2014-01-07 09:04:48 +00:00
from __future__ import unicode_literals
2021-01-10 21:37:45 +00:00
import re
import json
2017-02-16 15:42:36 +00:00
from . common import InfoExtractor
2021-01-10 21:37:45 +00:00
from . . compat import compat_urllib_parse_unquote_plus
2016-08-05 15:40:21 +00:00
from . . utils import (
2021-01-10 21:37:45 +00:00
KNOWN_EXTENSIONS ,
extract_attributes ,
2016-08-05 15:40:21 +00:00
unified_strdate ,
2021-01-10 21:37:45 +00:00
unified_timestamp ,
2016-08-05 15:40:21 +00:00
clean_html ,
2021-01-10 21:37:45 +00:00
dict_get ,
parse_duration ,
int_or_none ,
str_or_none ,
merge_dicts ,
2016-08-05 15:40:21 +00:00
)
2013-07-08 00:04:11 +00:00
2017-02-16 15:42:36 +00:00
class ArchiveOrgIE ( InfoExtractor ) :
2013-07-08 00:04:11 +00:00
IE_NAME = ' archive.org '
2021-01-10 21:37:45 +00:00
IE_DESC = ' archive.org video and audio '
_VALID_URL = r ' https?://(?:www \ .)?archive \ .org/(?:details|embed)/(?P<id>[^?#]+)(?:[?].*)?$ '
2014-12-28 20:08:46 +00:00
_TESTS = [ {
' url ' : ' http://archive.org/details/XD300-23_68HighlightsAResearchCntAugHumanIntellect ' ,
2014-01-07 09:04:48 +00:00
' md5 ' : ' 8af1d4cf447933ed3c7f4871162602db ' ,
' info_dict ' : {
2014-12-28 20:08:46 +00:00
' id ' : ' XD300-23_68HighlightsAResearchCntAugHumanIntellect ' ,
2021-01-10 21:37:45 +00:00
' ext ' : ' ogv ' ,
2014-12-28 20:08:46 +00:00
' title ' : ' 1968 Demo - FJCC Conference Presentation Reel #1 ' ,
2016-08-05 15:40:21 +00:00
' description ' : ' md5:da45c349df039f1cc8075268eb1b5c25 ' ,
2021-01-10 21:37:45 +00:00
' release_date ' : ' 19681210 ' ,
' timestamp ' : 1268695290 ,
' upload_date ' : ' 20100315 ' ,
' creator ' : ' SRI International ' ,
' uploader ' : ' laura@archive.org ' ,
} ,
2014-12-28 20:08:46 +00:00
} , {
' url ' : ' https://archive.org/details/Cops1922 ' ,
2017-04-27 20:48:32 +00:00
' md5 ' : ' 0869000b4ce265e8ca62738b336b268a ' ,
2014-12-28 20:08:46 +00:00
' info_dict ' : {
' id ' : ' Cops1922 ' ,
2016-08-05 15:40:21 +00:00
' ext ' : ' mp4 ' ,
2014-12-28 20:08:46 +00:00
' title ' : ' Buster Keaton \' s " Cops " (1922) ' ,
2021-01-10 21:37:45 +00:00
' description ' : ' md5:43a603fd6c5b4b90d12a96b921212b9c ' ,
' uploader ' : ' yorkmba99@hotmail.com ' ,
' timestamp ' : 1387699629 ,
' upload_date ' : " 20131222 " ,
} ,
2016-08-05 15:40:21 +00:00
} , {
' url ' : ' http://archive.org/embed/XD300-23_68HighlightsAResearchCntAugHumanIntellect ' ,
' only_matching ' : True ,
2021-01-10 21:37:45 +00:00
} , {
' url ' : ' https://archive.org/details/Election_Ads ' ,
' md5 ' : ' 284180e857160cf866358700bab668a3 ' ,
' info_dict ' : {
' id ' : ' Election_Ads/Commercial-JFK1960ElectionAdCampaignJingle.mpg ' ,
' title ' : ' Commercial-JFK1960ElectionAdCampaignJingle.mpg ' ,
' ext ' : ' mp4 ' ,
} ,
} , {
' url ' : ' https://archive.org/details/Election_Ads/Commercial-Nixon1960ElectionAdToughonDefense.mpg ' ,
' md5 ' : ' 7915213ef02559b5501fe630e1a53f59 ' ,
' info_dict ' : {
' id ' : ' Election_Ads/Commercial-Nixon1960ElectionAdToughonDefense.mpg ' ,
' title ' : ' Commercial-Nixon1960ElectionAdToughonDefense.mpg ' ,
' ext ' : ' mp4 ' ,
' timestamp ' : 1205588045 ,
' uploader ' : ' mikedavisstripmaster@yahoo.com ' ,
' description ' : ' 1960 Presidential Campaign Election Commercials John F Kennedy, Richard M Nixon ' ,
' upload_date ' : ' 20080315 ' ,
} ,
} , {
' url ' : ' https://archive.org/details/gd1977-05-08.shure57.stevenson.29303.flac16 ' ,
' md5 ' : ' 7d07ffb42aba6537c28e053efa4b54c9 ' ,
' info_dict ' : {
' id ' : ' gd1977-05-08.shure57.stevenson.29303.flac16/gd1977-05-08d01t01.flac ' ,
' title ' : ' Turning ' ,
' ext ' : ' flac ' ,
} ,
} , {
' url ' : ' https://archive.org/details/gd1977-05-08.shure57.stevenson.29303.flac16/gd1977-05-08d01t07.flac ' ,
' md5 ' : ' a07cd8c6ab4ee1560f8a0021717130f3 ' ,
' info_dict ' : {
' id ' : ' gd1977-05-08.shure57.stevenson.29303.flac16/gd1977-05-08d01t07.flac ' ,
' title ' : ' Deal ' ,
' ext ' : ' flac ' ,
' timestamp ' : 1205895624 ,
' uploader ' : ' mvernon54@yahoo.com ' ,
' description ' : ' md5:6a31f1996db0aa0fc9da6d6e708a1bb0 ' ,
' upload_date ' : ' 20080319 ' ,
' location ' : ' Barton Hall - Cornell University ' ,
} ,
} , {
' url ' : ' https://archive.org/details/lp_the-music-of-russia_various-artists-a-askaryan-alexander-melik ' ,
' md5 ' : ' 7cb019baa9b332e82ea7c10403acd180 ' ,
' info_dict ' : {
' id ' : ' lp_the-music-of-russia_various-artists-a-askaryan-alexander-melik/disc1/01.01. Bells Of Rostov.mp3 ' ,
' title ' : ' Bells Of Rostov ' ,
' ext ' : ' mp3 ' ,
} ,
} , {
' url ' : ' https://archive.org/details/lp_the-music-of-russia_various-artists-a-askaryan-alexander-melik/disc1/02.02.+Song+And+Chorus+In+The+Polovetsian+Camp+From+ % 22Prince+Igor % 22+(Act+2 % 2C+Scene+1).mp3 ' ,
' md5 ' : ' 1d0aabe03edca83ca58d9ed3b493a3c3 ' ,
' info_dict ' : {
' id ' : ' lp_the-music-of-russia_various-artists-a-askaryan-alexander-melik/disc1/02.02. Song And Chorus In The Polovetsian Camp From " Prince Igor " (Act 2, Scene 1).mp3 ' ,
' title ' : ' Song And Chorus In The Polovetsian Camp From " Prince Igor " (Act 2, Scene 1) ' ,
' ext ' : ' mp3 ' ,
' timestamp ' : 1569662587 ,
' uploader ' : ' associate-joygen-odiongan@archive.org ' ,
' description ' : ' md5:012b2d668ae753be36896f343d12a236 ' ,
' upload_date ' : ' 20190928 ' ,
} ,
2014-12-28 20:08:46 +00:00
} ]
2014-12-28 19:04:21 +00:00
2021-01-10 21:37:45 +00:00
@staticmethod
def _playlist_data ( webpage ) :
element = re . findall ( r ''' (?xs)
< input
( ? : \s + [ a - zA - Z0 - 9 : . _ - ] + ( ? := [ a - zA - Z0 - 9 : . _ - ] * | = " [^ " ] * " |= ' [^ ' ]* ' |))*?
\s + class = [ ' " ]?js-play8-playlist[ ' " ]?
( ? : \s + [ a - zA - Z0 - 9 : . _ - ] + ( ? := [ a - zA - Z0 - 9 : . _ - ] * | = " [^ " ] * " |= ' [^ ' ]* ' |))*?
\s * / >
''' , webpage)[0]
return json . loads ( extract_attributes ( element ) [ ' value ' ] )
2013-07-08 00:04:11 +00:00
def _real_extract ( self , url ) :
2021-01-10 21:37:45 +00:00
video_id = compat_urllib_parse_unquote_plus ( self . _match_id ( url ) )
identifier , entry_id = ( video_id . split ( ' / ' , 1 ) + [ None ] ) [ : 2 ]
# Archive.org metadata API doesn't clearly demarcate playlist entries
# or subtitle tracks, so we get them from the embeddable player.
embed_page = self . _download_webpage (
' https://archive.org/embed/ ' + identifier , identifier )
playlist = self . _playlist_data ( embed_page )
entries = { }
for p in playlist :
# If the user specified a playlist entry in the URL, ignore the
# rest of the playlist.
if entry_id and p [ ' orig ' ] != entry_id :
continue
entries [ p [ ' orig ' ] ] = {
' formats ' : [ ] ,
' thumbnails ' : [ ] ,
' artist ' : p . get ( ' artist ' ) ,
' track ' : p . get ( ' title ' ) ,
' subtitles ' : { } }
for track in p . get ( ' tracks ' , [ ] ) :
if track [ ' kind ' ] != ' subtitles ' :
continue
entries [ p [ ' orig ' ] ] [ track [ ' label ' ] ] = {
' url ' : ' https://archive.org/ ' + track [ ' file ' ] . lstrip ( ' / ' ) }
2013-07-08 00:04:11 +00:00
2016-08-05 15:40:21 +00:00
metadata = self . _download_json (
2021-01-10 21:37:45 +00:00
' http://archive.org/metadata/ ' + identifier , identifier )
m = metadata [ ' metadata ' ]
identifier = m [ ' identifier ' ]
info = {
' id ' : identifier ,
' title ' : m [ ' title ' ] ,
' description ' : clean_html ( m . get ( ' description ' ) ) ,
' uploader ' : dict_get ( m , [ ' uploader ' , ' adder ' ] ) ,
' creator ' : m . get ( ' creator ' ) ,
' license ' : m . get ( ' licenseurl ' ) ,
' release_date ' : unified_strdate ( m . get ( ' date ' ) ) ,
' timestamp ' : unified_timestamp ( dict_get ( m , [ ' publicdate ' , ' addeddate ' ] ) ) ,
' webpage_url ' : ' https://archive.org/details/ ' + identifier ,
' location ' : m . get ( ' venue ' ) ,
' release_year ' : int_or_none ( m . get ( ' year ' ) ) }
for f in metadata [ ' files ' ] :
if f [ ' name ' ] in entries :
entries [ f [ ' name ' ] ] = merge_dicts ( entries [ f [ ' name ' ] ] , {
' id ' : identifier + ' / ' + f [ ' name ' ] ,
' title ' : f . get ( ' title ' ) or f [ ' name ' ] ,
' display_id ' : f [ ' name ' ] ,
' description ' : clean_html ( f . get ( ' description ' ) ) ,
' creator ' : f . get ( ' creator ' ) ,
' duration ' : parse_duration ( f . get ( ' length ' ) ) ,
' track_number ' : int_or_none ( f . get ( ' track ' ) ) ,
' album ' : f . get ( ' album ' ) ,
' discnumber ' : int_or_none ( f . get ( ' disc ' ) ) ,
' release_year ' : int_or_none ( f . get ( ' year ' ) ) } )
entry = entries [ f [ ' name ' ] ]
elif f . get ( ' original ' ) in entries :
entry = entries [ f [ ' original ' ] ]
else :
continue
if f . get ( ' format ' ) == ' Thumbnail ' :
entry [ ' thumbnails ' ] . append ( {
' id ' : f [ ' name ' ] ,
' url ' : ' https://archive.org/download/ ' + identifier + ' / ' + f [ ' name ' ] ,
' width ' : int_or_none ( f . get ( ' width ' ) ) ,
' height ' : int_or_none ( f . get ( ' width ' ) ) ,
' filesize ' : int_or_none ( f . get ( ' size ' ) ) } )
extension = ( f [ ' name ' ] . rsplit ( ' . ' , 1 ) + [ None ] ) [ 1 ]
if extension in KNOWN_EXTENSIONS :
entry [ ' formats ' ] . append ( {
' url ' : ' https://archive.org/download/ ' + identifier + ' / ' + f [ ' name ' ] ,
' format ' : f . get ( ' format ' ) ,
' width ' : int_or_none ( f . get ( ' width ' ) ) ,
' height ' : int_or_none ( f . get ( ' height ' ) ) ,
' filesize ' : int_or_none ( f . get ( ' size ' ) ) ,
' protocol ' : ' https ' } )
# Sort available formats by filesize
for entry in entries . values ( ) :
entry [ ' formats ' ] = list ( sorted ( entry [ ' formats ' ] , key = lambda x : x . get ( ' filesize ' , - 1 ) ) )
if len ( entries ) == 1 :
# If there's only one item, use it as the main info dict
only_video = entries [ list ( entries . keys ( ) ) [ 0 ] ]
if entry_id :
info = merge_dicts ( only_video , info )
else :
info = merge_dicts ( info , only_video )
else :
# Otherwise, we have a playlist.
info [ ' _type ' ] = ' playlist '
info [ ' entries ' ] = list ( entries . values ( ) )
if metadata . get ( ' reviews ' ) :
info [ ' comments ' ] = [ ]
for review in metadata [ ' reviews ' ] :
info [ ' comments ' ] . append ( {
' id ' : review . get ( ' review_id ' ) ,
' author ' : review . get ( ' reviewer ' ) ,
' text ' : str_or_none ( review . get ( ' reviewtitle ' ) , ' ' ) + ' \n \n ' + review . get ( ' reviewbody ' ) ,
' timestamp ' : unified_timestamp ( review . get ( ' createdate ' ) ) ,
' parent ' : ' root ' } )
2016-08-05 16:16:19 +00:00
return info