Linking of message accounts is now significantly less likely to produce
false-positives.
Previously, any hostmasks with matching nick!*@* would be strongly linked
together. This led to falsely-linking accounts, either inadvertently or
intentionally.
For example, Bob might also be known as Bob_ and Bobby,
but primarily uses Bob as his main nick. Somebody else might join with
Bobby and end up being linked to Bob. Now both Bob and the new Bobby are
linked together as the same person, but likely with different *!user@host.
Now if the new Bobby ever gets banned, then Bob will also end up being
banned for evading Bobby's ban.
This was a sore spot in the previous linking implementation.
This new implementation has several adjustments to more intelligently link
accounts only when they're proven beyond a reasonable doubt to be the same
person (e.g. by matching nickserv accounts, etc).
Consequently, rather than aggressively linking accounts and catching more
ban-evaders at the risk of potentially falsely-linking accounts and falsely
detecting innocent people as ban-evaders, this new implementation will instead
link accounts more reliably at the risk of potential ban-evaders not yet being
linked together and thus being able to evade a ban.
This is a more preferable and reasonable risk. Active channel ops should be
able to catch any obnoxious ban-evaders that slip through this net.
PBot will now use weak links if the ip address portion of a hostmask
hasn't been seen in the last 48 hours in order to prevent false-positive
linking of dynamic ip addresses.
Weak links are excluded from ban-evasion logic.
Weak links can be manually upgraded to strong links with the `akalink`
command if a human confirms they are the same person through the `aka -w`
command.
Individuals with matching nicks or matching nickserv accounts, etc, will
automatically be strongly linked, as usual.
Improve linking of known aliases for users by using an aliases table
to track the linkages.
Improve check-bans implementation to use new aliases table for
significant performance gains and reduced resource usage.
This significantly speeds up certain SQL actions on
the Messages table that were done for anti-flood detection.
The bot should now be significantly more quicker to catch
floods promptly without delay.
(Also, comment out some logging messages. Perhaps revisit this
and wrap them with a debug registry variable instead.)
Add the -b/-before and -a/-after flags to the `recall` command to display
that many lines of surrounding context.
For example, `recall -c ##c -b 100` will show the last 100 messages in the ##c
channel.
Another example, `recall pragma- 50 -b 5 -a 5` will show the 5 messages before the 50th
most recent message spoken by pragma- and then show that 50th message and finally show
the 5 messages after it.
Allows changing of bot configuration values without needing to restart
bot instance or needing to edit pbot.pl script.
Registry will initially be populated with default values from pbot.pl,
but if a registry file exists then the registry values will take
precedence over the pbot.pl values. For instance, if you regset the
bot trigger to '%' then the trigger will be '%' even if pbot.pl has '!'
or something else explicitly set.
Some registry items can have trigger hooks associated with them. For
instance, the irc->botnick registry entry has a change_botnick_trigger
associated with it which changes the IRC nick on the server when a new
value is set via regset/regadd.
Tons of other fixes and improvements throughout.
The nick field is now optional and Getopt::Long is used to accept targeted
options for channel/history, allowing recalls by text/channel for any nick
which is useful when you know what text the message contained but not who
said it.
Added MessageHistory.pm and MessageHistory_SQLite.pm. May eventually port
and add MessageHistory_Hashtable.pm as was done with Quotegrabs, but this is
not particularly high on the todo list.
Antiflood.pm has been updated to use the new MessageHistory API.
The `recall` command has been moved from Quotegrabs into MessageHistory. It
also now has the ability to ignore messages containing the recall command
itself, for improved usability.
Likewise, the `grab` command will now ignore previous `grab` commands when
grabbing by regex in order to prevent accidentally grabbing previous grab
attempts.
The `join` and `part` commands have been improved to accept multiple channels,
and `part` will use the current channel if none is provided.
