michael.cadilhac@lrde.org (Michaėl Cadilhac) writes:

>   Hi !
>
>   I use the latest CVS version.
>
>   With the following :
>
> $ emacs -Q
>
> C-u M-x ispell-change-dictionary RET francais RET
> C-x b test1 RET
> M-x ispell-change-dictionary RET english RET
> M-x ispell-buffer
> C-x b test2 RET
> M-x ispell-change-dictionary RET english RET
>
>   The « english » dictionary will be loaded twice: on the
>   « ispell-buffer » and on the latest « ispell-change-dictionary ».
>
>   The rest of this message is IMHO :-)
>
>   What happens is the following on the second ispell-change-dictionary:
>
>   - ispell-buffer-local-dict is called,
>   - this one calls ispell-internal-change-dictionary,
>   - ispell-local-dictionary being nil, and current dictionary
>     (« english ») not being equal to default one (« francais »), the
>     ispell process is killed.
>
>   So it has to be restarted.
>
>   The call to ispell-buffer-local-dict is mandatory
>   (ispell-change-dictionary could be called with "" just to load
>   default buffer's one), but it should not call
>   ispell-internal-change-dictionary at this time.
>
>   The proposed patch :
>   - Delays this call,
>   - Makes ispell-internal-change-dictionary checks for pdict,
>   - Fixes the bug.

  Well,  this  bug  is  still  triggered in  some  cases,  but  before
  I investigate, I was  wondering why ispell-kill-process took so much
  time to execute.

  This function does the following:
  - Send EOF to ispell,
  - Read its output if there is, timeout to 1 sec,
  - Kill the process if it's still not,
  - Wait for it to be really killed (sleeping for 0.25 sec between checks).

  I don't want to be ... rude, I'm really a pacifist actually, but why
  not just delete-process it ?

  I often have some flyspellized buffers in English and some others in
  French, the time I have to wait on every C-x o is kind of
  disturbing...

  I propose the following change: