Kazz> It is very old article... But I saved this because I have an Kazz> idea. SL Baur <steve(a)xemacs.org&gt; writes: Kazz> No. As I said before (maybe), the current code lacks of Kazz> register/unregister the input method. Instead of the Kazz> implementation, input-method-xlib just opens XIM and attach Kazz> it to the context. (It is very same with Motif, but Motif Kazz> does this smart internally) It's not clear to me what happens in your method if you have just typed half of a kana in romaji.... Once the input context is open, I'm not sure what happens if the connection dies. The man page does not specify error returns for X*LookupString. If the call blocks, we're screwed anyway. This clearly reduces the possibility of the holocaust scenario, though. Kazz> Instead of just opening the XIM, we should set the code to: Kazz> 1. when opening the context, we should set the callback Kazz> for XIM that if the XIM is available for the context, use Kazz> it. Are you suggesting that the context should be opened and closed on every toggle of the input method? This reduces the chance of dying much more, but I don't know how inefficient it is. Kazz> XRegisterIMInstantiateCallback Kazz> 2. when invoking the XIM, we should register the Kazz> callback that if the XIM is unavailable, then release the Kazz> XIM. Kazz> XUnregisterIMInstantiateCallback Kazz> I'd been having this idea 1 year, but I have no coding power Kazz> and the enough time for this. Any Xpert? This looks like it solves part of the problem, but we should be doing something to integrate XIM with other methods anyway. XIM is the only major input protocol left that hasn't been LEIM-ized. (I guess the iso-*.el methods haven't been, but that's going to be trivial. (push todo "LEIM-ize iso-*.el methods") :-) What is needed is to move control of initialization and closing of XIM methods to the Lisp level so that XIM methods get opened and closed when we do select-input-method. I'm pretty sure Martin Buchholz has been thinking about this, too.