epicyon/conversation.py

395 lines
15 KiB
Python

__filename__ = "conversation.py"
__author__ = "Bob Mottram"
__license__ = "AGPL3+"
__version__ = "1.5.0"
__maintainer__ = "Bob Mottram"
__email__ = "bob@libreserver.org"
__status__ = "Production"
__module_group__ = "Timeline"
import os
from utils import has_object_dict
from utils import acct_dir
from utils import remove_id_ending
from utils import text_in_file
from utils import locate_post
from utils import load_json
from utils import harmless_markup
from utils import get_attributed_to
from utils import get_reply_to
from utils import resembles_url
from keys import get_instance_actor_key
from session import get_json
from session import get_json_valid
def _get_conversation_filename(base_dir: str, nickname: str, domain: str,
post_json_object: {}) -> str:
"""Returns the conversation filename
"""
if not has_object_dict(post_json_object):
return None
if not post_json_object['object'].get('conversation') and \
not post_json_object['object'].get('context'):
return None
if not post_json_object['object'].get('id'):
return None
conversation_dir = acct_dir(base_dir, nickname, domain) + '/conversation'
if not os.path.isdir(conversation_dir):
os.mkdir(conversation_dir)
if post_json_object['object'].get('conversation'):
conversation_id = post_json_object['object']['conversation']
else:
conversation_id = post_json_object['object']['context']
conversation_id = conversation_id.replace('/', '#')
return conversation_dir + '/' + conversation_id
def update_conversation(base_dir: str, nickname: str, domain: str,
post_json_object: {}) -> bool:
"""Adds a post to a conversation index in the /conversation subdirectory
"""
conversation_filename = \
_get_conversation_filename(base_dir, nickname, domain,
post_json_object)
if not conversation_filename:
return False
post_id = remove_id_ending(post_json_object['object']['id'])
if not os.path.isfile(conversation_filename):
try:
with open(conversation_filename, 'w+',
encoding='utf-8') as fp_conv:
fp_conv.write(post_id + '\n')
return True
except OSError:
print('EX: update_conversation ' +
'unable to write to ' + conversation_filename)
elif not text_in_file(post_id + '\n', conversation_filename):
try:
with open(conversation_filename, 'a+',
encoding='utf-8') as fp_conv:
fp_conv.write(post_id + '\n')
return True
except OSError:
print('EX: update_conversation 2 ' +
'unable to write to ' + conversation_filename)
return False
def mute_conversation(base_dir: str, nickname: str, domain: str,
conversation_id: str) -> None:
"""Mutes the given conversation
"""
conversation_dir = acct_dir(base_dir, nickname, domain) + '/conversation'
conversation_filename = \
conversation_dir + '/' + conversation_id.replace('/', '#')
if not os.path.isfile(conversation_filename):
return
if os.path.isfile(conversation_filename + '.muted'):
return
try:
with open(conversation_filename + '.muted', 'w+',
encoding='utf-8') as fp_conv:
fp_conv.write('\n')
except OSError:
print('EX: unable to write mute ' + conversation_filename)
def unmute_conversation(base_dir: str, nickname: str, domain: str,
conversation_id: str) -> None:
"""Unmutes the given conversation
"""
conversation_dir = acct_dir(base_dir, nickname, domain) + '/conversation'
conversation_filename = \
conversation_dir + '/' + conversation_id.replace('/', '#')
if not os.path.isfile(conversation_filename):
return
if not os.path.isfile(conversation_filename + '.muted'):
return
try:
os.remove(conversation_filename + '.muted')
except OSError:
print('EX: unmute_conversation unable to delete ' +
conversation_filename + '.muted')
def _get_replies_to_post(post_json_object: {},
signing_priv_key_pem: str,
session, as_header, debug: bool,
http_prefix: str, domain: str,
depth: int, ids: []) -> []:
"""Returns a list of reply posts to the given post as json
"""
result = []
post_obj = post_json_object
if has_object_dict(post_json_object):
post_obj = post_json_object['object']
if not post_obj.get('replies'):
return result
# get the replies collection url
replies_collection_id = None
if isinstance(post_obj['replies'], dict):
if post_obj['replies'].get('id'):
replies_collection_id = post_obj['replies']['id']
elif isinstance(post_obj['replies'], str):
replies_collection_id = post_obj['replies']
if replies_collection_id:
if debug:
print('DEBUG: get_replies_to_post replies_collection_id ' +
str(replies_collection_id))
replies_collection = \
get_json(signing_priv_key_pem, session, replies_collection_id,
as_header, None, debug, __version__,
http_prefix, domain)
if not get_json_valid(replies_collection):
return result
if debug:
print('DEBUG: get_replies_to_post replies_collection ' +
str(replies_collection))
# get the list of replies
if not replies_collection.get('first'):
return result
if not isinstance(replies_collection['first'], dict):
return result
if not replies_collection['first'].get('items'):
if not replies_collection['first'].get('next'):
return result
items_list = []
if replies_collection['first'].get('items'):
items_list = replies_collection['first']['items']
if not items_list:
# if there are no items try the next one
next_page_id = replies_collection['first']['next']
if not isinstance(next_page_id, str):
return result
replies_collection = \
get_json(signing_priv_key_pem, session, next_page_id,
as_header, None, debug, __version__,
http_prefix, domain)
if debug:
print('DEBUG: get_replies_to_post next replies_collection ' +
str(replies_collection))
if not get_json_valid(replies_collection):
return result
if not replies_collection.get('items'):
return result
if not isinstance(replies_collection['items'], list):
return result
items_list = replies_collection['items']
if debug:
print('DEBUG: get_replies_to_post items_list ' +
str(items_list))
if not isinstance(items_list, list):
return result
# check each item in the list
for item in items_list:
# download the item if needed
if isinstance(item, str):
if resembles_url(item):
if debug:
print('Downloading conversation item ' + item)
item_dict = \
get_json(signing_priv_key_pem, session, item,
as_header, None, debug, __version__,
http_prefix, domain)
if not get_json_valid(item_dict):
continue
item = item_dict
if not isinstance(item, dict):
continue
if not has_object_dict(item):
if not item.get('attributedTo'):
continue
attrib_str = get_attributed_to(item['attributedTo'])
if not attrib_str:
continue
if not item.get('published'):
continue
if not item.get('id'):
continue
if not isinstance(item['id'], str):
continue
if not item.get('to'):
continue
if not isinstance(item['to'], list):
continue
if 'cc' not in item:
continue
if not isinstance(item['cc'], list):
continue
wrapped_post = {
"@context": "https://www.w3.org/ns/activitystreams",
'id': item['id'] + '/activity',
'type': 'Create',
'actor': attrib_str,
'published': item['published'],
'to': item['to'],
'cc': item['cc'],
'object': item
}
item = wrapped_post
if not item['object'].get('published'):
continue
# render harmless any dangerous markup
harmless_markup(item)
# keep a list of ids encountered, to avoid circularity
reply_post_id = None
if item.get('id'):
if isinstance(item['id'], str):
reply_post_id = item['id']
if reply_post_id in ids:
continue
ids.append(reply_post_id)
# add it to the list
result.append(item)
if depth < 10 and reply_post_id:
result += \
_get_replies_to_post(item,
signing_priv_key_pem,
session, as_header,
debug,
http_prefix, domain,
depth + 1, ids)
return result
def download_conversation_posts(authorized: bool, session,
http_prefix: str, base_dir: str,
nickname: str, domain: str,
post_id: str, debug: bool) -> []:
"""Downloads all posts for a conversation and returns a list of the
json objects
"""
if '://' not in post_id:
return []
profile_str = 'https://www.w3.org/ns/activitystreams'
as_header = {
'Accept': 'application/ld+json; profile="' + profile_str + '"'
}
conversation_view = []
signing_priv_key_pem = get_instance_actor_key(base_dir, domain)
post_id = remove_id_ending(post_id)
post_filename = \
locate_post(base_dir, nickname, domain, post_id)
post_json_object = None
if post_filename:
post_json_object = load_json(post_filename)
else:
if authorized:
post_json_object = \
get_json(signing_priv_key_pem, session, post_id,
as_header, None, debug, __version__,
http_prefix, domain)
if debug:
if not get_json_valid(post_json_object):
print(post_id + ' returned no json')
# get any replies
replies_to_post = []
if get_json_valid(post_json_object):
replies_to_post = \
_get_replies_to_post(post_json_object,
signing_priv_key_pem,
session, as_header, debug,
http_prefix, domain, 0, [])
ids = []
while get_json_valid(post_json_object):
if not isinstance(post_json_object, dict):
break
if not has_object_dict(post_json_object):
if not post_json_object.get('id'):
break
if not isinstance(post_json_object['id'], str):
break
if not post_json_object.get('attributedTo'):
if debug:
print(str(post_json_object))
print(post_json_object['id'] + ' has no attributedTo')
break
attrib_str = get_attributed_to(post_json_object['attributedTo'])
if not attrib_str:
break
if not post_json_object.get('published'):
if debug:
print(str(post_json_object))
print(post_json_object['id'] + ' has no published date')
break
if not post_json_object.get('to'):
if debug:
print(str(post_json_object))
print(post_json_object['id'] + ' has no "to" list')
break
if not isinstance(post_json_object['to'], list):
break
if 'cc' not in post_json_object:
if debug:
print(str(post_json_object))
print(post_json_object['id'] + ' has no "cc" list')
break
if not isinstance(post_json_object['cc'], list):
break
wrapped_post = {
"@context": "https://www.w3.org/ns/activitystreams",
'id': post_json_object['id'] + '/activity',
'type': 'Create',
'actor': attrib_str,
'published': post_json_object['published'],
'to': post_json_object['to'],
'cc': post_json_object['cc'],
'object': post_json_object
}
post_json_object = wrapped_post
if not post_json_object['object'].get('published'):
break
# avoid any circularity in previous conversation posts
if post_json_object.get('id'):
if isinstance(post_json_object['id'], str):
if post_json_object['id'] in ids:
break
ids.append(post_json_object['id'])
# render harmless any dangerous markup
harmless_markup(post_json_object)
conversation_view = [post_json_object] + conversation_view
if not authorized:
# only show a single post to non-authorized viewers
break
post_id = get_reply_to(post_json_object['object'])
if not post_id:
if debug:
print(post_id + ' is not a reply')
break
post_id = remove_id_ending(post_id)
post_filename = \
locate_post(base_dir, nickname, domain, post_id)
post_json_object = None
if post_filename:
post_json_object = load_json(post_filename)
else:
if authorized:
post_json_object = \
get_json(signing_priv_key_pem, session, post_id,
as_header, None, debug, __version__,
http_prefix, domain)
if debug:
if not get_json_valid(post_json_object):
print(post_id + ' returned no json')
return conversation_view + replies_to_post