epicyon/posts.py

502 lines
19 KiB
Python

__filename__ = "posts.py"
__author__ = "Bob Mottram"
__license__ = "AGPL3+"
__version__ = "0.0.1"
__maintainer__ = "Bob Mottram"
__email__ = "bob@freedombone.net"
__status__ = "Production"
import requests
import json
import commentjson
import html
import datetime
import os
import shutil
import threading
import sys
import trace
from threads import threadWithTrace
from cache import storePersonInCache
from cache import getPersonFromCache
from pprint import pprint
from random import randint
from session import getJson
from session import postJson
from webfinger import webfingerHandle
try:
from BeautifulSoup import BeautifulSoup
except ImportError:
from bs4 import BeautifulSoup
def getPersonKey(username: str,domain: str,baseDir: str,keyType='public'):
"""Returns the public or private key of a person
"""
handle=username+'@'+domain
keyFilename=baseDir+'/keys/'+keyType+'/'+handle.lower()+'.key'
if not os.path.isfile(keyFilename):
return ''
keyPem=''
with open(keyFilename, "r") as pemFile:
keyPem=pemFile.read()
if len(keyPem)<20:
return ''
return keyPem
def permitted(url: str,federationList) -> bool:
"""Is a url from one of the permitted domains?
"""
for domain in federationList:
if domain in url:
return True
return False
def cleanHtml(rawHtml: str) -> str:
text = BeautifulSoup(rawHtml, 'html.parser').get_text()
return html.unescape(text)
def getUserUrl(wfRequest) -> str:
if wfRequest.get('links'):
for link in wfRequest['links']:
if link.get('type') and link.get('href'):
if link['type'] == 'application/activity+json':
return link['href']
return None
def parseUserFeed(session,feedUrl,asHeader) -> None:
feedJson = getJson(session,feedUrl,asHeader,None)
pprint(feedJson)
if 'orderedItems' in feedJson:
for item in feedJson['orderedItems']:
yield item
nextUrl = None
if 'first' in feedJson:
nextUrl = feedJson['first']
elif 'next' in feedJson:
nextUrl = feedJson['next']
if nextUrl:
for item in parseUserFeed(session,nextUrl,asHeader):
yield item
def getPersonBox(session,wfRequest,boxName='inbox'):
asHeader = {'Accept': 'application/ld+json; profile="https://www.w3.org/ns/activitystreams"'}
personUrl = getUserUrl(wfRequest)
if not personUrl:
return None
personJson = getPersonFromCache(personUrl)
if not personJson:
personJson = getJson(session,personUrl,asHeader,None)
if not personJson.get(boxName):
return personPosts
personId=None
if personJson.get('id'):
personId=personJson['id']
pubKey=None
if personJson.get('publicKey'):
if personJson['publicKey'].get('publicKeyPem'):
pubKey=personJson['publicKey']['publicKeyPem']
storePersonInCache(personUrl,personJson)
return personJson[boxName],pubKey,personId
def getUserPosts(session,wfRequest,maxPosts,maxMentions,maxEmoji,maxAttachments,federationList) -> {}:
userPosts={}
feedUrl,pubKey,personId = getPersonBox(session,wfRequest,'outbox')
if not feedUrl:
return userPosts
i = 0
for item in parseUserFeed(session,feedUrl,asHeader):
if not item.get('type'):
continue
if item['type'] != 'Create':
continue
if not item.get('object'):
continue
published = item['object']['published']
if not userPosts.get(published):
content = item['object']['content']
mentions=[]
emoji={}
if item['object'].get('tag'):
for tagItem in item['object']['tag']:
tagType=tagItem['type'].lower()
if tagType=='emoji':
if tagItem.get('name') and tagItem.get('icon'):
if tagItem['icon'].get('url'):
# No emoji from non-permitted domains
if permitted(tagItem['icon']['url'],federationList):
emojiName=tagItem['name']
emojiIcon=tagItem['icon']['url']
emoji[emojiName]=emojiIcon
if tagType=='mention':
if tagItem.get('name'):
if tagItem['name'] not in mentions:
mentions.append(tagItem['name'])
if len(mentions)>maxMentions:
continue
if len(emoji)>maxEmoji:
continue
summary = ''
if item['object'].get('summary'):
if item['object']['summary']:
summary = item['object']['summary']
inReplyTo = ''
if item['object'].get('inReplyTo'):
if item['object']['inReplyTo']:
# No replies to non-permitted domains
if not permitted(item['object']['inReplyTo'],federationList):
continue
inReplyTo = item['object']['inReplyTo']
conversation = ''
if item['object'].get('conversation'):
if item['object']['conversation']:
# no conversations originated in non-permitted domains
if permitted(item['object']['conversation'],federationList):
conversation = item['object']['conversation']
attachment = []
if item['object'].get('attachment'):
if item['object']['attachment']:
for attach in item['object']['attachment']:
if attach.get('name') and attach.get('url'):
# no attachments from non-permitted domains
if permitted(attach['url'],federationList):
attachment.append([attach['name'],attach['url']])
sensitive = False
if item['object'].get('sensitive'):
sensitive = item['object']['sensitive']
userPosts[published] = {
"sensitive": sensitive,
"inreplyto": inReplyTo,
"summary": summary,
"html": content,
"plaintext": cleanHtml(content),
"attachment": attachment,
"mentions": mentions,
"emoji": emoji,
"conversation": conversation
}
#print(str(item)+'\n')
i += 1
if i == maxPosts:
break
return userPosts
def createOutboxDir(username: str,domain: str,baseDir: str) -> str:
"""Create an outbox for a person and returns the feed filename and directory
"""
handle=username.lower()+'@'+domain.lower()
if not os.path.isdir(baseDir+'/accounts/'+handle):
os.mkdir(baseDir+'/accounts/'+handle)
outboxDir=baseDir+'/accounts/'+handle+'/outbox'
if not os.path.isdir(outboxDir):
os.mkdir(outboxDir)
return outboxDir
def createOutboxArchive(username: str,domain: str,baseDir: str) -> str:
"""Creates an archive directory for outbox posts
"""
handle=username.lower()+'@'+domain.lower()
if not os.path.isdir(baseDir+'/accounts/'+handle):
os.mkdir(baseDir+'/accounts/'+handle)
outboxArchiveDir=baseDir+'/accounts/'+handle+'/outboxarchive'
if not os.path.isdir(outboxArchiveDir):
os.mkdir(outboxArchiveDir)
return outboxArchiveDir
def deleteAllPosts(username: str, domain: str,baseDir: str) -> None:
"""Deletes all posts for a person
"""
outboxDir = createOutboxDir(username,domain,baseDir)
for deleteFilename in os.listdir(outboxDir):
filePath = os.path.join(outboxDir, deleteFilename)
try:
if os.path.isfile(filePath):
os.unlink(filePath)
elif os.path.isdir(filePath): shutil.rmtree(filePath)
except Exception as e:
print(e)
def getStatusNumber() -> (str,str):
"""Returns the status number and published date
"""
currTime=datetime.datetime.utcnow()
daysSinceEpoch=(currTime - datetime.datetime(1970,1,1)).days
# status is the number of seconds since epoch
statusNumber=str((daysSinceEpoch*24*60*60) + (currTime.hour*60*60) + (currTime.minute*60) + currTime.second)
published=currTime.strftime("%Y-%m-%dT%H:%M:%SZ")
conversationDate=currTime.strftime("%Y-%m-%d")
return statusNumber,published
def createPostBase(baseDir: str,username: str, domain: str, toUrl: str, ccUrl: str, https: bool, content: str, followersOnly: bool, saveToFile: bool, inReplyTo=None, inReplyToAtomUri=None, subject=None) -> {}:
"""Creates a public post
"""
prefix='https'
if not https:
prefix='http'
statusNumber,published = getStatusNumber()
conversationDate=published.split('T')[0]
conversationId=statusNumber
postTo='https://www.w3.org/ns/activitystreams#Public'
postCC=prefix+'://'+domain+'/users/'+username+'/followers'
if followersOnly:
postTo=postCC
postCC=''
newPostId=prefix+'://'+domain+'/users/'+username+'/statuses/'+statusNumber
sensitive=False
summary=None
if subject:
summary=subject
sensitive=True
newPost = {
'id': newPostId+'/activity',
'type': 'Create',
'actor': prefix+'://'+domain+'/users/'+username,
'published': published,
'to': [toUrl],
'cc': [ccUrl],
'object': {'id': newPostId,
'type': 'Note',
'summary': summary,
'inReplyTo': inReplyTo,
'published': published,
'url': prefix+'://'+domain+'/@'+username+'/'+statusNumber,
'attributedTo': prefix+'://'+domain+'/users/'+username,
'to': [toUrl],
'cc': [ccUrl],
'sensitive': sensitive,
'atomUri': prefix+'://'+domain+'/users/'+username+'/statuses/'+statusNumber,
'inReplyToAtomUri': inReplyToAtomUri,
'conversation': 'tag:'+domain+','+conversationDate+':objectId='+conversationId+':objectType=Conversation',
'content': content,
'contentMap': {
'en': content
},
'attachment': [],
'tag': [],
'replies': {}
# 'id': 'https://'+domain+'/users/'+username+'/statuses/'+statusNumber+'/replies',
# 'type': 'Collection',
# 'first': {
# 'type': 'CollectionPage',
# 'partOf': 'https://'+domain+'/users/'+username+'/statuses/'+statusNumber+'/replies',
# 'items': []
# }
#}
}
}
if saveToFile:
outboxDir = createOutboxDir(username,domain,baseDir)
filename=outboxDir+'/'+newPostId.replace('/','#')+'.json'
with open(filename, 'w') as fp:
commentjson.dump(newPost, fp, indent=4, sort_keys=False)
return newPost
def createPublicPost(username: str, domain: str, https: bool, content: str, followersOnly: bool, saveToFile: bool, inReplyTo=None, inReplyToAtomUri=None, subject=None) -> {}:
"""Public post to the outbox
"""
prefix='https'
if not https:
prefix='http'
return createPostBase(username, domain, 'https://www.w3.org/ns/activitystreams#Public', prefix+'://'+domain+'/users/'+username+'/followers', https, content, followersOnly, saveToFile, inReplyTo, inReplyToAtomUri, subject)
def threadSendPost(session,postJsonObject,federationList,inboxUrl: str,baseDir: str,signatureHeader,postLog) -> None:
"""Sends a post with exponential backoff
"""
tries=0
backoffTime=60
for attempt in range(20):
postResult = postJson(session,postJsonObject,federationList,inboxUrl,signatureHeader)
if postResult:
postLog.append(postJsonObject['published']+' '+postResult+'\n')
# keep the length of the log finite
# Don't accumulate massive files on systems with limited resources
while len(postLog)>64:
postlog.pop(0)
# save the log file
filename=baseDir+'/post.log'
with open(filename, "w") as logFile:
for line in postLog:
print(line, file=logFile)
# our work here is done
break
time.sleep(backoffTime)
backoffTime *= 2
def sendPost(session,baseDir,username: str, domain: str, port: int, toUsername: str, toDomain: str, toPort: int, cc: str, https: bool, content: str, followersOnly: bool, saveToFile: bool, federationList, sendThreads, postLog, inReplyTo=None, inReplyToAtomUri=None, subject=None) -> int:
"""Post to another inbox
"""
prefix='https'
if not https:
prefix='http'
if toPort!=80 and toPort!=443:
toDomain=toDomain+':'+str(toPort)
handle=prefix+'://'+toDomain+'/@'+toUsername
# lookup the inbox for the To handle
wfRequest = webfingerHandle(session,handle,https)
if not wfRequest:
return 1
# get the actor inbox for the To handle
inboxUrl,pubKey,toPersonId = getPersonBox(session,wfRequest,'inbox')
if not inboxUrl:
return 2
if not pubKey:
return 3
if not toPersonId:
return 4
postJsonObject=createPostBase(username, domain, toPersonId, cc, https, content, followersOnly, saveToFile, inReplyTo, inReplyToAtomUri, subject)
# get the senders private key
privateKeyPem=getPersonKey(username,domain,baseDir,'private')
if len(privateKeyPem)==0:
return 5
# construct the http header
signatureHeader = signPostHeaders(privateKeyPem, username, domain, '/inbox', https, postJsonObject)
signatureHeader['Content-type'] = 'application/json'
# Keep the number of threads being used small
while len(sendThreads)>10:
sendThreads[0].kill()
sendThreads.pop(0)
thr = threadWithTrace(target=threadSendPost,args=(session,postJsonObject.copy(),federationList,inboxUrl,baseDir,signatureHeader.copy(),postLog),daemon=True)
sendThreads.append(thr)
thr.start()
return 0
def createOutbox(baseDir: str,username: str,domain: str,port: int,https: bool,itemsPerPage: int,headerOnly: bool,pageNumber=None) -> {}:
"""Constructs the outbox feed
"""
prefix='https'
if not https:
prefix='http'
outboxDir = createOutboxDir(username,domain,baseDir)
if port!=80 and port!=443:
domain = domain+':'+str(port)
pageStr='?page=true'
if pageNumber:
try:
pageStr='?page='+str(pageNumber)
except:
pass
outboxHeader = {'@context': 'https://www.w3.org/ns/activitystreams',
'first': prefix+'://'+domain+'/users/'+username+'/outbox?page=true',
'id': prefix+'://'+domain+'/users/'+username+'/outbox',
'last': prefix+'://'+domain+'/users/'+username+'/outbox?page=true',
'totalItems': 0,
'type': 'OrderedCollection'}
outboxItems = {'@context': 'https://www.w3.org/ns/activitystreams',
'id': prefix+'://'+domain+'/users/'+username+'/outbox'+pageStr,
'orderedItems': [
],
'partOf': prefix+'://'+domain+'/users/'+username+'/outbox',
'type': 'OrderedCollectionPage'}
# counter for posts loop
postsOnPageCtr=0
# post filenames sorted in descending order
postsInOutbox=sorted(os.listdir(outboxDir), reverse=True)
# number of posts in outbox
outboxHeader['totalItems']=len(postsInOutbox)
prevPostFilename=None
if not pageNumber:
pageNumber=1
# Generate first and last entries within header
if len(postsInOutbox)>0:
lastPage=int(len(postsInOutbox)/itemsPerPage)
if lastPage<1:
lastPage=1
outboxHeader['last']= \
prefix+'://'+domain+'/users/'+username+'/outbox?page='+str(lastPage)
# Insert posts
currPage=1
postsCtr=0
for postFilename in postsInOutbox:
# Are we at the starting page yet?
if prevPostFilename and currPage==pageNumber and postsCtr==0:
# update the prev entry for the last message id
postId = prevPostFilename.split('#statuses#')[1].replace('#activity','')
outboxHeader['prev']= \
prefix+'://'+domain+'/users/'+username+'/outbox?min_id='+postId+'&page=true'
# get the full path of the post file
filePath = os.path.join(outboxDir, postFilename)
try:
if os.path.isfile(filePath):
if currPage == pageNumber and postsOnPageCtr <= itemsPerPage:
# get the post as json
with open(filePath, 'r') as fp:
p=commentjson.load(fp)
# insert it into the outbox feed
if postsOnPageCtr < itemsPerPage:
if not headerOnly:
outboxItems['orderedItems'].append(p)
elif postsOnPageCtr == itemsPerPage:
# if this is the last post update the next message ID
if '/statuses/' in p['id']:
postId = p['id'].split('/statuses/')[1].replace('/activity','')
outboxHeader['next']= \
prefix+'://'+domain+'/users/'+ \
username+'/outbox?max_id='+ \
postId+'&page=true'
postsOnPageCtr += 1
# remember the last post filename for use with prev
prevPostFilename = postFilename
if postsOnPageCtr > itemsPerPage:
break
# count the pages
postsCtr += 1
if postsCtr >= itemsPerPage:
postsCtr = 0
currPage += 1
except Exception as e:
print(e)
if headerOnly:
return outboxHeader
return outboxItems
def archivePosts(username: str,domain: str,baseDir: str,maxPostsInOutbox=256) -> None:
"""Retain a maximum number of posts within the outbox
Move any others to an archive directory
"""
outboxDir = createOutboxDir(username,domain,baseDir)
archiveDir = createOutboxArchive(username,domain,baseDir)
postsInOutbox=sorted(os.listdir(outboxDir), reverse=False)
noOfPosts=len(postsInOutbox)
if noOfPosts<=maxPostsInOutbox:
return
for postFilename in postsInOutbox:
filePath = os.path.join(outboxDir, postFilename)
if os.path.isfile(filePath):
archivePath = os.path.join(archiveDir, postFilename)
os.rename(filePath,archivePath)
# TODO: possibly archive any associated media files
noOfPosts -= 1
if noOfPosts <= maxPostsInOutbox:
break