| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
|
|
|
|
When using gettext, the bot will look for English messages in the 'en'
locale, rather than en_US. This causes some Ruby GetText version to
not find the .mo files, which used to be stored as en_US instead. Since
Ruby's GetText seems to be too stupid to fall back to en_US when en is
not found, we move the messages to en.
This is also probably more correct from a linguistic standpoint, since
the messages are actually mostly British English due to the bot origin.
|
|
This prevents a dead loop when learning lines starting with whitespace,
which could otherwise happen e.g. when learning from a file.
|
|
|
|
|
|
|
|
|
|
This allows us to support servers that (ab)use the CASEMAPPING
to indicate the set of allowed characters for nicks (e.g. Inspire
IRCd).
|
|
|
|
Fix an error in commit 027b6965f5abf5d05a95 which causes @lang_list to
be undefined but used in BabelFish.
|
|
Previously the translation services in the plugin would connect to their
respective websites on initialize. Now they will only do this when the first
time they are used. WorldLingo still connects on initialialize in order to list
the supported language pairs, but the Mechanize object is not saved here, so
the connection should be closed after initialize, and only reopened if
translation by WorldLingo is requested later.
Previously the services are assumed to fail if they raise any Exception in
initialize, and in that case the service is disabled, and its command as well
as the help translate [failed] commands state this. Now this exception catch is
done for any do_translate method call.
|
|
Somebody should hang the ActiveSupport developers by their balls
with barbed wire. Their MissingSourceFile extension to LoadError
_expects_ a second argument, breaking the usual Exception interface
(instead, the smart thing to do would have been to make the second
parameter optional and run the code in the from_message method if
it was missing).
|
|
|
|
We get cooler github readme.
|
|
When translate command is used without source language, "auto" as source
language is assumed. It means that google translator is used and we let
google figure out what the source language is.
Problem is that the google translator will fail if the system that the
bot is running on does not have the json gem installed. This commit
addresses that problem by showing more informative error message instead
of a weird "none of the translators supports auto to en translation".
|
|
|
|
|
|
|
|
|
|
|
|
|
|
For the 'before' part of the reply, make sure we use the where
from the correct Saw. Otherwise, for example, a public message
preceeding a quit would show a dangling 'in ' at the end of the reply.
|
|
|
|
|
|
|
|
This commit introduces the ability to note that a user was doing
something, optionally without disclosing what or where it was.
Users themselves do not get the chance to be hidden, because you can ask
the /WHOIS service if they logged on at all.
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
|
|
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
|
|
This gets the list of channels we think we are in (not what the server says we are in).
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
|
|
My fault for not finding it in review
|
|
|
|
|
|
The new Freenode ircd returns multiple RPL_WHOISCHANNELS lines (split
based on length), and they must be merged, otherwise only the last ends
up as the list of channels that the bot is in.
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
|
|
|
|
It should now follow this logic:
* Allow everyone to create basic replies
* Require specific permissions for cmd and ruby reactions
|
|
|
|
|
|
Anybody can create a dangerous reaction, and then trick the owner into
triggering it.
Credits to apoc for spotting this.
|
|
and filter it in smart way
|
|
markov.delay had the wrong default, and its use didn't have any
degree of randomness in it. Set the default to 5 and use rand()
again.
|
|
Unify probability check for addressed and non-addressed case by
passing the message as a parameter to should_talk and picking the
probability accordingly.
|
|
|
|
These were slowing down learning process greatly and do not make sense as far
as I can see: learning is always done only by single thread anyway.
|
|
|
|
|
|
|
|
|
