On 03/28/2014 07:18 PM, Daniel Colascione wrote:
> Say we're using a mode derived from comint that implements completion by
> using the comint redirection functionality to send commands to the
> process associated with the comint buffer. Say we have TAB bound to
> complete-symbol. If the user presses TAB (to create a list of
> completions) and then immediately presses RET to run comint-send-input,
> we send the input to the subprocess, but don't wait for a reply. Then we
> run post-command-hook; completion-in-region--postch is on the list of
> hooks to run. This function runs completion-in-region-mode--predicate,
> which makes a hidden call to the subprocess; this hidden call involves
> writing a command waiting for a reply. But because we just sent the
> *user* line in comint-send-input, we might actually read the response to
> *that* line instead of the internal completion command, causing an
> error. The response to the internal completion command then appears in
> the comint buffer.
> 
> Why can't we do the completion-in-region--postch stuff in pre-command-hook?

I think this fix works, but I'm not particularly familiar with the
completion code. Can someone review?

=== modified file 'lisp/comint.el'
--- lisp/comint.el	2014-03-22 22:12:52 +0000
+++ lisp/comint.el	2014-03-29 02:36:49 +0000
@@ -1769,6 +1769,12 @@

 Similarly for Soar, Scheme, etc."
   (interactive)
+  ;; If we're currently completing, stop.  We're definitely done
+  ;; completing, and by sending the input, we might cause side effects
+  ;; that will confuse the code running in the completion
+  ;; post-command-hook.
+  (when completion-in-region-mode
+    (completion-in-region-mode -1))
   ;; Note that the input string does not include its terminal newline.
   (let ((proc (get-buffer-process (current-buffer))))
     (if (not proc) (user-error "Current buffer has no process")