epicyon/webapp_media.py

227 lines
9.0 KiB
Python
Raw Normal View History

2020-11-09 19:41:01 +00:00
__filename__ = "webapp_media.py"
__author__ = "Bob Mottram"
__license__ = "AGPL3+"
__version__ = "1.1.0"
__maintainer__ = "Bob Mottram"
__email__ = "bob@freedombone.net"
__status__ = "Production"
def _addEmbeddedVideoFromSites(translate: {}, content: str,
width=400, height=300) -> str:
2020-11-09 19:41:01 +00:00
"""Adds embedded videos
"""
if '>vimeo.com/' in content:
url = content.split('>vimeo.com/')[1]
if '<' in url:
url = url.split('<')[0]
content = \
content + "<center>\n<iframe loading=\"lazy\" " + \
"src=\"https://player.vimeo.com/video/" + \
url + "\" width=\"" + str(width) + \
"\" height=\"" + str(height) + \
"\" frameborder=\"0\" allow=\"autoplay; " + \
"fullscreen\" allowfullscreen></iframe>\n</center>\n"
return content
videoSite = 'https://www.youtube.com'
if '"' + videoSite in content:
url = content.split('"' + videoSite)[1]
if '"' in url:
url = url.split('"')[0].replace('/watch?v=', '/embed/')
if '&' in url:
url = url.split('&')[0]
content = \
content + "<center>\n<iframe loading=\"lazy\" src=\"" + \
videoSite + url + "\" width=\"" + str(width) + \
"\" height=\"" + str(height) + \
"\" frameborder=\"0\" allow=\"autoplay; fullscreen\" " + \
"allowfullscreen></iframe>\n</center>\n"
return content
invidiousSites = ('https://invidio.us',
'https://invidious.snopyta.org',
'http://c7hqkpkpemu6e7emz5b4vy' +
'z7idjgdvgaaa3dyimmeojqbgpea3xqjoid.onion',
'http://axqzx4s6s54s32yentfqojs3x5i7faxza6xo3ehd4' +
'bzzsg2ii4fv2iid.onion')
for videoSite in invidiousSites:
if '"' + videoSite in content:
url = content.split('"' + videoSite)[1]
if '"' in url:
url = url.split('"')[0].replace('/watch?v=', '/embed/')
if '&' in url:
url = url.split('&')[0]
content = \
content + "<center>\n<iframe loading=\"lazy\" src=\"" + \
videoSite + url + "\" width=\"" + \
str(width) + "\" height=\"" + str(height) + \
"\" frameborder=\"0\" allow=\"autoplay; fullscreen\" " + \
"allowfullscreen></iframe>\n</center>\n"
return content
videoSite = 'https://media.ccc.de'
if '"' + videoSite in content:
url = content.split('"' + videoSite)[1]
if '"' in url:
url = url.split('"')[0]
if not url.endswith('/oembed'):
url = url + '/oembed'
content = \
content + "<center>\n<iframe loading=\"lazy\" src=\"" + \
videoSite + url + "\" width=\"" + \
str(width) + "\" height=\"" + str(height) + \
"\" frameborder=\"0\" allow=\"fullscreen\" " + \
"allowfullscreen></iframe>\n</center>\n"
return content
if '"https://' in content:
# A selection of the current larger peertube sites, mostly
# French and German language
# These have been chosen based on reported numbers of users
# and the content of each has not been reviewed, so mileage could vary
2020-12-23 22:47:07 +00:00
# Also see https://peertube_isolation.frama.io/list/ for
# adversarial instances
2020-12-23 22:45:12 +00:00
peerTubeSites = ('peertube.mastodon.host', 'share.tube',
2020-11-09 19:41:01 +00:00
'tube.tr4sk.me', 'videos.elbinario.net',
'hkvideo.live',
'peertube.snargol.com', 'tube.22decembre.eu',
'tube.fabrigli.fr', 'libretube.net', 'libre.video',
'peertube.linuxrocks.online', 'spacepub.space',
'video.ploud.jp', 'video.omniatv.com',
'peertube.servebeer.com',
'tube.tchncs.de', 'tubee.fr', 'video.alternanet.fr',
'devtube.dev-wiki.de', 'video.samedi.pm',
'video.irem.univ-paris-diderot.fr',
'peertube.openstreetmap.fr', 'video.antopie.org',
'scitech.video', 'tube.4aem.com', 'video.ploud.fr',
'peervideo.net', 'video.valme.io',
'videos.pair2jeux.tube',
'vault.mle.party', 'hostyour.tv',
'diode.zone', 'visionon.tv',
'artitube.artifaille.fr', 'peertube.fr',
2020-11-27 18:57:17 +00:00
'peertube.live', 'kolektiva.media',
2020-11-09 19:41:01 +00:00
'tube.ac-lyon.fr', 'www.yiny.org', 'betamax.video',
'tube.piweb.be', 'pe.ertu.be', 'peertube.social',
'videos.lescommuns.org', 'peertube.nogafa.org',
'skeptikon.fr', 'video.tedomum.net',
2020-11-24 18:32:06 +00:00
'tube.p2p.legal', 'tilvids.com',
2020-11-09 19:41:01 +00:00
'sikke.fi', 'exode.me', 'peertube.video')
for site in peerTubeSites:
if '"https://' + site in content:
url = content.split('"https://' + site)[1]
if '"' in url:
url = url.split('"')[0].replace('/watch/', '/embed/')
content = \
content + "<center>\n<iframe loading=\"lazy\" " + \
"sandbox=\"allow-same-origin " + \
"allow-scripts\" src=\"https://" + \
site + url + "\" width=\"" + str(width) + \
"\" height=\"" + str(height) + \
"\" frameborder=\"0\" allow=\"autoplay; " + \
"fullscreen\" allowfullscreen></iframe>\n</center>\n"
return content
return content
def _addEmbeddedAudio(translate: {}, content: str) -> str:
2020-11-09 19:41:01 +00:00
"""Adds embedded audio for mp3/ogg
"""
if not ('.mp3' in content or '.ogg' in content):
return content
if '<audio ' in content:
return content
extension = '.mp3'
if '.ogg' in content:
extension = '.ogg'
words = content.strip('\n').split(' ')
for w in words:
if extension not in w:
continue
w = w.replace('href="', '').replace('">', '')
if w.endswith('.'):
w = w[:-1]
if w.endswith('"'):
w = w[:-1]
if w.endswith(';'):
w = w[:-1]
if w.endswith(':'):
w = w[:-1]
if not w.endswith(extension):
continue
if not (w.startswith('http') or w.startswith('dat:') or
w.startswith('hyper:') or w.startswith('i2p:') or
w.startswith('gnunet:') or
'/' in w):
continue
url = w
content += '<center>\n<audio controls>\n'
content += \
'<source src="' + url + '" type="audio/' + \
extension.replace('.', '') + '">'
content += \
translate['Your browser does not support the audio element.']
content += '</audio>\n</center>\n'
return content
def _addEmbeddedVideo(translate: {}, content: str,
width=400, height=300) -> str:
2020-11-09 19:41:01 +00:00
"""Adds embedded video for mp4/webm/ogv
"""
if not ('.mp4' in content or '.webm' in content or '.ogv' in content):
return content
if '<video ' in content:
return content
extension = '.mp4'
if '.webm' in content:
extension = '.webm'
elif '.ogv' in content:
extension = '.ogv'
words = content.strip('\n').split(' ')
for w in words:
if extension not in w:
continue
w = w.replace('href="', '').replace('">', '')
if w.endswith('.'):
w = w[:-1]
if w.endswith('"'):
w = w[:-1]
if w.endswith(';'):
w = w[:-1]
if w.endswith(':'):
w = w[:-1]
if not w.endswith(extension):
continue
if not (w.startswith('http') or w.startswith('dat:') or
w.startswith('hyper:') or w.startswith('i2p:') or
w.startswith('gnunet:') or
'/' in w):
continue
url = w
content += \
'<center>\n<video width="' + str(width) + '" height="' + \
str(height) + '" controls>\n'
content += \
'<source src="' + url + '" type="video/' + \
extension.replace('.', '') + '">\n'
content += \
translate['Your browser does not support the video element.']
content += '</video>\n</center>\n'
return content
def addEmbeddedElements(translate: {}, content: str) -> str:
"""Adds embedded elements for various media types
"""
content = _addEmbeddedVideoFromSites(translate, content)
content = _addEmbeddedAudio(translate, content)
return _addEmbeddedVideo(translate, content)