[PATCH 00/15] Hardware keyboard input for input methods

[Bill Spitzak]

It appears that both the input method and the client have to do the xkb 
decoding from hardware events to keysyms and shift information. And that 
the method used by the input method to produce "other" keys is to, in 
effect, run xkb backwards to produce fake hardware events from the 
desired result.

This seems rather wasteful to me, and I would expect some really 
confusing bugs if/when the two xkb interpreters disagree with each other.

Can't the input method send the result of running xkb? This would make 
it trivial for it to fake any xkb result. It also means the clients 
don't have to run an extra xkb decoder and respond to keymap changes. 
The main change is to make all the example clients use the input method 
for all keystroke events.

I may be misunderstanding this. Perhaps what I am suggesting is what is 
already being done, or what is intended before Wayland is finished? It 
just does not sound this way from the comments on the patches.