Emacs dedicated completion window

I don't believe this is possible, since each time you're invoking a subshell to source the script you're invoking a child process from the original process where the emacs applications was launched.

The exporting of environment variables is a one way street where only the parent can provide variables to any child processes, but no child processes can manipulate the parent's environment.

Experiment

I'm using vim but the same should apply to emacs. Sample file to source.

$ more ~/vars.bash 
export VAR=somevalue

1. Initial Parent environment, $VAR is unset

   $ echo $VAR
   
   $
   

2. Launch vim. Then invoke a subshell to source the above file (:sh).

   # check variable
   $ echo $VAR
   
   $
   
   # source and re-check
   $ source ~/vars.bash
   $ echo $VAR
   somevalue
   

3. Exit subshell, return to vim. Then invoke another subshell (:sh).

   $ exit
   
   ... back in vim, do another `:sh` ...
   
   # check variable
   $ echo $VAR
   
   $ 
   

Sorry for your trouble. I don't see the problem you report, following your recipe. Perhaps the description is not complete? I can turn on both pretty-control-l-mode and whitespace-mode, and the behavior I see for each seems normal. Perhaps there is some custom setting you use for whitespace-style or something?

Anyway, maybe it would help if you make a change like this to pretty-control-l-mode. If so, let me know and I will apply it to pp-c-l.el. (To test, set the new option to nil.)

 (defcustom pp^L-use-window-display-table-flag t
   "Non-nil: use `window-display-table'; nil: use `buffer-display-table`."
   :type 'boolean :group 'Pretty-Control-L)

 (define-minor-mode pretty-control-l-mode
     "Toggle pretty display of Control-l (`^L') characters.
 With ARG, turn pretty display of `^L' on if and only if ARG is positive."
   :init-value nil :global t :group 'Pretty-Control-L
   (if pretty-control-l-mode
       (add-hook 'window-configuration-change-hook 'refresh-pretty-control-l)
     (remove-hook 'window-configuration-change-hook 'refresh-pretty-control-l))
   (walk-windows
    (lambda (window)
      (let ((display-table  (if pp^L-use-window-display-table-flag ; <=========
                                (or (window-display-table window)
                                    (make-display-table))
                              (if buffer-display-table
                                  (copy-sequence buffer-display-table)
                                (make-display-table)))))
        (aset display-table ?\014 (and pretty-control-l-mode
                                       (pp^L-^L-display-table-entry window)))
        (if pp^L-use-window-display-table-flag                     ; <=========
            (set-window-display-table window display-table)
          (setq buffer-display-table display-table))))
    'no-minibuf
    'visible))

UPDATED to add comment thread, in case comments get deleted at some point:

BTW, I wonder if the hierarchy of display tables described in the doc shouldn't perhaps be applied using inheritance of some kind. Seems a bit primitive for one level (e.g. window) to completely shadow a lower level (e.g. buffer). You might consider sending a question about this to M-x report-emacs-bug. – Drew Sep 24 '14 at 16:36

Ping? Could you please let me know if the change above helps? Thx. – Drew Oct 14 '14 at 18:12

I just read this answer (I have not been around this part of the Internet for a while...). I will check this when I get round to it, perhaps in a few days or so. I'll get back with an ‘Answer approved’ (if it works), or comments (otherwise), as appropriate, later. – Johan E Oct 25 '14 at 22:32

I edited the question to add a more fleshed-out recipe for showing the problem. I'd be interested whether you get the same results. --- Also, is there a way to shadow a system installed .el-file with a user-supplied one (I'm really just a “user”, not a lisp-programmer...)? I don't really feel like messing with the files installed by deb-packages. (That's why I did the problem-recipe before testing your answer...) – Johan E Oct 27 '14 at 1:02

Five seconds after I wrote the last comment I realized that I could just paste the code into scratch and C-j-run it to test. (No need to edit any files.) The results: It works a charm! Thank you! (=> Answer accepted) However, I'd still like to know if you get the same results as I from my problem-recipe (before patching the code). – Johan E Oct 27 '14 at 1:09

I just followed your new recipe, and I saw everything you described (so clearly). And then I read the new comment that you just added. Glad to know that things work OK. Thx for your feedback. – Drew Oct 27 '14 at 1:12