$airOrder ' ) )
rating = video_data . get ( ' rating ' )
if rating is None :
try_get ( tpm , lambda x : x [ ' ratings ' ] [ 0 ] [ ' rating ' ] )
season_number = int_or_none ( video_data . get ( ' seasonNumber ' ) )
if season_number is None :
season_number = int_or_none ( tpm . get ( ' nbcu$seasonNumber ' ) )
series = video_data . get ( ' seriesShortTitle ' )
if series is None :
series = tpm . get ( ' nbcu$seriesShortTitle ' )
tags = video_data . get ( ' keywords ' )
if tags is None or len ( tags ) == 0 :
tags = tpm . get ( ' keywords ' )
2017-05-07 09:58:34 +02:00
return {
2016-01-30 21:34:46 +01:00
' _type ' : ' url_transparent ' ,
2021-03-21 16:56:31 +01:00
' age_limit ' : parse_age_limit ( rating ) ,
' description ' : description ,
' episode ' : title ,
' episode_number ' : episode_number ,
2016-01-30 21:34:46 +01:00
' id ' : video_id ,
2021-03-21 16:56:31 +01:00
' ie_key ' : ' ThePlatform ' ,
' season_number ' : season_number ,
' series ' : series ,
' tags ' : tags ,
2017-05-07 09:58:34 +02:00
' title ' : title ,
' url ' : theplatform_url ,
2016-01-30 21:34:46 +01:00
}
2014-02-25 23:57:54 +01:00
2015-03-30 20:47:18 +02:00
class NBCSportsVPlayerIE ( InfoExtractor ) :
2021-01-01 13:26:37 +01:00
_VALID_URL_BASE = r ' https?://(?:vplayer \ .nbcsports \ .com|(?:www \ .)?nbcsports \ .com/vplayer)/ '
_VALID_URL = _VALID_URL_BASE + r ' (?:[^/]+/)+(?P<id>[0-9a-zA-Z_]+) '
2015-03-30 20:21:27 +02:00
2015-03-30 21:38:45 +02:00
_TESTS = [ {
2017-09-20 21:53:06 +02:00
' url ' : ' https://vplayer.nbcsports.com/p/BxmELC/nbcsports_embed/select/9CsDKds0kvHI ' ,
2015-03-30 20:21:27 +02:00
' info_dict ' : {
' id ' : ' 9CsDKds0kvHI ' ,
2017-09-20 21:53:06 +02:00
' ext ' : ' mp4 ' ,
2015-03-30 20:21:27 +02:00
' description ' : ' md5:df390f70a9ba7c95ff1daace988f0d8d ' ,
' title ' : ' Tyler Kalinoski hits buzzer-beater to lift Davidson ' ,
2016-04-01 19:06:11 +02:00
' timestamp ' : 1426270238 ,
' upload_date ' : ' 20150313 ' ,
' uploader ' : ' NBCU-SPORTS ' ,
2015-03-30 20:21:27 +02:00
}
2015-03-30 21:38:45 +02:00
} , {
2017-09-20 21:53:06 +02:00
' url ' : ' https://vplayer.nbcsports.com/p/BxmELC/nbcsports_embed/select/media/_hqLjQ95yx8Z ' ,
2015-03-30 21:38:45 +02:00
' only_matching ' : True ,
2021-01-01 13:26:37 +01:00
} , {
' url ' : ' https://www.nbcsports.com/vplayer/p/BxmELC/nbcsports/select/PHJSaFWbrTY9?form=html&autoPlay=true ' ,
' only_matching ' : True ,
2015-03-30 21:38:45 +02:00
} ]
2015-03-30 20:21:27 +02:00
2015-03-30 20:47:18 +02:00
@staticmethod
def _extract_url ( webpage ) :
iframe_m = re . search (
2021-01-01 13:26:37 +01:00
r ' <(?:iframe[^>]+|div[^>]+data-(?:mpx-)?)src= " (?P<url> %s [^ " ]+) " ' % NBCSportsVPlayerIE . _VALID_URL_BASE , webpage )
2015-03-30 20:47:18 +02:00
if iframe_m :
return iframe_m . group ( ' url ' )
2015-03-30 20:21:27 +02:00
def _real_extract ( self , url ) :
video_id = self . _match_id ( url )
webpage = self . _download_webpage ( url , video_id )
2017-09-20 21:53:06 +02:00
theplatform_url = self . _og_search_video_url ( webpage ) . replace (
' vplayer.nbcsports.com ' , ' player.theplatform.com ' )
2015-03-30 20:21:27 +02:00
return self . url_result ( theplatform_url , ' ThePlatform ' )
2015-03-30 20:47:18 +02:00
class NBCSportsIE ( InfoExtractor ) :
2021-01-01 13:26:37 +01:00
_VALID_URL = r ' https?://(?:www \ .)?nbcsports \ .com//?(?!vplayer/)(?:[^/]+/)+(?P<id>[0-9a-z-]+) '
2015-03-30 20:47:18 +02:00
2021-01-01 13:26:37 +01:00
_TESTS = [ {
# iframe src
2015-03-30 20:47:18 +02:00
' url ' : ' http://www.nbcsports.com//college-basketball/ncaab/tom-izzo-michigan-st-has-so-much-respect-duke ' ,
' info_dict ' : {
' id ' : ' PHJSaFWbrTY9 ' ,
2021-01-01 13:26:37 +01:00
' ext ' : ' mp4 ' ,
2015-03-30 20:47:18 +02:00
' title ' : ' Tom Izzo, Michigan St. has \' so much respect \' for Duke ' ,
' description ' : ' md5:ecb459c9d59e0766ac9c7d5d0eda8113 ' ,
2016-04-24 14:44:52 +02:00
' uploader ' : ' NBCU-SPORTS ' ,
' upload_date ' : ' 20150330 ' ,
' timestamp ' : 1427726529 ,
2015-03-30 20:47:18 +02:00
}
2021-01-01 13:26:37 +01:00
} , {
# data-mpx-src
' url ' : ' https://www.nbcsports.com/philadelphia/philadelphia-phillies/bruce-bochy-hector-neris-hes-idiot ' ,
' only_matching ' : True ,
} , {
# data-src
' url ' : ' https://www.nbcsports.com/boston/video/report-card-pats-secondary-no-match-josh-allen ' ,
' only_matching ' : True ,
} ]
2015-03-30 20:47:18 +02:00
def _real_extract ( self , url ) :
video_id = self . _match_id ( url )
webpage = self . _download_webpage ( url , video_id )
return self . url_result (
NBCSportsVPlayerIE . _extract_url ( webpage ) , ' NBCSportsVPlayer ' )
2018-05-24 03:51:47 +02:00
class NBCSportsStreamIE ( AdobePassIE ) :
_VALID_URL = r ' https?://stream \ .nbcsports \ .com/.+? \ bpid=(?P<id> \ d+) '
_TEST = {
' url ' : ' http://stream.nbcsports.com/nbcsn/generic?pid=206559 ' ,
' info_dict ' : {
' id ' : ' 206559 ' ,
' ext ' : ' mp4 ' ,
' title ' : ' Amgen Tour of California Women \' s Recap ' ,
' description ' : ' md5:66520066b3b5281ada7698d0ea2aa894 ' ,
} ,
' params ' : {
# m3u8 download
' skip_download ' : True ,
} ,
' skip ' : ' Requires Adobe Pass Authentication ' ,
}
def _real_extract ( self , url ) :
video_id = self . _match_id ( url )
live_source = self . _download_json (
' http://stream.nbcsports.com/data/live_sources_ %s .json ' % video_id ,
video_id )
video_source = live_source [ ' videoSources ' ] [ 0 ]
title = video_source [ ' title ' ]
source_url = None
for k in ( ' source ' , ' msl4source ' , ' iossource ' , ' hlsv4 ' ) :
sk = k + ' Url '
source_url = video_source . get ( sk ) or video_source . get ( sk + ' Alt ' )
if source_url :
break
else :
source_url = video_source [ ' ottStreamUrl ' ]
is_live = video_source . get ( ' type ' ) == ' live ' or video_source . get ( ' status ' ) == ' Live '
resource = self . _get_mvpd_resource ( ' nbcsports ' , title , video_id , ' ' )
token = self . _extract_mvpd_auth ( url , video_id , ' nbcsports ' , resource )
tokenized_url = self . _download_json (
' https://token.playmakerservices.com/cdn ' ,
video_id , data = json . dumps ( {
' requestorId ' : ' nbcsports ' ,
' pid ' : video_id ,
' application ' : ' NBCSports ' ,
' version ' : ' v1 ' ,
' platform ' : ' desktop ' ,
' cdn ' : ' akamai ' ,
' url ' : video_source [ ' sourceUrl ' ] ,
' token ' : base64 . b64encode ( token . encode ( ) ) . decode ( ) ,
' resourceId ' : base64 . b64encode ( resource . encode ( ) ) . decode ( ) ,
} ) . encode ( ) ) [ ' tokenizedUrl ' ]
formats = self . _extract_m3u8_formats ( tokenized_url , video_id , ' mp4 ' )
self . _sort_formats ( formats )
return {
' id ' : video_id ,
' title ' : self . _live_title ( title ) if is_live else title ,
' description ' : live_source . get ( ' description ' ) ,
' formats ' : formats ,
' is_live ' : is_live ,
}
2016-03-17 16:07:36 +01:00
class NBCNewsIE ( ThePlatformIE ) :
2018-11-21 12:10:06 +01:00
_VALID_URL = r ' (?x)https?://(?:www \ .)?(?:nbcnews|today|msnbc) \ .com/([^/]+/)*(?:.*-)?(?P<id>[^/?]+) '
2013-08-27 12:38:30 +02:00
2014-05-30 00:38:57 +02:00
_TESTS = [
{
2016-03-17 16:07:36 +01:00
' url ' : ' http://www.nbcnews.com/watch/nbcnews-com/how-twitter-reacted-to-the-snowden-interview-269389891880 ' ,
2019-10-28 19:31:20 +01:00
' md5 ' : ' cf4bc9e6ce0130f00f545d80ecedd4bf ' ,
2014-05-30 00:38:57 +02:00
' info_dict ' : {
2018-11-21 12:10:06 +01:00
' id ' : ' 269389891880 ' ,
2014-10-27 01:14:17 +01:00
' ext ' : ' mp4 ' ,
2014-05-30 00:38:57 +02:00
' title ' : ' How Twitter Reacted To The Snowden Interview ' ,
' description ' : ' md5:65a0bd5d76fe114f3c2727aa3a81fe64 ' ,
2016-06-23 02:36:19 +02:00
' timestamp ' : 1401363060 ,
' upload_date ' : ' 20140529 ' ,
2014-05-30 00:38:57 +02:00
} ,
} ,
2015-01-20 16:23:51 +01:00
{
' url ' : ' http://www.nbcnews.com/feature/dateline-full-episodes/full-episode-family-business-n285156 ' ,
' md5 ' : ' fdbf39ab73a72df5896b6234ff98518a ' ,
' info_dict ' : {
2016-06-23 02:36:19 +02:00
' id ' : ' 529953347624 ' ,
2015-01-20 16:23:51 +01:00
' ext ' : ' mp4 ' ,
' title ' : ' FULL EPISODE: Family Business ' ,
' description ' : ' md5:757988edbaae9d7be1d585eb5d55cc04 ' ,
} ,
2016-03-17 16:07:36 +01:00
' skip ' : ' This page is unavailable. ' ,
2015-01-20 16:23:51 +01:00
} ,
2015-02-14 11:10:23 +01:00
{
' url ' : ' http://www.nbcnews.com/nightly-news/video/nightly-news-with-brian-williams-full-broadcast-february-4-394064451844 ' ,
2019-10-28 19:31:20 +01:00
' md5 ' : ' 8eb831eca25bfa7d25ddd83e85946548 ' ,
2015-02-14 11:10:23 +01:00
' info_dict ' : {
2018-11-21 12:10:06 +01:00
' id ' : ' 394064451844 ' ,
2015-02-14 11:10:23 +01:00
' ext ' : ' mp4 ' ,
' title ' : ' Nightly News with Brian Williams Full Broadcast (February 4) ' ,
' description ' : ' md5:1c10c1eccbe84a26e5debb4381e2d3c5 ' ,
2016-06-23 02:36:19 +02:00
' timestamp ' : 1423104900 ,
' upload_date ' : ' 20150205 ' ,
2015-02-14 11:10:23 +01:00
} ,
} ,
2016-03-17 16:07:36 +01:00
{
' url ' : ' http://www.nbcnews.com/business/autos/volkswagen-11-million-vehicles-could-have-suspect-software-emissions-scandal-n431456 ' ,
2019-10-28 19:31:20 +01:00
' md5 ' : ' 4a8c4cec9e1ded51060bdda36ff0a5c0 ' ,
2016-03-17 16:07:36 +01:00
' info_dict ' : {
2019-10-28 19:31:20 +01:00
' id ' : ' n431456 ' ,
2016-03-17 16:07:36 +01:00
' ext ' : ' mp4 ' ,
2019-10-28 19:31:20 +01:00
' title ' : " Volkswagen U.S. Chief: We ' Totally Screwed Up ' " ,
' description ' : ' md5:d22d1281a24f22ea0880741bb4dd6301 ' ,
2016-06-23 02:36:19 +02:00
' upload_date ' : ' 20150922 ' ,
' timestamp ' : 1442917800 ,
2016-03-17 16:07:36 +01:00
} ,
} ,
2016-04-22 19:07:40 +02:00
{
' url ' : ' http://www.today.com/video/see-the-aurora-borealis-from-space-in-stunning-new-nasa-video-669831235788 ' ,
' md5 ' : ' 118d7ca3f0bea6534f119c68ef539f71 ' ,
' info_dict ' : {
2018-11-21 12:10:06 +01:00
' id ' : ' 669831235788 ' ,
2016-04-22 19:07:40 +02:00
' ext ' : ' mp4 ' ,
' title ' : ' See the aurora borealis from space in stunning new NASA video ' ,
' description ' : ' md5:74752b7358afb99939c5f8bb2d1d04b1 ' ,
' upload_date ' : ' 20160420 ' ,
' timestamp ' : 1461152093 ,
2016-06-23 02:36:19 +02:00
} ,
} ,
{
' url ' : ' http://www.msnbc.com/all-in-with-chris-hayes/watch/the-chaotic-gop-immigration-vote-314487875924 ' ,
' md5 ' : ' 6d236bf4f3dddc226633ce6e2c3f814d ' ,
' info_dict ' : {
2018-11-21 12:10:06 +01:00
' id ' : ' 314487875924 ' ,
2016-06-23 02:36:19 +02:00
' ext ' : ' mp4 ' ,
' title ' : ' The chaotic GOP immigration vote ' ,
' description ' : ' The Republican House votes on a border bill that has no chance of getting through the Senate or signed by the President and is drawing criticism from all sides. ' ,
2017-01-02 13:08:07 +01:00
' thumbnail ' : r ' re:^https?://.* \ .jpg$ ' ,
2016-06-23 02:36:19 +02:00
' timestamp ' : 1406937606 ,
' upload_date ' : ' 20140802 ' ,
2016-04-22 19:07:40 +02:00
} ,
} ,
2015-08-01 17:43:33 +02:00
{
' url ' : ' http://www.nbcnews.com/watch/dateline/full-episode--deadly-betrayal-386250819952 ' ,
' only_matching ' : True ,
} ,
2016-06-10 07:31:55 +02:00
{
# From http://www.vulture.com/2016/06/letterman-couldnt-care-less-about-late-night.html
' url ' : ' http://www.nbcnews.com/widget/video-embed/701714499682 ' ,
' only_matching ' : True ,
} ,
2014-05-30 00:38:57 +02:00
]
2013-08-27 12:38:30 +02:00
def _real_extract ( self , url ) :
2018-11-21 12:10:06 +01:00
video_id = self . _match_id ( url )
2019-10-28 19:31:20 +01:00
webpage = self . _download_webpage ( url , video_id )
data = self . _parse_json ( self . _search_regex (
2020-11-19 20:22:59 +01:00
r ' <script[^>]+id= " __NEXT_DATA__ " [^>]*>( { .+?})</script> ' ,
webpage , ' bootstrap json ' ) , video_id ) [ ' props ' ] [ ' initialState ' ]
2019-10-28 19:31:20 +01:00
video_data = try_get ( data , lambda x : x [ ' video ' ] [ ' current ' ] , dict )
if not video_data :
video_data = data [ ' article ' ] [ ' content ' ] [ 0 ] [ ' primaryMedia ' ] [ ' video ' ]
title = video_data [ ' headline ' ] [ ' primary ' ]
formats = [ ]
for va in video_data . get ( ' videoAssets ' , [ ] ) :
public_url = va . get ( ' publicUrl ' )
if not public_url :
continue
if ' ://link.theplatform.com/ ' in public_url :
public_url = update_url_query ( public_url , { ' format ' : ' redirect ' } )
format_id = va . get ( ' format ' )
if format_id == ' M3U ' :
formats . extend ( self . _extract_m3u8_formats (
public_url , video_id , ' mp4 ' , ' m3u8_native ' ,
m3u8_id = format_id , fatal = False ) )
continue
tbr = int_or_none ( va . get ( ' bitrate ' ) , 1000 )
if tbr :
format_id + = ' - %d ' % tbr
formats . append ( {
' format_id ' : format_id ,
' url ' : public_url ,
' width ' : int_or_none ( va . get ( ' width ' ) ) ,
' height ' : int_or_none ( va . get ( ' height ' ) ) ,
' tbr ' : tbr ,
' ext ' : ' mp4 ' ,
} )
self . _sort_formats ( formats )
2016-12-17 18:11:13 +01:00
2019-10-28 19:31:20 +01:00
subtitles = { }
closed_captioning = video_data . get ( ' closedCaptioning ' )
if closed_captioning :
for cc_url in closed_captioning . values ( ) :
if not cc_url :
continue
subtitles . setdefault ( ' en ' , [ ] ) . append ( {
' url ' : cc_url ,
} )
2016-12-17 18:11:13 +01:00
2018-11-21 12:10:06 +01:00
return {
' id ' : video_id ,
2019-10-28 19:31:20 +01:00
' title ' : title ,
' description ' : try_get ( video_data , lambda x : x [ ' description ' ] [ ' primary ' ] ) ,
' thumbnail ' : try_get ( video_data , lambda x : x [ ' primaryImage ' ] [ ' url ' ] [ ' primary ' ] ) ,
' duration ' : parse_duration ( video_data . get ( ' duration ' ) ) ,
' timestamp ' : unified_timestamp ( video_data . get ( ' datePublished ' ) ) ,
' formats ' : formats ,
' subtitles ' : subtitles ,
2018-11-21 12:10:06 +01:00
}
2016-09-11 20:55:15 +02:00
class NBCOlympicsIE ( InfoExtractor ) :
2018-02-09 02:05:12 +01:00
IE_NAME = ' nbcolympics '
2021-08-04 06:19:44 +02:00
_VALID_URL = r ' https?://www \ .nbcolympics \ .com/videos?/(?P<id>[0-9a-z-]+) '
2016-09-11 20:55:15 +02:00
_TEST = {
# Geo-restricted to US
' url ' : ' http://www.nbcolympics.com/video/justin-roses-son-leo-was-tears-after-his-dad-won-gold ' ,
' md5 ' : ' 54fecf846d05429fbaa18af557ee523a ' ,
' info_dict ' : {
' id ' : ' WjTBzDXx5AUq ' ,
' display_id ' : ' justin-roses-son-leo-was-tears-after-his-dad-won-gold ' ,
' ext ' : ' mp4 ' ,
' title ' : ' Rose \' s son Leo was in tears after his dad won gold ' ,
' description ' : ' Olympic gold medalist Justin Rose gets emotional talking to the impact his win in men \' s golf has already had on his children. ' ,
' timestamp ' : 1471274964 ,
' upload_date ' : ' 20160815 ' ,
' uploader ' : ' NBCU-SPORTS ' ,
} ,
}
def _real_extract ( self , url ) :
display_id = self . _match_id ( url )
webpage = self . _download_webpage ( url , display_id )
2021-08-04 06:19:44 +02:00
try :
drupal_settings = self . _parse_json ( self . _search_regex (
r ' jQuery \ .extend \ (Drupal \ .settings \ s*, \ s*( { .+?}) \ ); ' ,
webpage , ' drupal settings ' ) , display_id )
iframe_url = drupal_settings [ ' vod ' ] [ ' iframe_url ' ]
theplatform_url = iframe_url . replace (
' vplayer.nbcolympics.com ' , ' player.theplatform.com ' )
except RegexNotFoundError :
theplatform_url = self . _search_regex (
r " ([ \" ' ])embedUrl \ 1: *([ \" ' ])(?P<embedUrl>.+) \ 2 " ,
webpage , ' embedding URL ' , group = " embedUrl " )
2016-09-11 20:55:15 +02:00
return {
' _type ' : ' url_transparent ' ,
' url ' : theplatform_url ,
' ie_key ' : ThePlatformIE . ie_key ( ) ,
' display_id ' : display_id ,
}
2018-02-09 02:05:12 +01:00
class NBCOlympicsStreamIE ( AdobePassIE ) :
IE_NAME = ' nbcolympics:stream '
_VALID_URL = r ' https?://stream \ .nbcolympics \ .com/(?P<id>[0-9a-z-]+) '
2021-07-29 17:31:27 +02:00
_TESTS = [
{
' note ' : ' Tokenized m3u8 source URL ' ,
' url ' : ' https://stream.nbcolympics.com/womens-soccer-group-round-11 ' ,
' info_dict ' : {
' id ' : ' 2019740 ' ,
' ext ' : ' mp4 ' ,
' title ' : r " re:Women ' s Group Stage - Netherlands vs \ . Brazil [0-9] {4} -[0-9] {2} -[0-9] {2} [0-9] {2} :[0-9] {2} $ " ,
} ,
' params ' : {
' skip_download ' : ' m3u8 ' ,
} ,
} , {
' note ' : ' Plain m3u8 source URL ' ,
' url ' : ' https://stream.nbcolympics.com/gymnastics-event-finals-mens-floor-pommel-horse-womens-vault-bars ' ,
' info_dict ' : {
' id ' : ' 2021729 ' ,
' ext ' : ' mp4 ' ,
' title ' : r ' re:Event Finals: M Floor, W Vault, M Pommel, W Uneven Bars [0-9] {4} -[0-9] {2} -[0-9] {2} [0-9] {2} :[0-9] {2} $ ' ,
} ,
' params ' : {
' skip_download ' : ' m3u8 ' ,
} ,
2018-02-09 02:05:12 +01:00
} ,
2021-07-29 17:31:27 +02:00
]
2018-02-09 02:05:12 +01:00
def _real_extract ( self , url ) :
display_id = self . _match_id ( url )
webpage = self . _download_webpage ( url , display_id )
pid = self . _search_regex ( r ' pid \ s*= \ s*( \ d+); ' , webpage , ' pid ' )
2021-07-29 17:31:27 +02:00
2018-02-09 02:05:12 +01:00
event_config = self . _download_json (
2021-07-29 17:31:27 +02:00
f ' http://stream.nbcolympics.com/data/event_config_ { pid } .json ' ,
pid , ' Downloading event config ' ) [ ' eventConfig ' ]
title = event_config [ ' eventTitle ' ]
is_live = { ' live ' : True , ' replay ' : False } . get ( event_config . get ( ' eventStatus ' ) )
if is_live :
title = self . _live_title ( title )
2018-02-09 02:05:12 +01:00
source_url = self . _download_json (
2021-07-29 17:31:27 +02:00
f ' https://api-leap.nbcsports.com/feeds/assets/ { pid } ?application=NBCOlympics&platform=desktop&format=nbc-player&env=staging ' ,
pid , ' Downloading leap config '
) [ ' videoSources ' ] [ 0 ] [ ' cdnSources ' ] [ ' primary ' ] [ 0 ] [ ' sourceUrl ' ]
if event_config . get ( ' cdnToken ' ) :
ap_resource = self . _get_mvpd_resource (
event_config . get ( ' resourceId ' , ' NBCOlympics ' ) ,
re . sub ( r ' [^ \ w \ d ]+ ' , ' ' , event_config [ ' eventTitle ' ] ) , pid ,
event_config . get ( ' ratingId ' , ' NO VALUE ' ) )
media_token = self . _extract_mvpd_auth ( url , pid , event_config . get ( ' requestorId ' , ' NBCOlympics ' ) , ap_resource )
source_url = self . _download_json (
' https://tokens.playmakerservices.com/ ' , pid , ' Retrieving tokenized URL ' ,
data = json . dumps ( {
' application ' : ' NBCSports ' ,
' authentication-type ' : ' adobe-pass ' ,
' cdn ' : ' akamai ' ,
' pid ' : pid ,
' platform ' : ' desktop ' ,
' requestorId ' : ' NBCOlympics ' ,
' resourceId ' : base64 . b64encode ( ap_resource . encode ( ) ) . decode ( ) ,
' token ' : base64 . b64encode ( media_token . encode ( ) ) . decode ( ) ,
' url ' : source_url ,
' version ' : ' v1 ' ,
} ) . encode ( ) ,
) [ ' akamai ' ] [ 0 ] [ ' tokenizedUrl ' ]
formats = self . _extract_m3u8_formats ( source_url , pid , ' mp4 ' , live = is_live )
for f in formats :
# -http_seekable requires ffmpeg 4.3+ but it doesnt seem possible to
# download with ffmpeg without this option
f [ ' _ffmpeg_args ' ] = [ ' -seekable ' , ' 0 ' , ' -http_seekable ' , ' 0 ' , ' -icy ' , ' 0 ' ]
2018-02-09 02:05:12 +01:00
self . _sort_formats ( formats )
return {
' id ' : pid ,
' display_id ' : display_id ,
' title ' : title ,
' formats ' : formats ,
2021-07-29 17:31:27 +02:00
' is_live ' : is_live ,
2018-02-09 02:05:12 +01:00
}