yt-dlp/youtube_dl/extractor/bbc.py
Aakash Gajjar b827ee921f
pull changes from remote master (#190)
* [scrippsnetworks] Add new extractor(closes #19857)(closes #22981)

* [teachable] Improve locked lessons detection (#23528)

* [teachable] Fail with error message if no video URL found

* [extractors] add missing import for ScrippsNetworksIE

* [brightcove] cache brightcove player policy keys

* [prosiebensat1] improve geo restriction handling(closes #23571)

* [soundcloud] automatically update client id on failing requests

* [spankbang] Fix extraction (closes #23307, closes #23423, closes #23444)

* [spankbang] Improve removed video detection (#23423)

* [brightcove] update policy key on failing requests

* [pornhub] Fix extraction and add support for m3u8 formats (closes #22749, closes #23082)

* [pornhub] Improve locked videos detection (closes #22449, closes #22780)

* [brightcove] invalidate policy key cache on failing requests

* [soundcloud] fix client id extraction for non fatal requests

* [ChangeLog] Actualize
[ci skip]

* [devscripts/create-github-release] Switch to using PAT for authentication

Basic authentication will be deprecated soon

* release 2020.01.01

* [redtube] Detect private videos (#23518)

* [vice] improve extraction(closes #23631)

* [devscripts/create-github-release] Remove unused import

* [wistia] improve format extraction and extract subtitles(closes #22590)

* [nrktv:seriebase] Fix extraction (closes #23625) (#23537)

* [discovery] fix anonymous token extraction(closes #23650)

* [scrippsnetworks] add support for www.discovery.com videos

* [scrippsnetworks] correct test case URL

* [dctp] fix format extraction(closes #23656)

* [pandatv] Remove extractor (#23630)

* [naver] improve extraction

- improve geo-restriction handling
- extract automatic captions
- extract uploader metadata
- extract VLive HLS formats

* [naver] improve metadata extraction

* [cloudflarestream] improve extraction

- add support for bytehighway.net domain
- add support for signed URLs
- extract thumbnail

* [cloudflarestream] import embed URL extraction

* [lego] fix extraction and extract subtitle(closes #23687)

* [safari] Fix kaltura session extraction (closes #23679) (#23670)

* [orf:fm4] Fix extraction (#23599)

* [orf:radio] Clean description and improve extraction

* [twitter] add support for promo_video_website cards(closes #23711)

* [vodplatform] add support for embed.kwikmotion.com domain

* [ndr:base:embed] Improve thumbnails extraction (closes #23731)

* [canvas] Add support for new API endpoint and update tests (closes #17680, closes #18629)

* [travis] Add flake8 job (#23720)

* [yourporn] Fix extraction (closes #21645, closes #22255, closes #23459)

* [ChangeLog] Actualize
[ci skip]

* release 2020.01.15

* [soundcloud] Restore previews extraction (closes #23739)

* [orf:tvthek] Improve geo restricted videos detection (closes #23741)

* [zype] improve extraction

- extract subtitles(closes #21258)
- support URLs with alternative keys/tokens(#21258)
- extract more metadata

* [americastestkitchen] fix extraction

* [nbc] add support for nbc multi network URLs(closes #23049)

* [ard] improve extraction(closes #23761)

- simplify extraction
- extract age limit and series
- bypass geo-restriction

* [ivi:compilation] Fix entries extraction (closes #23770)

* [24video] Add support for 24video.vip (closes #23753)

* [businessinsider] Fix jwplatform id extraction (closes #22929) (#22954)

* [ard] add a missing condition

* [azmedien] fix extraction(closes #23783)

* [voicerepublic] fix extraction

* [stretchinternet] fix extraction(closes #4319)

* [youtube] Fix sigfunc name extraction (closes #23819)

* [ChangeLog] Actualize
[ci skip]

* release 2020.01.24

* [soundcloud] imporve private playlist/set tracks extraction

https://github.com/ytdl-org/youtube-dl/issues/3707#issuecomment-577873539

* [svt] fix article extraction(closes #22897)(closes #22919)

* [svt] fix series extraction(closes #22297)

* [viewlift] improve extraction

- fix extraction(closes #23851)
- add add support for authentication
- add support for more domains

* [vimeo] fix album extraction(closes #23864)

* [tva] Relax _VALID_URL (closes #23903)

* [tv5mondeplus] Fix extraction (closes #23907, closes #23911)

* [twitch:stream] Lowercase channel id for stream request (closes #23917)

* [sportdeutschland] Update to new sportdeutschland API

They switched to SSL, but under a different host AND path...
Remove the old test cases because these videos have become unavailable.

* [popcorntimes] Add extractor (closes #23949)

* [thisoldhouse] fix extraction(closes #23951)

* [toggle] Add support for mewatch.sg (closes #23895) (#23930)

* [compat] Introduce compat_realpath (refs #23991)

* [update] Fix updating via symlinks (closes #23991)

* [nytimes] improve format sorting(closes #24010)

* [abc:iview] Support 720p (#22907) (#22921)

* [nova:embed] Fix extraction (closes #23672)

* [nova:embed] Improve (closes #23690)

* [nova] Improve extraction (refs #23690)

* [jpopsuki] Remove extractor (closes #23858)

* [YoutubeDL] Fix playlist entry indexing with --playlist-items (closes #10591, closes #10622)

* [test_YoutubeDL] Fix get_ids

* [test_YoutubeDL] Add tests for #10591 (closes #23873)

* [24video] Add support for porn.24video.net (closes #23779, closes #23784)

* [npr] Add support for streams (closes #24042)

* [ChangeLog] Actualize
[ci skip]

* release 2020.02.16

* [tv2dk:bornholm:play] Fix extraction (#24076)

* [imdb] Fix extraction (closes #23443)

* [wistia] Add support for multiple generic embeds (closes #8347, closes #11385)

* [teachable] Add support for multiple videos per lecture (closes #24101)

* [pornhd] Fix extraction (closes #24128)

* [options] Remove duplicate short option -v for --version (#24162)

* [extractor/common] Convert ISM manifest to unicode before processing on python 2 (#24152)

* [YoutubeDL] Force redirect URL to unicode on python 2

* Remove no longer needed compat_str around geturl

* [youjizz] Fix extraction (closes #24181)

* [test_subtitles] Remove obsolete test

* [zdf:channel] Fix tests

* [zapiks] Fix test

* [xtube] Fix metadata extraction (closes #21073, closes #22455)

* [xtube:user] Fix test

* [telecinco] Fix extraction (refs #24195)

* [telecinco] Add support for article opening videos

* [franceculture] Fix extraction (closes #24204)

* [xhamster] Fix extraction (closes #24205)

* [ChangeLog] Actualize
[ci skip]

* release 2020.03.01

* [vimeo] Fix subtitles URLs (#24209)

* [servus] Add support for new URL schema (closes #23475, closes #23583, closes #24142)

* [youtube:playlist] Fix tests (closes #23872) (#23885)

* [peertube] Improve extraction

* [peertube] Fix issues and improve extraction (closes #23657)

* [pornhub] Improve title extraction (closes #24184)

* [vimeo] fix showcase password protected video extraction(closes #24224)

* [youtube] Fix age-gated videos support without login (closes #24248)

* [youtube] Fix tests

* [ChangeLog] Actualize
[ci skip]

* release 2020.03.06

* [nhk] update API version(closes #24270)

* [youtube] Improve extraction in 429 error conditions (closes #24283)

* [youtube] Improve age-gated videos extraction in 429 error conditions (refs #24283)

* [youtube] Remove outdated code

Additional get_video_info requests don't seem to provide any extra itags any longer

* [README.md] Clarify 429 error

* [pornhub] Add support for pornhubpremium.com (#24288)

* [utils] Add support for cookies with spaces used instead of tabs

* [ChangeLog] Actualize
[ci skip]

* release 2020.03.08

* Revert "[utils] Add support for cookies with spaces used instead of tabs"

According to [1] TABs must be used as separators between fields.
Files produces by some tools with spaces as separators are considered
malformed.

1. https://curl.haxx.se/docs/http-cookies.html

This reverts commit cff99c91d1.

* [utils] Add reference to cookie file format

* Revert "[vimeo] fix showcase password protected video extraction(closes #24224)"

This reverts commit 12ee431676.

* [nhk] Relax _VALID_URL (#24329)

* [nhk] Remove obsolete rtmp formats (closes #24329)

* [nhk] Update m3u8 URL and use native hls (#24329)

* [ndr] Fix extraction (closes #24326)

* [xtube] Fix formats extraction (closes #24348)

* [xtube] Fix typo

* [hellporno] Fix extraction (closes #24399)

* [cbc:watch] Add support for authentication

* [cbc:watch] Fix authenticated device token caching (closes #19160)

* [soundcloud] fix download url extraction(closes #24394)

* [limelight] remove disabled API requests(closes #24255)

* [bilibili] Add support for new URL schema with BV ids (closes #24439, closes #24442)

* [bilibili] Add support for player.bilibili.com (closes #24402)

* [teachable] Extract chapter metadata (closes #24421)

* [generic] Look for teachable embeds before wistia

* [teachable] Update upskillcourses domain

New version does not use teachable platform any longer

* [teachable] Update gns3 domain

* [teachable] Update test

* [ChangeLog] Actualize
[ci skip]

* [ChangeLog] Actualize
[ci skip]

* release 2020.03.24

* [spankwire] Fix extraction (closes #18924, closes #20648)

* [spankwire] Add support for generic embeds (refs #24633)

* [youporn] Add support form generic embeds

* [mofosex] Add support for generic embeds (closes #24633)

* [tele5] Fix extraction (closes #24553)

* [extractor/common] Skip malformed ISM manifest XMLs while extracting ISM formats (#24667)

* [tv4] Fix ISM formats extraction (closes #24667)

* [twitch:clips] Extend _VALID_URL (closes #24290) (#24642)

* [motherless] Fix extraction (closes #24699)

* [nova:embed] Fix extraction (closes #24700)

* [youtube] Skip broken multifeed videos (closes #24711)

* [soundcloud] Extract AAC format

* [soundcloud] Improve AAC format extraction (closes #19173, closes #24708)

* [thisoldhouse] Fix video id extraction (closes #24548)

Added support for:
with of without "www."
and either  ".chorus.build" or ".com"

It now validated correctly on older URL's
```
<iframe src="https://thisoldhouse.chorus.build/videos/zype/5e33baec27d2e50001d5f52f
```
and newer ones
```
<iframe src="https://www.thisoldhouse.com/videos/zype/5e2b70e95216cc0001615120
```

* [thisoldhouse] Improve video id extraction (closes #24549)

* [youtube] Fix DRM videos detection (refs #24736)

* [options] Clarify doc on --exec command (closes #19087) (#24883)

* [prosiebensat1] Improve extraction and remove 7tv.de support (#24948)

* [prosiebensat1] Extract series metadata

* [tenplay] Relax _VALID_URL (closes #25001)

* [tvplay] fix Viafree extraction(closes #15189)(closes #24473)(closes #24789)

* [yahoo] fix GYAO Player extraction and relax title URL regex(closes #24178)(closes #24778)

* [youtube] Use redirected video id if any (closes #25063)

* [youtube] Improve player id extraction and add tests

* [extractor/common] Extract multiple JSON-LD entries

* [crunchyroll] Fix and improve extraction (closes #25096, closes #25060)

* [ChangeLog] Actualize
[ci skip]

* release 2020.05.03

* [puhutv] Remove no longer available HTTP formats (closes #25124)

* [utils] Improve cookie files support

+ Add support for UTF-8 in cookie files
* Skip malformed cookie file entries instead of crashing (invalid entry len, invalid expires at)

* [dailymotion] Fix typo

* [compat] Introduce compat_cookiejar_Cookie

* [extractor/common] Use compat_cookiejar_Cookie for _set_cookie (closes #23256, closes #24776)

To always ensure cookie name and value are bytestrings on python 2.

* [orf] Add support for more radio stations (closes #24938) (#24968)

* [uol] fix extraction(closes #22007)

* [downloader/http] Finish downloading once received data length matches expected

Always do this if possible, i.e. if Content-Length or expected length is known, not only in test.
This will save unnecessary last extra loop trying to read 0 bytes.

* [downloader/http] Request last data block of exact remaining size

Always request last data block of exact size remaining to download if possible not the current block size.

* [iprima] Improve extraction (closes #25138)

* [youtube] Improve signature cipher extraction (closes #25188)

* [ChangeLog] Actualize
[ci skip]

* release 2020.05.08

* [spike] fix Bellator mgid extraction(closes #25195)

* [bbccouk] PEP8

* [mailru] Fix extraction (closes #24530) (#25239)

* [README.md] flake8 HTTPS URL (#25230)

* [youtube] Add support for yewtu.be (#25226)

* [soundcloud] reduce API playlist page limit(closes #25274)

* [vimeo] improve format extraction and sorting(closes #25285)

* [redtube] Improve title extraction (#25208)

* [indavideo] Switch to HTTPS for API request (#25191)

* [utils] Fix file permissions in write_json_file (closes #12471) (#25122)

* [redtube] Improve formats extraction and extract m3u8 formats (closes #25311, closes #25321)

* [ard] Improve _VALID_URL (closes #25134) (#25198)

* [giantbomb] Extend _VALID_URL (#25222)

* [postprocessor/ffmpeg] Embed series metadata with --add-metadata

* [youtube] Add support for more invidious instances (#25417)

* [ard:beta] Extend _VALID_URL (closes #25405)

* [ChangeLog] Actualize
[ci skip]

* release 2020.05.29

* [jwplatform] Improve embeds extraction (closes #25467)

* [periscope] Fix untitled broadcasts (#25482)

* [twitter:broadcast] Add untitled periscope broadcast test

* [malltv] Add support for sk.mall.tv (#25445)

* [brightcove] Fix subtitles extraction (closes #25540)

* [brightcove] Sort imports

* [twitch] Pass v5 accept header and fix thumbnails extraction (closes #25531)

* [twitch:stream] Fix extraction (closes #25528)

* [twitch:stream] Expect 400 and 410 HTTP errors from API

* [tele5] Prefer jwplatform over nexx (closes #25533)

* [jwplatform] Add support for bypass geo restriction

* [tele5] Bypass geo restriction

* [ChangeLog] Actualize
[ci skip]

* release 2020.06.06

* [kaltura] Add support for multiple embeds on a webpage (closes #25523)

* [youtube] Extract chapters from JSON (closes #24819)

* [facebook] Support single-video ID links

I stumbled upon this at https://www.facebook.com/bwfbadminton/posts/10157127020046316 . No idea how prevalent it is yet.

* [youtube] Fix playlist and feed extraction (closes #25675)

* [youtube] Fix thumbnails extraction and remove uploader id extraction warning (closes #25676)

* [youtube] Fix upload date extraction

* [youtube] Improve view count extraction

* [youtube] Fix uploader id and uploader URL extraction

* [ChangeLog] Actualize
[ci skip]

* release 2020.06.16

* [youtube] Fix categories and improve tags extraction

* [youtube] Force old layout (closes #25682, closes #25683, closes #25680, closes #25686)

* [ChangeLog] Actualize
[ci skip]

* release 2020.06.16.1

* [brightcove] Improve embed detection (closes #25674)

* [bellmedia] add support for cp24.com clip URLs(closes #25764)

* [youtube:playlists] Extend _VALID_URL (closes #25810)

* [youtube] Prevent excess HTTP 301 (#25786)

* [wistia] Restrict embed regex (closes #25969)

* [youtube] Improve description extraction (closes #25937) (#25980)

* [youtube] Fix sigfunc name extraction (closes #26134, closes #26135, closes #26136, closes #26137)

* [ChangeLog] Actualize
[ci skip]

* release 2020.07.28

* [xhamster] Extend _VALID_URL (closes #25789) (#25804)

* [xhamster] Fix extraction (closes #26157) (#26254)

* [xhamster] Extend _VALID_URL (closes #25927)

Co-authored-by: Remita Amine <remitamine@gmail.com>
Co-authored-by: Sergey M․ <dstftw@gmail.com>
Co-authored-by: nmeum <soeren+github@soeren-tempel.net>
Co-authored-by: Roxedus <me@roxedus.dev>
Co-authored-by: Singwai Chan <c.singwai@gmail.com>
Co-authored-by: cdarlint <cdarlint@users.noreply.github.com>
Co-authored-by: Johannes N <31795504+jonolt@users.noreply.github.com>
Co-authored-by: jnozsc <jnozsc@gmail.com>
Co-authored-by: Moritz Patelscheck <moritz.patelscheck@campus.tu-berlin.de>
Co-authored-by: PB <3854688+uno20001@users.noreply.github.com>
Co-authored-by: Philipp Hagemeister <phihag@phihag.de>
Co-authored-by: Xaver Hellauer <software@hellauer.bayern>
Co-authored-by: d2au <d2au.dev@gmail.com>
Co-authored-by: Jan 'Yenda' Trmal <jtrmal@gmail.com>
Co-authored-by: jxu <7989982+jxu@users.noreply.github.com>
Co-authored-by: Martin Ström <name@my-domain.se>
Co-authored-by: The Hatsune Daishi <nao20010128@gmail.com>
Co-authored-by: tsia <github@tsia.de>
Co-authored-by: 3risian <59593325+3risian@users.noreply.github.com>
Co-authored-by: Tristan Waddington <tristan.waddington@gmail.com>
Co-authored-by: Devon Meunier <devon.meunier@gmail.com>
Co-authored-by: Felix Stupp <felix.stupp@outlook.com>
Co-authored-by: tom <tomster954@gmail.com>
Co-authored-by: AndrewMBL <62922222+AndrewMBL@users.noreply.github.com>
Co-authored-by: willbeaufoy <will@willbeaufoy.net>
Co-authored-by: Philipp Stehle <anderschwiedu@googlemail.com>
Co-authored-by: hh0rva1h <61889859+hh0rva1h@users.noreply.github.com>
Co-authored-by: comsomisha <shmelev1996@mail.ru>
Co-authored-by: TotalCaesar659 <14265316+TotalCaesar659@users.noreply.github.com>
Co-authored-by: Juan Francisco Cantero Hurtado <iam@juanfra.info>
Co-authored-by: Dave Loyall <dave@the-good-guys.net>
Co-authored-by: tlsssl <63866177+tlsssl@users.noreply.github.com>
Co-authored-by: Rob <ankenyr@gmail.com>
Co-authored-by: Michael Klein <github@a98shuttle.de>
Co-authored-by: JordanWeatherby <47519158+JordanWeatherby@users.noreply.github.com>
Co-authored-by: striker.sh <19488257+strikersh@users.noreply.github.com>
Co-authored-by: Matej Dujava <mdujava@gmail.com>
Co-authored-by: Glenn Slayden <5589855+glenn-slayden@users.noreply.github.com>
Co-authored-by: MRWITEK <mrvvitek@gmail.com>
Co-authored-by: JChris246 <43832407+JChris246@users.noreply.github.com>
Co-authored-by: TheRealDude2 <the.real.dude@gmx.de>
2020-08-25 20:23:34 +05:30

1359 lines
58 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# coding: utf-8
from __future__ import unicode_literals
import itertools
import re
from .common import InfoExtractor
from ..utils import (
clean_html,
dict_get,
ExtractorError,
float_or_none,
get_element_by_class,
int_or_none,
js_to_json,
parse_duration,
parse_iso8601,
try_get,
unescapeHTML,
url_or_none,
urlencode_postdata,
urljoin,
)
from ..compat import (
compat_etree_Element,
compat_HTTPError,
compat_urlparse,
)
class BBCCoUkIE(InfoExtractor):
IE_NAME = 'bbc.co.uk'
IE_DESC = 'BBC iPlayer'
_ID_REGEX = r'(?:[pbm][\da-z]{7}|w[\da-z]{7,14})'
_VALID_URL = r'''(?x)
https?://
(?:www\.)?bbc\.co\.uk/
(?:
programmes/(?!articles/)|
iplayer(?:/[^/]+)?/(?:episode/|playlist/)|
music/(?:clips|audiovideo/popular)[/#]|
radio/player/|
sounds/play/|
events/[^/]+/play/[^/]+/
)
(?P<id>%s)(?!/(?:episodes|broadcasts|clips))
''' % _ID_REGEX
_LOGIN_URL = 'https://account.bbc.com/signin'
_NETRC_MACHINE = 'bbc'
_MEDIASELECTOR_URLS = [
# Provides HQ HLS streams with even better quality that pc mediaset but fails
# with geolocation in some cases when it's even not geo restricted at all (e.g.
# http://www.bbc.co.uk/programmes/b06bp7lf). Also may fail with selectionunavailable.
'http://open.live.bbc.co.uk/mediaselector/5/select/version/2.0/mediaset/iptv-all/vpid/%s',
'http://open.live.bbc.co.uk/mediaselector/5/select/version/2.0/mediaset/pc/vpid/%s',
]
_MEDIASELECTION_NS = 'http://bbc.co.uk/2008/mp/mediaselection'
_EMP_PLAYLIST_NS = 'http://bbc.co.uk/2008/emp/playlist'
_NAMESPACES = (
_MEDIASELECTION_NS,
_EMP_PLAYLIST_NS,
)
_TESTS = [
{
'url': 'http://www.bbc.co.uk/programmes/b039g8p7',
'info_dict': {
'id': 'b039d07m',
'ext': 'flv',
'title': 'Kaleidoscope, Leonard Cohen',
'description': 'The Canadian poet and songwriter reflects on his musical career.',
},
'params': {
# rtmp download
'skip_download': True,
}
},
{
'url': 'http://www.bbc.co.uk/iplayer/episode/b00yng5w/The_Man_in_Black_Series_3_The_Printed_Name/',
'info_dict': {
'id': 'b00yng1d',
'ext': 'flv',
'title': 'The Man in Black: Series 3: The Printed Name',
'description': "Mark Gatiss introduces Nicholas Pierpan's chilling tale of a writer's devilish pact with a mysterious man. Stars Ewan Bailey.",
'duration': 1800,
},
'params': {
# rtmp download
'skip_download': True,
},
'skip': 'Episode is no longer available on BBC iPlayer Radio',
},
{
'url': 'http://www.bbc.co.uk/iplayer/episode/b03vhd1f/The_Voice_UK_Series_3_Blind_Auditions_5/',
'info_dict': {
'id': 'b00yng1d',
'ext': 'flv',
'title': 'The Voice UK: Series 3: Blind Auditions 5',
'description': 'Emma Willis and Marvin Humes present the fifth set of blind auditions in the singing competition, as the coaches continue to build their teams based on voice alone.',
'duration': 5100,
},
'params': {
# rtmp download
'skip_download': True,
},
'skip': 'Currently BBC iPlayer TV programmes are available to play in the UK only',
},
{
'url': 'http://www.bbc.co.uk/iplayer/episode/p026c7jt/tomorrows-worlds-the-unearthly-history-of-science-fiction-2-invasion',
'info_dict': {
'id': 'b03k3pb7',
'ext': 'flv',
'title': "Tomorrow's Worlds: The Unearthly History of Science Fiction",
'description': '2. Invasion',
'duration': 3600,
},
'params': {
# rtmp download
'skip_download': True,
},
'skip': 'Currently BBC iPlayer TV programmes are available to play in the UK only',
}, {
'url': 'http://www.bbc.co.uk/programmes/b04v20dw',
'info_dict': {
'id': 'b04v209v',
'ext': 'flv',
'title': 'Pete Tong, The Essential New Tune Special',
'description': "Pete has a very special mix - all of 2014's Essential New Tunes!",
'duration': 10800,
},
'params': {
# rtmp download
'skip_download': True,
},
'skip': 'Episode is no longer available on BBC iPlayer Radio',
}, {
'url': 'http://www.bbc.co.uk/music/clips/p022h44b',
'note': 'Audio',
'info_dict': {
'id': 'p022h44j',
'ext': 'flv',
'title': 'BBC Proms Music Guides, Rachmaninov: Symphonic Dances',
'description': "In this Proms Music Guide, Andrew McGregor looks at Rachmaninov's Symphonic Dances.",
'duration': 227,
},
'params': {
# rtmp download
'skip_download': True,
}
}, {
'url': 'http://www.bbc.co.uk/music/clips/p025c0zz',
'note': 'Video',
'info_dict': {
'id': 'p025c103',
'ext': 'flv',
'title': 'Reading and Leeds Festival, 2014, Rae Morris - Closer (Live on BBC Three)',
'description': 'Rae Morris performs Closer for BBC Three at Reading 2014',
'duration': 226,
},
'params': {
# rtmp download
'skip_download': True,
}
}, {
'url': 'http://www.bbc.co.uk/iplayer/episode/b054fn09/ad/natural-world-20152016-2-super-powered-owls',
'info_dict': {
'id': 'p02n76xf',
'ext': 'flv',
'title': 'Natural World, 2015-2016: 2. Super Powered Owls',
'description': 'md5:e4db5c937d0e95a7c6b5e654d429183d',
'duration': 3540,
},
'params': {
# rtmp download
'skip_download': True,
},
'skip': 'geolocation',
}, {
'url': 'http://www.bbc.co.uk/iplayer/episode/b05zmgwn/royal-academy-summer-exhibition',
'info_dict': {
'id': 'b05zmgw1',
'ext': 'flv',
'description': 'Kirsty Wark and Morgan Quaintance visit the Royal Academy as it prepares for its annual artistic extravaganza, meeting people who have come together to make the show unique.',
'title': 'Royal Academy Summer Exhibition',
'duration': 3540,
},
'params': {
# rtmp download
'skip_download': True,
},
'skip': 'geolocation',
}, {
# iptv-all mediaset fails with geolocation however there is no geo restriction
# for this programme at all
'url': 'http://www.bbc.co.uk/programmes/b06rkn85',
'info_dict': {
'id': 'b06rkms3',
'ext': 'flv',
'title': "Best of the Mini-Mixes 2015: Part 3, Annie Mac's Friday Night - BBC Radio 1",
'description': "Annie has part three in the Best of the Mini-Mixes 2015, plus the year's Most Played!",
},
'params': {
# rtmp download
'skip_download': True,
},
'skip': 'Now it\'s really geo-restricted',
}, {
# compact player (https://github.com/ytdl-org/youtube-dl/issues/8147)
'url': 'http://www.bbc.co.uk/programmes/p028bfkf/player',
'info_dict': {
'id': 'p028bfkj',
'ext': 'flv',
'title': 'Extract from BBC documentary Look Stranger - Giant Leeks and Magic Brews',
'description': 'Extract from BBC documentary Look Stranger - Giant Leeks and Magic Brews',
},
'params': {
# rtmp download
'skip_download': True,
},
}, {
'url': 'https://www.bbc.co.uk/sounds/play/m0007jzb',
'note': 'Audio',
'info_dict': {
'id': 'm0007jz9',
'ext': 'mp4',
'title': 'BBC Proms, 2019, Prom 34: WestEastern Divan Orchestra',
'description': "Live BBC Proms. WestEastern Divan Orchestra with Daniel Barenboim and Martha Argerich.",
'duration': 9840,
},
'params': {
# rtmp download
'skip_download': True,
}
}, {
'url': 'http://www.bbc.co.uk/iplayer/playlist/p01dvks4',
'only_matching': True,
}, {
'url': 'http://www.bbc.co.uk/music/clips#p02frcc3',
'only_matching': True,
}, {
'url': 'http://www.bbc.co.uk/iplayer/cbeebies/episode/b0480276/bing-14-atchoo',
'only_matching': True,
}, {
'url': 'http://www.bbc.co.uk/radio/player/p03cchwf',
'only_matching': True,
}, {
'url': 'https://www.bbc.co.uk/music/audiovideo/popular#p055bc55',
'only_matching': True,
}, {
'url': 'http://www.bbc.co.uk/programmes/w3csv1y9',
'only_matching': True,
}, {
'url': 'https://www.bbc.co.uk/programmes/m00005xn',
'only_matching': True,
}, {
'url': 'https://www.bbc.co.uk/programmes/w172w4dww1jqt5s',
'only_matching': True,
}]
_USP_RE = r'/([^/]+?)\.ism(?:\.hlsv2\.ism)?/[^/]+\.m3u8'
def _login(self):
username, password = self._get_login_info()
if username is None:
return
login_page = self._download_webpage(
self._LOGIN_URL, None, 'Downloading signin page')
login_form = self._hidden_inputs(login_page)
login_form.update({
'username': username,
'password': password,
})
post_url = urljoin(self._LOGIN_URL, self._search_regex(
r'<form[^>]+action=(["\'])(?P<url>.+?)\1', login_page,
'post url', default=self._LOGIN_URL, group='url'))
response, urlh = self._download_webpage_handle(
post_url, None, 'Logging in', data=urlencode_postdata(login_form),
headers={'Referer': self._LOGIN_URL})
if self._LOGIN_URL in urlh.geturl():
error = clean_html(get_element_by_class('form-message', response))
if error:
raise ExtractorError(
'Unable to login: %s' % error, expected=True)
raise ExtractorError('Unable to log in')
def _real_initialize(self):
self._login()
class MediaSelectionError(Exception):
def __init__(self, id):
self.id = id
def _extract_asx_playlist(self, connection, programme_id):
asx = self._download_xml(connection.get('href'), programme_id, 'Downloading ASX playlist')
return [ref.get('href') for ref in asx.findall('./Entry/ref')]
def _extract_items(self, playlist):
return playlist.findall('./{%s}item' % self._EMP_PLAYLIST_NS)
def _findall_ns(self, element, xpath):
elements = []
for ns in self._NAMESPACES:
elements.extend(element.findall(xpath % ns))
return elements
def _extract_medias(self, media_selection):
error = media_selection.find('./{%s}error' % self._MEDIASELECTION_NS)
if error is None:
media_selection.find('./{%s}error' % self._EMP_PLAYLIST_NS)
if error is not None:
raise BBCCoUkIE.MediaSelectionError(error.get('id'))
return self._findall_ns(media_selection, './{%s}media')
def _extract_connections(self, media):
return self._findall_ns(media, './{%s}connection')
def _get_subtitles(self, media, programme_id):
subtitles = {}
for connection in self._extract_connections(media):
cc_url = url_or_none(connection.get('href'))
if not cc_url:
continue
captions = self._download_xml(
cc_url, programme_id, 'Downloading captions', fatal=False)
if not isinstance(captions, compat_etree_Element):
continue
lang = captions.get('{http://www.w3.org/XML/1998/namespace}lang', 'en')
subtitles[lang] = [
{
'url': connection.get('href'),
'ext': 'ttml',
},
]
return subtitles
def _raise_extractor_error(self, media_selection_error):
raise ExtractorError(
'%s returned error: %s' % (self.IE_NAME, media_selection_error.id),
expected=True)
def _download_media_selector(self, programme_id):
last_exception = None
for mediaselector_url in self._MEDIASELECTOR_URLS:
try:
return self._download_media_selector_url(
mediaselector_url % programme_id, programme_id)
except BBCCoUkIE.MediaSelectionError as e:
if e.id in ('notukerror', 'geolocation', 'selectionunavailable'):
last_exception = e
continue
self._raise_extractor_error(e)
self._raise_extractor_error(last_exception)
def _download_media_selector_url(self, url, programme_id=None):
media_selection = self._download_xml(
url, programme_id, 'Downloading media selection XML',
expected_status=(403, 404))
return self._process_media_selector(media_selection, programme_id)
def _process_media_selector(self, media_selection, programme_id):
formats = []
subtitles = None
urls = []
for media in self._extract_medias(media_selection):
kind = media.get('kind')
if kind in ('video', 'audio'):
bitrate = int_or_none(media.get('bitrate'))
encoding = media.get('encoding')
service = media.get('service')
width = int_or_none(media.get('width'))
height = int_or_none(media.get('height'))
file_size = int_or_none(media.get('media_file_size'))
for connection in self._extract_connections(media):
href = connection.get('href')
if href in urls:
continue
if href:
urls.append(href)
conn_kind = connection.get('kind')
protocol = connection.get('protocol')
supplier = connection.get('supplier')
transfer_format = connection.get('transferFormat')
format_id = supplier or conn_kind or protocol
if service:
format_id = '%s_%s' % (service, format_id)
# ASX playlist
if supplier == 'asx':
for i, ref in enumerate(self._extract_asx_playlist(connection, programme_id)):
formats.append({
'url': ref,
'format_id': 'ref%s_%s' % (i, format_id),
})
elif transfer_format == 'dash':
formats.extend(self._extract_mpd_formats(
href, programme_id, mpd_id=format_id, fatal=False))
elif transfer_format == 'hls':
formats.extend(self._extract_m3u8_formats(
href, programme_id, ext='mp4', entry_protocol='m3u8_native',
m3u8_id=format_id, fatal=False))
if re.search(self._USP_RE, href):
usp_formats = self._extract_m3u8_formats(
re.sub(self._USP_RE, r'/\1.ism/\1.m3u8', href),
programme_id, ext='mp4', entry_protocol='m3u8_native',
m3u8_id=format_id, fatal=False)
for f in usp_formats:
if f.get('height') and f['height'] > 720:
continue
formats.append(f)
elif transfer_format == 'hds':
formats.extend(self._extract_f4m_formats(
href, programme_id, f4m_id=format_id, fatal=False))
else:
if not service and not supplier and bitrate:
format_id += '-%d' % bitrate
fmt = {
'format_id': format_id,
'filesize': file_size,
}
if kind == 'video':
fmt.update({
'width': width,
'height': height,
'tbr': bitrate,
'vcodec': encoding,
})
else:
fmt.update({
'abr': bitrate,
'acodec': encoding,
'vcodec': 'none',
})
if protocol in ('http', 'https'):
# Direct link
fmt.update({
'url': href,
})
elif protocol == 'rtmp':
application = connection.get('application', 'ondemand')
auth_string = connection.get('authString')
identifier = connection.get('identifier')
server = connection.get('server')
fmt.update({
'url': '%s://%s/%s?%s' % (protocol, server, application, auth_string),
'play_path': identifier,
'app': '%s?%s' % (application, auth_string),
'page_url': 'http://www.bbc.co.uk',
'player_url': 'http://www.bbc.co.uk/emp/releases/iplayer/revisions/617463_618125_4/617463_618125_4_emp.swf',
'rtmp_live': False,
'ext': 'flv',
})
else:
continue
formats.append(fmt)
elif kind == 'captions':
subtitles = self.extract_subtitles(media, programme_id)
return formats, subtitles
def _download_playlist(self, playlist_id):
try:
playlist = self._download_json(
'http://www.bbc.co.uk/programmes/%s/playlist.json' % playlist_id,
playlist_id, 'Downloading playlist JSON')
version = playlist.get('defaultAvailableVersion')
if version:
smp_config = version['smpConfig']
title = smp_config['title']
description = smp_config['summary']
for item in smp_config['items']:
kind = item['kind']
if kind not in ('programme', 'radioProgramme'):
continue
programme_id = item.get('vpid')
duration = int_or_none(item.get('duration'))
formats, subtitles = self._download_media_selector(programme_id)
return programme_id, title, description, duration, formats, subtitles
except ExtractorError as ee:
if not (isinstance(ee.cause, compat_HTTPError) and ee.cause.code == 404):
raise
# fallback to legacy playlist
return self._process_legacy_playlist(playlist_id)
def _process_legacy_playlist_url(self, url, display_id):
playlist = self._download_legacy_playlist_url(url, display_id)
return self._extract_from_legacy_playlist(playlist, display_id)
def _process_legacy_playlist(self, playlist_id):
return self._process_legacy_playlist_url(
'http://www.bbc.co.uk/iplayer/playlist/%s' % playlist_id, playlist_id)
def _download_legacy_playlist_url(self, url, playlist_id=None):
return self._download_xml(
url, playlist_id, 'Downloading legacy playlist XML')
def _extract_from_legacy_playlist(self, playlist, playlist_id):
no_items = playlist.find('./{%s}noItems' % self._EMP_PLAYLIST_NS)
if no_items is not None:
reason = no_items.get('reason')
if reason == 'preAvailability':
msg = 'Episode %s is not yet available' % playlist_id
elif reason == 'postAvailability':
msg = 'Episode %s is no longer available' % playlist_id
elif reason == 'noMedia':
msg = 'Episode %s is not currently available' % playlist_id
else:
msg = 'Episode %s is not available: %s' % (playlist_id, reason)
raise ExtractorError(msg, expected=True)
for item in self._extract_items(playlist):
kind = item.get('kind')
if kind not in ('programme', 'radioProgramme'):
continue
title = playlist.find('./{%s}title' % self._EMP_PLAYLIST_NS).text
description_el = playlist.find('./{%s}summary' % self._EMP_PLAYLIST_NS)
description = description_el.text if description_el is not None else None
def get_programme_id(item):
def get_from_attributes(item):
for p in ('identifier', 'group'):
value = item.get(p)
if value and re.match(r'^[pb][\da-z]{7}$', value):
return value
get_from_attributes(item)
mediator = item.find('./{%s}mediator' % self._EMP_PLAYLIST_NS)
if mediator is not None:
return get_from_attributes(mediator)
programme_id = get_programme_id(item)
duration = int_or_none(item.get('duration'))
if programme_id:
formats, subtitles = self._download_media_selector(programme_id)
else:
formats, subtitles = self._process_media_selector(item, playlist_id)
programme_id = playlist_id
return programme_id, title, description, duration, formats, subtitles
def _real_extract(self, url):
group_id = self._match_id(url)
webpage = self._download_webpage(url, group_id, 'Downloading video page')
error = self._search_regex(
r'<div\b[^>]+\bclass=["\']smp__message delta["\'][^>]*>([^<]+)<',
webpage, 'error', default=None)
if error:
raise ExtractorError(error, expected=True)
programme_id = None
duration = None
tviplayer = self._search_regex(
r'mediator\.bind\(({.+?})\s*,\s*document\.getElementById',
webpage, 'player', default=None)
if tviplayer:
player = self._parse_json(tviplayer, group_id).get('player', {})
duration = int_or_none(player.get('duration'))
programme_id = player.get('vpid')
if not programme_id:
programme_id = self._search_regex(
r'"vpid"\s*:\s*"(%s)"' % self._ID_REGEX, webpage, 'vpid', fatal=False, default=None)
if programme_id:
formats, subtitles = self._download_media_selector(programme_id)
title = self._og_search_title(webpage, default=None) or self._html_search_regex(
(r'<h2[^>]+id="parent-title"[^>]*>(.+?)</h2>',
r'<div[^>]+class="info"[^>]*>\s*<h1>(.+?)</h1>'), webpage, 'title')
description = self._search_regex(
(r'<p class="[^"]*medium-description[^"]*">([^<]+)</p>',
r'<div[^>]+class="info_+synopsis"[^>]*>([^<]+)</div>'),
webpage, 'description', default=None)
if not description:
description = self._html_search_meta('description', webpage)
else:
programme_id, title, description, duration, formats, subtitles = self._download_playlist(group_id)
self._sort_formats(formats)
return {
'id': programme_id,
'title': title,
'description': description,
'thumbnail': self._og_search_thumbnail(webpage, default=None),
'duration': duration,
'formats': formats,
'subtitles': subtitles,
}
class BBCIE(BBCCoUkIE):
IE_NAME = 'bbc'
IE_DESC = 'BBC'
_VALID_URL = r'https?://(?:www\.)?bbc\.(?:com|co\.uk)/(?:[^/]+/)+(?P<id>[^/#?]+)'
_MEDIASELECTOR_URLS = [
# Provides HQ HLS streams but fails with geolocation in some cases when it's
# even not geo restricted at all
'http://open.live.bbc.co.uk/mediaselector/5/select/version/2.0/mediaset/iptv-all/vpid/%s',
# Provides more formats, namely direct mp4 links, but fails on some videos with
# notukerror for non UK (?) users (e.g.
# http://www.bbc.com/travel/story/20150625-sri-lankas-spicy-secret)
'http://open.live.bbc.co.uk/mediaselector/4/mtis/stream/%s',
# Provides fewer formats, but works everywhere for everybody (hopefully)
'http://open.live.bbc.co.uk/mediaselector/5/select/version/2.0/mediaset/journalism-pc/vpid/%s',
]
_TESTS = [{
# article with multiple videos embedded with data-playable containing vpids
'url': 'http://www.bbc.com/news/world-europe-32668511',
'info_dict': {
'id': 'world-europe-32668511',
'title': 'Russia stages massive WW2 parade',
'description': 'md5:00ff61976f6081841f759a08bf78cc9c',
},
'playlist_count': 2,
}, {
# article with multiple videos embedded with data-playable (more videos)
'url': 'http://www.bbc.com/news/business-28299555',
'info_dict': {
'id': 'business-28299555',
'title': 'Farnborough Airshow: Video highlights',
'description': 'BBC reports and video highlights at the Farnborough Airshow.',
},
'playlist_count': 9,
'skip': 'Save time',
}, {
# article with multiple videos embedded with `new SMP()`
# broken
'url': 'http://www.bbc.co.uk/blogs/adamcurtis/entries/3662a707-0af9-3149-963f-47bea720b460',
'info_dict': {
'id': '3662a707-0af9-3149-963f-47bea720b460',
'title': 'BUGGER',
},
'playlist_count': 18,
}, {
# single video embedded with data-playable containing vpid
'url': 'http://www.bbc.com/news/world-europe-32041533',
'info_dict': {
'id': 'p02mprgb',
'ext': 'mp4',
'title': 'Aerial footage showed the site of the crash in the Alps - courtesy BFM TV',
'description': 'md5:2868290467291b37feda7863f7a83f54',
'duration': 47,
'timestamp': 1427219242,
'upload_date': '20150324',
},
'params': {
# rtmp download
'skip_download': True,
}
}, {
# article with single video embedded with data-playable containing XML playlist
# with direct video links as progressiveDownloadUrl (for now these are extracted)
# and playlist with f4m and m3u8 as streamingUrl
'url': 'http://www.bbc.com/turkce/haberler/2015/06/150615_telabyad_kentin_cogu',
'info_dict': {
'id': '150615_telabyad_kentin_cogu',
'ext': 'mp4',
'title': "YPG: Tel Abyad'ın tamamı kontrolümüzde",
'description': 'md5:33a4805a855c9baf7115fcbde57e7025',
'timestamp': 1434397334,
'upload_date': '20150615',
},
'params': {
'skip_download': True,
}
}, {
# single video embedded with data-playable containing XML playlists (regional section)
'url': 'http://www.bbc.com/mundo/video_fotos/2015/06/150619_video_honduras_militares_hospitales_corrupcion_aw',
'info_dict': {
'id': '150619_video_honduras_militares_hospitales_corrupcion_aw',
'ext': 'mp4',
'title': 'Honduras militariza sus hospitales por nuevo escándalo de corrupción',
'description': 'md5:1525f17448c4ee262b64b8f0c9ce66c8',
'timestamp': 1434713142,
'upload_date': '20150619',
},
'params': {
'skip_download': True,
}
}, {
# single video from video playlist embedded with vxp-playlist-data JSON
'url': 'http://www.bbc.com/news/video_and_audio/must_see/33376376',
'info_dict': {
'id': 'p02w6qjc',
'ext': 'mp4',
'title': '''Judge Mindy Glazer: "I'm sorry to see you here... I always wondered what happened to you"''',
'duration': 56,
'description': '''Judge Mindy Glazer: "I'm sorry to see you here... I always wondered what happened to you"''',
},
'params': {
'skip_download': True,
}
}, {
# single video story with digitalData
'url': 'http://www.bbc.com/travel/story/20150625-sri-lankas-spicy-secret',
'info_dict': {
'id': 'p02q6gc4',
'ext': 'flv',
'title': 'Sri Lankas spicy secret',
'description': 'As a new train line to Jaffna opens up the countrys north, travellers can experience a truly distinct slice of Tamil culture.',
'timestamp': 1437674293,
'upload_date': '20150723',
},
'params': {
# rtmp download
'skip_download': True,
}
}, {
# single video story without digitalData
'url': 'http://www.bbc.com/autos/story/20130513-hyundais-rock-star',
'info_dict': {
'id': 'p018zqqg',
'ext': 'mp4',
'title': 'Hyundai Santa Fe Sport: Rock star',
'description': 'md5:b042a26142c4154a6e472933cf20793d',
'timestamp': 1415867444,
'upload_date': '20141113',
},
'params': {
# rtmp download
'skip_download': True,
}
}, {
# single video embedded with Morph
'url': 'http://www.bbc.co.uk/sport/live/olympics/36895975',
'info_dict': {
'id': 'p041vhd0',
'ext': 'mp4',
'title': "Nigeria v Japan - Men's First Round",
'description': 'Live coverage of the first round from Group B at the Amazonia Arena.',
'duration': 7980,
'uploader': 'BBC Sport',
'uploader_id': 'bbc_sport',
},
'params': {
# m3u8 download
'skip_download': True,
},
'skip': 'Georestricted to UK',
}, {
# single video with playlist.sxml URL in playlist param
'url': 'http://www.bbc.com/sport/0/football/33653409',
'info_dict': {
'id': 'p02xycnp',
'ext': 'mp4',
'title': 'Transfers: Cristiano Ronaldo to Man Utd, Arsenal to spend?',
'description': 'BBC Sport\'s David Ornstein has the latest transfer gossip, including rumours of a Manchester United return for Cristiano Ronaldo.',
'duration': 140,
},
'params': {
# rtmp download
'skip_download': True,
}
}, {
# article with multiple videos embedded with playlist.sxml in playlist param
'url': 'http://www.bbc.com/sport/0/football/34475836',
'info_dict': {
'id': '34475836',
'title': 'Jurgen Klopp: Furious football from a witty and winning coach',
'description': 'Fast-paced football, wit, wisdom and a ready smile - why Liverpool fans should come to love new boss Jurgen Klopp.',
},
'playlist_count': 3,
}, {
# school report article with single video
'url': 'http://www.bbc.co.uk/schoolreport/35744779',
'info_dict': {
'id': '35744779',
'title': 'School which breaks down barriers in Jerusalem',
},
'playlist_count': 1,
}, {
# single video with playlist URL from weather section
'url': 'http://www.bbc.com/weather/features/33601775',
'only_matching': True,
}, {
# custom redirection to www.bbc.com
'url': 'http://www.bbc.co.uk/news/science-environment-33661876',
'only_matching': True,
}, {
# single video article embedded with data-media-vpid
'url': 'http://www.bbc.co.uk/sport/rowing/35908187',
'only_matching': True,
}, {
'url': 'https://www.bbc.co.uk/bbcthree/clip/73d0bbd0-abc3-4cea-b3c0-cdae21905eb1',
'info_dict': {
'id': 'p06556y7',
'ext': 'mp4',
'title': 'Transfers: Cristiano Ronaldo to Man Utd, Arsenal to spend?',
'description': 'md5:4b7dfd063d5a789a1512e99662be3ddd',
},
'params': {
'skip_download': True,
}
}, {
# window.__PRELOADED_STATE__
'url': 'https://www.bbc.co.uk/radio/play/b0b9z4yl',
'info_dict': {
'id': 'b0b9z4vz',
'ext': 'mp4',
'title': 'Prom 6: An American in Paris and Turangalila',
'description': 'md5:51cf7d6f5c8553f197e58203bc78dff8',
'uploader': 'Radio 3',
'uploader_id': 'bbc_radio_three',
},
}, {
'url': 'http://www.bbc.co.uk/learningenglish/chinese/features/lingohack/ep-181227',
'info_dict': {
'id': 'p06w9tws',
'ext': 'mp4',
'title': 'md5:2fabf12a726603193a2879a055f72514',
'description': 'Learn English words and phrases from this story',
},
'add_ie': [BBCCoUkIE.ie_key()],
}]
@classmethod
def suitable(cls, url):
EXCLUDE_IE = (BBCCoUkIE, BBCCoUkArticleIE, BBCCoUkIPlayerPlaylistIE, BBCCoUkPlaylistIE)
return (False if any(ie.suitable(url) for ie in EXCLUDE_IE)
else super(BBCIE, cls).suitable(url))
def _extract_from_media_meta(self, media_meta, video_id):
# Direct links to media in media metadata (e.g.
# http://www.bbc.com/turkce/haberler/2015/06/150615_telabyad_kentin_cogu)
# TODO: there are also f4m and m3u8 streams incorporated in playlist.sxml
source_files = media_meta.get('sourceFiles')
if source_files:
return [{
'url': f['url'],
'format_id': format_id,
'ext': f.get('encoding'),
'tbr': float_or_none(f.get('bitrate'), 1000),
'filesize': int_or_none(f.get('filesize')),
} for format_id, f in source_files.items() if f.get('url')], []
programme_id = media_meta.get('externalId')
if programme_id:
return self._download_media_selector(programme_id)
# Process playlist.sxml as legacy playlist
href = media_meta.get('href')
if href:
playlist = self._download_legacy_playlist_url(href)
_, _, _, _, formats, subtitles = self._extract_from_legacy_playlist(playlist, video_id)
return formats, subtitles
return [], []
def _extract_from_playlist_sxml(self, url, playlist_id, timestamp):
programme_id, title, description, duration, formats, subtitles = \
self._process_legacy_playlist_url(url, playlist_id)
self._sort_formats(formats)
return {
'id': programme_id,
'title': title,
'description': description,
'duration': duration,
'timestamp': timestamp,
'formats': formats,
'subtitles': subtitles,
}
def _real_extract(self, url):
playlist_id = self._match_id(url)
webpage = self._download_webpage(url, playlist_id)
json_ld_info = self._search_json_ld(webpage, playlist_id, default={})
timestamp = json_ld_info.get('timestamp')
playlist_title = json_ld_info.get('title')
if not playlist_title:
playlist_title = self._og_search_title(
webpage, default=None) or self._html_search_regex(
r'<title>(.+?)</title>', webpage, 'playlist title', default=None)
if playlist_title:
playlist_title = re.sub(r'(.+)\s*-\s*BBC.*?$', r'\1', playlist_title).strip()
playlist_description = json_ld_info.get(
'description') or self._og_search_description(webpage, default=None)
if not timestamp:
timestamp = parse_iso8601(self._search_regex(
[r'<meta[^>]+property="article:published_time"[^>]+content="([^"]+)"',
r'itemprop="datePublished"[^>]+datetime="([^"]+)"',
r'"datePublished":\s*"([^"]+)'],
webpage, 'date', default=None))
entries = []
# article with multiple videos embedded with playlist.sxml (e.g.
# http://www.bbc.com/sport/0/football/34475836)
playlists = re.findall(r'<param[^>]+name="playlist"[^>]+value="([^"]+)"', webpage)
playlists.extend(re.findall(r'data-media-id="([^"]+/playlist\.sxml)"', webpage))
if playlists:
entries = [
self._extract_from_playlist_sxml(playlist_url, playlist_id, timestamp)
for playlist_url in playlists]
# news article with multiple videos embedded with data-playable
data_playables = re.findall(r'data-playable=(["\'])({.+?})\1', webpage)
if data_playables:
for _, data_playable_json in data_playables:
data_playable = self._parse_json(
unescapeHTML(data_playable_json), playlist_id, fatal=False)
if not data_playable:
continue
settings = data_playable.get('settings', {})
if settings:
# data-playable with video vpid in settings.playlistObject.items (e.g.
# http://www.bbc.com/news/world-us-canada-34473351)
playlist_object = settings.get('playlistObject', {})
if playlist_object:
items = playlist_object.get('items')
if items and isinstance(items, list):
title = playlist_object['title']
description = playlist_object.get('summary')
duration = int_or_none(items[0].get('duration'))
programme_id = items[0].get('vpid')
formats, subtitles = self._download_media_selector(programme_id)
self._sort_formats(formats)
entries.append({
'id': programme_id,
'title': title,
'description': description,
'timestamp': timestamp,
'duration': duration,
'formats': formats,
'subtitles': subtitles,
})
else:
# data-playable without vpid but with a playlist.sxml URLs
# in otherSettings.playlist (e.g.
# http://www.bbc.com/turkce/multimedya/2015/10/151010_vid_ankara_patlama_ani)
playlist = data_playable.get('otherSettings', {}).get('playlist', {})
if playlist:
entry = None
for key in ('streaming', 'progressiveDownload'):
playlist_url = playlist.get('%sUrl' % key)
if not playlist_url:
continue
try:
info = self._extract_from_playlist_sxml(
playlist_url, playlist_id, timestamp)
if not entry:
entry = info
else:
entry['title'] = info['title']
entry['formats'].extend(info['formats'])
except Exception as e:
# Some playlist URL may fail with 500, at the same time
# the other one may work fine (e.g.
# http://www.bbc.com/turkce/haberler/2015/06/150615_telabyad_kentin_cogu)
if isinstance(e.cause, compat_HTTPError) and e.cause.code == 500:
continue
raise
if entry:
self._sort_formats(entry['formats'])
entries.append(entry)
if entries:
return self.playlist_result(entries, playlist_id, playlist_title, playlist_description)
# http://www.bbc.co.uk/learningenglish/chinese/features/lingohack/ep-181227
group_id = self._search_regex(
r'<div[^>]+\bclass=["\']video["\'][^>]+\bdata-pid=["\'](%s)' % self._ID_REGEX,
webpage, 'group id', default=None)
if playlist_id:
return self.url_result(
'https://www.bbc.co.uk/programmes/%s' % group_id,
ie=BBCCoUkIE.ie_key())
# single video story (e.g. http://www.bbc.com/travel/story/20150625-sri-lankas-spicy-secret)
programme_id = self._search_regex(
[r'data-(?:video-player|media)-vpid="(%s)"' % self._ID_REGEX,
r'<param[^>]+name="externalIdentifier"[^>]+value="(%s)"' % self._ID_REGEX,
r'videoId\s*:\s*["\'](%s)["\']' % self._ID_REGEX],
webpage, 'vpid', default=None)
if programme_id:
formats, subtitles = self._download_media_selector(programme_id)
self._sort_formats(formats)
# digitalData may be missing (e.g. http://www.bbc.com/autos/story/20130513-hyundais-rock-star)
digital_data = self._parse_json(
self._search_regex(
r'var\s+digitalData\s*=\s*({.+?});?\n', webpage, 'digital data', default='{}'),
programme_id, fatal=False)
page_info = digital_data.get('page', {}).get('pageInfo', {})
title = page_info.get('pageName') or self._og_search_title(webpage)
description = page_info.get('description') or self._og_search_description(webpage)
timestamp = parse_iso8601(page_info.get('publicationDate')) or timestamp
return {
'id': programme_id,
'title': title,
'description': description,
'timestamp': timestamp,
'formats': formats,
'subtitles': subtitles,
}
# Morph based embed (e.g. http://www.bbc.co.uk/sport/live/olympics/36895975)
# There are several setPayload calls may be present but the video
# seems to be always related to the first one
morph_payload = self._parse_json(
self._search_regex(
r'Morph\.setPayload\([^,]+,\s*({.+?})\);',
webpage, 'morph payload', default='{}'),
playlist_id, fatal=False)
if morph_payload:
components = try_get(morph_payload, lambda x: x['body']['components'], list) or []
for component in components:
if not isinstance(component, dict):
continue
lead_media = try_get(component, lambda x: x['props']['leadMedia'], dict)
if not lead_media:
continue
identifiers = lead_media.get('identifiers')
if not identifiers or not isinstance(identifiers, dict):
continue
programme_id = identifiers.get('vpid') or identifiers.get('playablePid')
if not programme_id:
continue
title = lead_media.get('title') or self._og_search_title(webpage)
formats, subtitles = self._download_media_selector(programme_id)
self._sort_formats(formats)
description = lead_media.get('summary')
uploader = lead_media.get('masterBrand')
uploader_id = lead_media.get('mid')
duration = None
duration_d = lead_media.get('duration')
if isinstance(duration_d, dict):
duration = parse_duration(dict_get(
duration_d, ('rawDuration', 'formattedDuration', 'spokenDuration')))
return {
'id': programme_id,
'title': title,
'description': description,
'duration': duration,
'uploader': uploader,
'uploader_id': uploader_id,
'formats': formats,
'subtitles': subtitles,
}
preload_state = self._parse_json(self._search_regex(
r'window\.__PRELOADED_STATE__\s*=\s*({.+?});', webpage,
'preload state', default='{}'), playlist_id, fatal=False)
if preload_state:
current_programme = preload_state.get('programmes', {}).get('current') or {}
programme_id = current_programme.get('id')
if current_programme and programme_id and current_programme.get('type') == 'playable_item':
title = current_programme.get('titles', {}).get('tertiary') or playlist_title
formats, subtitles = self._download_media_selector(programme_id)
self._sort_formats(formats)
synopses = current_programme.get('synopses') or {}
network = current_programme.get('network') or {}
duration = int_or_none(
current_programme.get('duration', {}).get('value'))
thumbnail = None
image_url = current_programme.get('image_url')
if image_url:
thumbnail = image_url.replace('{recipe}', '1920x1920')
return {
'id': programme_id,
'title': title,
'description': dict_get(synopses, ('long', 'medium', 'short')),
'thumbnail': thumbnail,
'duration': duration,
'uploader': network.get('short_title'),
'uploader_id': network.get('id'),
'formats': formats,
'subtitles': subtitles,
}
bbc3_config = self._parse_json(
self._search_regex(
r'(?s)bbcthreeConfig\s*=\s*({.+?})\s*;\s*<', webpage,
'bbcthree config', default='{}'),
playlist_id, transform_source=js_to_json, fatal=False)
if bbc3_config:
bbc3_playlist = try_get(
bbc3_config, lambda x: x['payload']['content']['bbcMedia']['playlist'],
dict)
if bbc3_playlist:
playlist_title = bbc3_playlist.get('title') or playlist_title
thumbnail = bbc3_playlist.get('holdingImageURL')
entries = []
for bbc3_item in bbc3_playlist['items']:
programme_id = bbc3_item.get('versionID')
if not programme_id:
continue
formats, subtitles = self._download_media_selector(programme_id)
self._sort_formats(formats)
entries.append({
'id': programme_id,
'title': playlist_title,
'thumbnail': thumbnail,
'timestamp': timestamp,
'formats': formats,
'subtitles': subtitles,
})
return self.playlist_result(
entries, playlist_id, playlist_title, playlist_description)
def extract_all(pattern):
return list(filter(None, map(
lambda s: self._parse_json(s, playlist_id, fatal=False),
re.findall(pattern, webpage))))
# Multiple video article (e.g.
# http://www.bbc.co.uk/blogs/adamcurtis/entries/3662a707-0af9-3149-963f-47bea720b460)
EMBED_URL = r'https?://(?:www\.)?bbc\.co\.uk/(?:[^/]+/)+%s(?:\b[^"]+)?' % self._ID_REGEX
entries = []
for match in extract_all(r'new\s+SMP\(({.+?})\)'):
embed_url = match.get('playerSettings', {}).get('externalEmbedUrl')
if embed_url and re.match(EMBED_URL, embed_url):
entries.append(embed_url)
entries.extend(re.findall(
r'setPlaylist\("(%s)"\)' % EMBED_URL, webpage))
if entries:
return self.playlist_result(
[self.url_result(entry_, 'BBCCoUk') for entry_ in entries],
playlist_id, playlist_title, playlist_description)
# Multiple video article (e.g. http://www.bbc.com/news/world-europe-32668511)
medias = extract_all(r"data-media-meta='({[^']+})'")
if not medias:
# Single video article (e.g. http://www.bbc.com/news/video_and_audio/international)
media_asset = self._search_regex(
r'mediaAssetPage\.init\(\s*({.+?}), "/',
webpage, 'media asset', default=None)
if media_asset:
media_asset_page = self._parse_json(media_asset, playlist_id, fatal=False)
medias = []
for video in media_asset_page.get('videos', {}).values():
medias.extend(video.values())
if not medias:
# Multiple video playlist with single `now playing` entry (e.g.
# http://www.bbc.com/news/video_and_audio/must_see/33767813)
vxp_playlist = self._parse_json(
self._search_regex(
r'<script[^>]+class="vxp-playlist-data"[^>]+type="application/json"[^>]*>([^<]+)</script>',
webpage, 'playlist data'),
playlist_id)
playlist_medias = []
for item in vxp_playlist:
media = item.get('media')
if not media:
continue
playlist_medias.append(media)
# Download single video if found media with asset id matching the video id from URL
if item.get('advert', {}).get('assetId') == playlist_id:
medias = [media]
break
# Fallback to the whole playlist
if not medias:
medias = playlist_medias
entries = []
for num, media_meta in enumerate(medias, start=1):
formats, subtitles = self._extract_from_media_meta(media_meta, playlist_id)
if not formats:
continue
self._sort_formats(formats)
video_id = media_meta.get('externalId')
if not video_id:
video_id = playlist_id if len(medias) == 1 else '%s-%s' % (playlist_id, num)
title = media_meta.get('caption')
if not title:
title = playlist_title if len(medias) == 1 else '%s - Video %s' % (playlist_title, num)
duration = int_or_none(media_meta.get('durationInSeconds')) or parse_duration(media_meta.get('duration'))
images = []
for image in media_meta.get('images', {}).values():
images.extend(image.values())
if 'image' in media_meta:
images.append(media_meta['image'])
thumbnails = [{
'url': image.get('href'),
'width': int_or_none(image.get('width')),
'height': int_or_none(image.get('height')),
} for image in images]
entries.append({
'id': video_id,
'title': title,
'thumbnails': thumbnails,
'duration': duration,
'timestamp': timestamp,
'formats': formats,
'subtitles': subtitles,
})
return self.playlist_result(entries, playlist_id, playlist_title, playlist_description)
class BBCCoUkArticleIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?bbc\.co\.uk/programmes/articles/(?P<id>[a-zA-Z0-9]+)'
IE_NAME = 'bbc.co.uk:article'
IE_DESC = 'BBC articles'
_TEST = {
'url': 'http://www.bbc.co.uk/programmes/articles/3jNQLTMrPlYGTBn0WV6M2MS/not-your-typical-role-model-ada-lovelace-the-19th-century-programmer',
'info_dict': {
'id': '3jNQLTMrPlYGTBn0WV6M2MS',
'title': 'Calculating Ada: The Countess of Computing - Not your typical role model: Ada Lovelace the 19th century programmer - BBC Four',
'description': 'Hannah Fry reveals some of her surprising discoveries about Ada Lovelace during filming.',
},
'playlist_count': 4,
'add_ie': ['BBCCoUk'],
}
def _real_extract(self, url):
playlist_id = self._match_id(url)
webpage = self._download_webpage(url, playlist_id)
title = self._og_search_title(webpage)
description = self._og_search_description(webpage).strip()
entries = [self.url_result(programme_url) for programme_url in re.findall(
r'<div[^>]+typeof="Clip"[^>]+resource="([^"]+)"', webpage)]
return self.playlist_result(entries, playlist_id, title, description)
class BBCCoUkPlaylistBaseIE(InfoExtractor):
def _entries(self, webpage, url, playlist_id):
single_page = 'page' in compat_urlparse.parse_qs(
compat_urlparse.urlparse(url).query)
for page_num in itertools.count(2):
for video_id in re.findall(
self._VIDEO_ID_TEMPLATE % BBCCoUkIE._ID_REGEX, webpage):
yield self.url_result(
self._URL_TEMPLATE % video_id, BBCCoUkIE.ie_key())
if single_page:
return
next_page = self._search_regex(
r'<li[^>]+class=(["\'])pagination_+next\1[^>]*><a[^>]+href=(["\'])(?P<url>(?:(?!\2).)+)\2',
webpage, 'next page url', default=None, group='url')
if not next_page:
break
webpage = self._download_webpage(
compat_urlparse.urljoin(url, next_page), playlist_id,
'Downloading page %d' % page_num, page_num)
def _real_extract(self, url):
playlist_id = self._match_id(url)
webpage = self._download_webpage(url, playlist_id)
title, description = self._extract_title_and_description(webpage)
return self.playlist_result(
self._entries(webpage, url, playlist_id),
playlist_id, title, description)
class BBCCoUkIPlayerPlaylistIE(BBCCoUkPlaylistBaseIE):
IE_NAME = 'bbc.co.uk:iplayer:playlist'
_VALID_URL = r'https?://(?:www\.)?bbc\.co\.uk/iplayer/(?:episodes|group)/(?P<id>%s)' % BBCCoUkIE._ID_REGEX
_URL_TEMPLATE = 'http://www.bbc.co.uk/iplayer/episode/%s'
_VIDEO_ID_TEMPLATE = r'data-ip-id=["\'](%s)'
_TESTS = [{
'url': 'http://www.bbc.co.uk/iplayer/episodes/b05rcz9v',
'info_dict': {
'id': 'b05rcz9v',
'title': 'The Disappearance',
'description': 'French thriller serial about a missing teenager.',
},
'playlist_mincount': 6,
'skip': 'This programme is not currently available on BBC iPlayer',
}, {
# Available for over a year unlike 30 days for most other programmes
'url': 'http://www.bbc.co.uk/iplayer/group/p02tcc32',
'info_dict': {
'id': 'p02tcc32',
'title': 'Bohemian Icons',
'description': 'md5:683e901041b2fe9ba596f2ab04c4dbe7',
},
'playlist_mincount': 10,
}]
def _extract_title_and_description(self, webpage):
title = self._search_regex(r'<h1>([^<]+)</h1>', webpage, 'title', fatal=False)
description = self._search_regex(
r'<p[^>]+class=(["\'])subtitle\1[^>]*>(?P<value>[^<]+)</p>',
webpage, 'description', fatal=False, group='value')
return title, description
class BBCCoUkPlaylistIE(BBCCoUkPlaylistBaseIE):
IE_NAME = 'bbc.co.uk:playlist'
_VALID_URL = r'https?://(?:www\.)?bbc\.co\.uk/programmes/(?P<id>%s)/(?:episodes|broadcasts|clips)' % BBCCoUkIE._ID_REGEX
_URL_TEMPLATE = 'http://www.bbc.co.uk/programmes/%s'
_VIDEO_ID_TEMPLATE = r'data-pid=["\'](%s)'
_TESTS = [{
'url': 'http://www.bbc.co.uk/programmes/b05rcz9v/clips',
'info_dict': {
'id': 'b05rcz9v',
'title': 'The Disappearance - Clips - BBC Four',
'description': 'French thriller serial about a missing teenager.',
},
'playlist_mincount': 7,
}, {
# multipage playlist, explicit page
'url': 'http://www.bbc.co.uk/programmes/b00mfl7n/clips?page=1',
'info_dict': {
'id': 'b00mfl7n',
'title': 'Frozen Planet - Clips - BBC One',
'description': 'md5:65dcbf591ae628dafe32aa6c4a4a0d8c',
},
'playlist_mincount': 24,
}, {
# multipage playlist, all pages
'url': 'http://www.bbc.co.uk/programmes/b00mfl7n/clips',
'info_dict': {
'id': 'b00mfl7n',
'title': 'Frozen Planet - Clips - BBC One',
'description': 'md5:65dcbf591ae628dafe32aa6c4a4a0d8c',
},
'playlist_mincount': 142,
}, {
'url': 'http://www.bbc.co.uk/programmes/b05rcz9v/broadcasts/2016/06',
'only_matching': True,
}, {
'url': 'http://www.bbc.co.uk/programmes/b05rcz9v/clips',
'only_matching': True,
}, {
'url': 'http://www.bbc.co.uk/programmes/b055jkys/episodes/player',
'only_matching': True,
}]
def _extract_title_and_description(self, webpage):
title = self._og_search_title(webpage, fatal=False)
description = self._og_search_description(webpage)
return title, description