Bug 15853

For some reason, Qt3 plugin instance for input methods is created (and destroyed) twice, and this seems to cause a problem for uim-pref-qt when libquiminputcontextplugin.so is installed.

One way to avoid such conditions is to count initialization of libuim (see attached patch).  Or just remove uim_quit() in UimInputContextPlugin::uimQuit() as a workaround.

Created attachment 16424 [details]
Patch to count libuim's initialization

uim-pref-qt works fine now.
Thank you so much.

I think this bug is critical.
Someone might want uim-1.5.1.

Thank you for the report and the patch.

But IMHO, uim_init() and uim_quit() is not the place to be added the workaround, at least in 1.5 branch since it alters the API specification of the functions. Please resolve the problem in bridge layer.

See [Anthy-dev 1563] and following messages for further consideration. We had discussed such behavior once, and reverted to the current specification.

(In reply to comment #4) 
> But IMHO, uim_init() and uim_quit() is not the place to be added the
> workaround, at least in 1.5 branch since it alters the API specification of the
> functions. Please resolve the problem in bridge layer.
> 
> See [Anthy-dev 1563] and following messages for further consideration. We had
> discussed such behavior once, and reverted to the current specification.

Hehe, you already considered and tried this init count three years ago, which I don't remember well ;)

Do you think just removing uim_quit() in UimInputContextPlugin::uimQuit() is OK for now?  (Personally, I don't prefer this workaround because it must be a bug in immodule patch for Qt3).

> Do you think just removing uim_quit() in UimInputContextPlugin::uimQuit() is OK
> for now?  (Personally, I don't prefer this workaround because it must be a bug
> in immodule patch for Qt3).
Though I don't know sufficient information about the Qt plugin management on immodule arena, the init count workaround is better than just removing uim_quit() since it is more proper and can behave as expected by Qt on any situations we don't know. I don't prefer introducing future possibility of another problem by removing uim_quit().

What about writing a wrapper functions for uim_init() and uim_quit() contaning the your init count workaround in the bridges and replace the uim_init/quit call with them?

Created attachment 16476 [details] [review]
init wrapper patch

(In reply to comment #6)

> What about writing a wrapper functions for uim_init() and uim_quit() contaning
> the your init count workaround in the bridges and replace the uim_init/quit
> call with them?

That should be fine.  I'll commit the above new patch if there is no further discussion. 

(In reply to comment #8)
> That should be fine.  I'll commit the above new patch if there is no further
> discussion. 


I noticed that the using convenience library cannot share a variable among plugins (uim_panelapplet.so and libquiminputcontextplugin.so in this case).  So I'll commit the patch except for kde applet part.  In addition, I'm going to remove unused reloadUim() in immodule.

Created attachment 16479 [details] [review]
Revised patch

What about these names?

uim_counted_init()
uim_counted_quit()
uim_force_quit_regardless_of_count()

The 3rd name is bit too long, but I think it'll be used rare and so such descriptive name is preferable/acceptable.

(In reply to comment #11)
> What about these names?
> 
> uim_counted_init()
> uim_counted_quit()
> uim_force_quit_regardless_of_count()

OK.  I've committed this with putting the convenience library into uim/.