Re: pre-command-hook with input methods

[Stefan Monnier <monnier@IRO.UMontreal.CA>]

> So, the pre-command-hook runs only *after* the multi-key press behaviour
> has been completed (either by introducing a diacritical letter or by not
> doing so). Hence on pressing "b" it gets run twice -- once to say "a has
> been entered" and once to say "b has been entered".

This is correct, since when you hit `a' the self-insert-command is not
run yet (instead the `a' char is inserted in the buffer by the input
method, which then waits for the next key to see which character was
really meant); it's only run after you hit `b'.

> All fine, but it's breaking my completion framework which removes
> previously offered completions on the pre-command-hook.

Indeed, I wouldn't be surprised if several other uses of
pre-command-hook could suffer from unexpected behaviors in
such situations.

> I need it to run as soon as the "a" key has been pressed.

Of course, you can do that using one of the input-method hooks (maybe
input-method-after-insert-chunk-hook), or you could rely on
before-change-functions.

> Is there a better hook?

I don't think so, sadly.

Maybe we could introduce a new hook like `after-idle-hook'.  Part of the
question is what should happen if a timer or a process filter runs
(these can run while Emacs is "idle")?  How 'bout when that
timer/process-filter inserts text near/at point?


        Stefan