scim-mozc occasionally ignores a first character of preediting after focus_in about mozc HOT 6 CLOSED

Thanks for the report. Reproduced on Ubuntu Lucid x86 (VMWare Player)

Actually here's a simpler repo step.

1. Type with SCIM-mozc
2. Focus out
3. kill mozc_server (execute killall mozc_server
4. Focus in and restart editing.

Let me explain the background of this bug.

Mozc client (e.g. SCIM-mozc, Mozc clients on TSF/IMKit) first creates a session 
(context) by calling CreateSession command. When a client no longer requires 
the session, the client has to call DeleteSession. However, it is not always 
guaranteed that all clients call DeleteSession method. For instance, 
DeleteSession might not be called when applications crash unexceptionally. When 
DeleteSession is not called, Zombie sessions remain in mozc_server heap memory, 
which causes a kind of memory leak.

To prevent such case, mozc server internally launches a watch dog thread 
(session/session_watch_dog.cc) and kills sessions that are not accessed for a 
fixed timeout (default timeout is  1 hour).  Mozc client library 
(client/session.cc) must be implemented so that the it works as expected even 
when the session is removed.

We have been working on allowing user to customize IME on/off key on Windows. 
This new feature causes a regression bug in the session management.
Similar issue is discovered in Google Japanese Input on Windows.  IME/On Off 
state is reset when session is killed.

http://www.google.co.jp/support/forum/p/ime/thread?tid=4e3f9d3a29e401d0&hl=ja

I got it. Thanks for your detailed description.

I think this bug is more terrible than the issue occurring on Windows because 
there is no way to know that the session has been deleted until the incorrect 
text is input. Please fix it as soon as possible.

@irieshinsuke

r31 has just been released. In the revision, the issue 17 should be fixed as 
well as other scim-mozc issues like state inconsistency on focus_out/reset 
events, Shift modifier problem on Kana input.

Could you check if they're fixed when you have time?

I tested r31 just now. The issues I pointed out about scim-mozc seem to be 
resolved. Thanks!

Thanks for the quick confirmation! Closing the bug.