pcomplete makes assumptions about its interactive entry points

pcomplete makes assumptions about its interactive entry points

[Vivek Dasmohapatra]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 481 bytes --]

Forwarded at author's request

---------- Forwarded message ----------
Date: Tue, 3 Sep 2013 01:35:06 +0100 (BST)
From: Vivek Dasmohapatra <vivek@etla.org>
To: John Wiegley <johnw@gnu.org>
Subject: pcomplete

Hi - pcomplete makes an assumption (I think) about the interactive
entry point(s) which can be used to drive it which means that non-
native entry points don't get completion cycling.

Attached is a possible fix plus the alteration required
to make erc completion use it.

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: Type: TEXT/X-DIFF; NAME=pcomplete-entry-points.patch, Size: 2136 bytes --]

diff --git a/lisp/erc/erc.el b/lisp/erc/erc.el
index 01efd83..3917638 100644
--- a/lisp/erc/erc.el
+++ b/lisp/erc/erc.el
@@ -3964,7 +3964,9 @@ This places `point' just after the prompt, or at the beginning of the line."
     (kill-line)))
 
 (defun erc-complete-word-at-point ()
-  (run-hook-with-args-until-success 'erc-complete-functions))
+  (let ((pcomplete-extra-entry-points
+	 (cons 'completion-at-point pcomplete-extra-entry-points)))
+  (run-hook-with-args-until-success 'erc-complete-functions)))
 
 (define-obsolete-function-alias 'erc-complete-word 'completion-at-point "24.1")
 
diff --git a/lisp/pcomplete.el b/lisp/pcomplete.el
index 957505f..30a743b 100644
--- a/lisp/pcomplete.el
+++ b/lisp/pcomplete.el
@@ -471,6 +471,15 @@ Same as `pcomplete' but using the standard completion UI."
                           (plist-get :predicate (nthcdr 3 data)))))
 
 ;;; Pcomplete's native UI.
+(defvar pcomplete-extra-entry-points nil
+  "When an non-native command is driving `pcomplete', that command's
+symbol must be placed in this variable while `pcomplete' is invoked
+in order to allow completion cycling to occur.\n
+For example:\n
+  When completing ab → aba abc abd able via `completion-at-point',\n
+which [may] delegate to `pcomplete' via `completion-at-point-functions',
+the functions in that list should bind this variable to include the
+symbol 'completion-at-point.")
 
 ;;;###autoload
 (defun pcomplete (&optional interactively)
@@ -481,9 +490,10 @@ completion functions list (it should occur fairly early in the list)."
   (if (and interactively
 	   pcomplete-cycle-completions
 	   pcomplete-current-completions
-	   (memq last-command '(pcomplete
-				pcomplete-expand-and-complete
-				pcomplete-reverse)))
+           (or (memq last-command '(pcomplete
+                                    pcomplete-expand-and-complete
+                                    pcomplete-reverse))
+               (memq last-command pcomplete-extra-entry-points)))
       (progn
 	(delete-char (- pcomplete-last-completion-length))
 	(if (eq this-command 'pcomplete-reverse)

Re: pcomplete makes assumptions about its interactive entry points

[Stefan Monnier]

> Hi - pcomplete makes an assumption (I think) about the interactive
> entry point(s) which can be used to drive it which means that non-
> native entry points don't get completion cycling.

> Attached is a possible fix plus the alteration required
> to make erc completion use it.

Could you explain in which circumstance this is needed?


        Stefan


PS: And could you make a bug report about it, so we can track it better,
rather than discussing it on emacs-devel?

Re: pcomplete makes assumptions about its interactive entry points

[Vivek Dasmohapatra]

On Thu, 12 Sep 2013, Stefan Monnier wrote:

>> Hi - pcomplete makes an assumption (I think) about the interactive
>> entry point(s) which can be used to drive it which means that non-
>> native entry points don't get completion cycling.
>
>> Attached is a possible fix plus the alteration required
>> to make erc completion use it.
>
> Could you explain in which circumstance this is needed?

In an erc buffer, I tried to tab-complete a nick: The initial tab completion
showed many possibilities - I typed an extra character, which should have
limited it to a small enough number for tab-completion-cycling to kick in...
but instead TAB just unconditionally completed to the first of the remaining
candidates, and never offered me the chance to pick any of the others.

It turns out that pcomplete assumes that the interactive entry point
(activating command) can only ever be one of a handful of commands,
and specifically checks for those when deciding whether to enter the
code path that starts cycling through the remaining completions.

So modes that use pcomplete but define their own commands for tab 
completion lose the ability to use pcomplete cycling unless they have
a way of adding themselves to the list of entry points that pcomplete
knows about.