161 lines
4.7 KiB
Python
Executable File
161 lines
4.7 KiB
Python
Executable File
#!/usr/bin/env python
|
|
import os
|
|
import sre
|
|
|
|
from pysqlite2 import dbapi2 as sqlite
|
|
|
|
if os.name == 'nt':
|
|
try:
|
|
PATH_TO_LOGS_BASE_DIR = os.path.join(os.environ['appdata'], 'Gajim', 'Logs')
|
|
PATH_TO_DB = os.path.join(os.environ['appdata'], 'Gajim', 'logs.db') # database is called logs.db
|
|
except KeyError:
|
|
# win9x
|
|
PATH_TO_LOGS_BASE_DIR = 'Logs'
|
|
PATH_TO_DB = 'logs.db'
|
|
else:
|
|
PATH_TO_LOGS_BASE_DIR = os.path.expanduser('~/.gajim/logs')
|
|
PATH_TO_DB = os.path.expanduser('~/.gajim/logs.db') # database is called logs.db
|
|
|
|
jids_already_in = [] # jid we already put in DB
|
|
con = sqlite.connect(PATH_TO_DB)
|
|
cur = con.cursor()
|
|
# create the tables
|
|
# type can be 'gc', 'gcstatus', 'recv', 'sent', 'status'
|
|
# logs --> jids.jid_id but Sqlite doesn't do FK etc so it's done in python code
|
|
cur.executescript(
|
|
'''
|
|
CREATE TABLE jids(
|
|
jid_id INTEGER PRIMARY KEY AUTOINCREMENT,
|
|
jid TEXT UNIQUE
|
|
);
|
|
|
|
CREATE TABLE logs(
|
|
log_line_id INTEGER PRIMARY KEY AUTOINCREMENT UNIQUE,
|
|
jid_id INTEGER,
|
|
contact_name TEXT,
|
|
time INTEGER,
|
|
type TEXT,
|
|
show TEXT,
|
|
message TEXT
|
|
);
|
|
'''
|
|
)
|
|
|
|
con.commit()
|
|
|
|
# (?<!\\) is a lookbehind assertion which asks anything but '\'
|
|
# to match the regexp that follows it
|
|
re = sre.compile(r'(?<!\\)\\n')
|
|
|
|
def from_one_line(msg):
|
|
# So here match '\\n' but not if you have a '\' before that
|
|
msg = re.sub('\n', msg)
|
|
msg = msg.replace('\\\\', '\\')
|
|
# s12 = 'test\\ntest\\\\ntest'
|
|
# s13 = re.sub('\n', s12)
|
|
# s14 s13.replace('\\\\', '\\')
|
|
# s14
|
|
# 'test\ntest\\ntest'
|
|
return msg
|
|
|
|
|
|
def get_jid(dirname, filename):
|
|
# TABLE NAME will be JID if TC-related, room_jid if GC-related,
|
|
# ROOM_JID/nick if pm-related
|
|
if dirname.endswith('logs') or dirname.endswith('Logs'):
|
|
# we have file (not dir) in logs base dir, so it's TC
|
|
jid = filename # file is JID
|
|
else:
|
|
# we are in a room folder (so it can be either pm or message in room)
|
|
if filename == os.path.basename(dirname): # room/room
|
|
jid = dirname # filename is ROOM_JID
|
|
else: #room/nick it's pm
|
|
jid = dirname + '/' + filename
|
|
|
|
if jid.startswith('/'):
|
|
p = len(PATH_TO_LOGS_BASE_DIR)
|
|
jid = jid[p:]
|
|
jid = jid.lower()
|
|
return jid
|
|
|
|
def visit(arg, dirname, filenames):
|
|
for filename in filenames:
|
|
# Don't take this file into account, this is dup info
|
|
# notifications are also in contact log file
|
|
if filename == 'notify.log':
|
|
continue
|
|
path_to_text_file = os.path.join(dirname, filename)
|
|
if os.path.isdir(path_to_text_file):
|
|
continue
|
|
|
|
jid = get_jid(dirname, filename)
|
|
print 'Processing', jid
|
|
# jid is already in the DB, don't create the table, just get his jid_id
|
|
if jid in jids_already_in:
|
|
cur.execute('SELECT jid_id FROM jids WHERE jid="%s"' % jid)
|
|
else:
|
|
jids_already_in.append(jid)
|
|
cur.execute('INSERT INTO jids (jid) VALUES (?)', (jid,))
|
|
con.commit()
|
|
|
|
cur.execute('SELECT MAX(jid) FROM jids')
|
|
JID_ID = cur.fetchone()[0]
|
|
|
|
f = open(path_to_text_file, 'r')
|
|
lines = f.readlines()
|
|
for line in lines:
|
|
line = from_one_line(line)
|
|
splitted_line = line.split(':')
|
|
if len(splitted_line) > 2:
|
|
# 'gc', 'gcstatus', 'recv', 'sent' and if nothing of those
|
|
# it is status
|
|
type = splitted_line[1] # line[1] has type of logged message
|
|
message_data = splitted_line[2:] # line[2:] has message data
|
|
# line[0] is date,
|
|
|
|
# some lines can be fucked up, just trop them
|
|
try:
|
|
tim = int(float(splitted_line[0]))
|
|
except:
|
|
continue
|
|
|
|
sql = 'INSERT INTO logs (jid_id, contact_name, time, type, show, message) '\
|
|
'VALUES (?, ?, ?, ?, ?, ?)'
|
|
|
|
contact_name = None
|
|
show = None
|
|
if type == 'gc':
|
|
contact_name = message_data[0]
|
|
message = ':'.join(message_data[1:])
|
|
elif type == 'gcstatus':
|
|
contact_name = message_data[0]
|
|
show = message_data[1]
|
|
message = ':'.join(message_data[2:]) # status msg
|
|
elif type in ('recv', 'sent'):
|
|
message = ':'.join(message_data[0:])
|
|
else: # status
|
|
type = 'status'
|
|
show = message_data[0]
|
|
message = ':'.join(message_data[1:]) # status msg
|
|
|
|
values = (JID_ID, contact_name, tim, type, show, message)
|
|
cur.execute(sql, values)
|
|
con.commit()
|
|
|
|
if __name__ == '__main__':
|
|
os.path.walk(PATH_TO_LOGS_BASE_DIR, visit, None)
|
|
f = open(os.path.join(PATH_TO_LOGS_BASE_DIR, 'README'), 'w')
|
|
f.write('We do not use plain-text files anymore, because they do not scale.\n')
|
|
f.write('Those files here are logs for Gajim up until 0.8.2\n')
|
|
f.write('We now use an sqlite database called logs.db found in ~/.gajim\n')
|
|
f.write('You can always run the migration script to import your old logs to the database\n')
|
|
f.write('Thank you\n')
|
|
f.close()
|
|
# after huge import create the indices (they are slow on massive insert)
|
|
cur.executescript(
|
|
'''
|
|
CREATE UNIQUE INDEX jids_already_index ON jids (jid);
|
|
CREATE INDEX JID_ID_Index ON logs (jid_id);
|
|
'''
|
|
)
|