bug#20454: 24.4; Emacs keyboard macros not working correctly

[Eli Zaretskii]

> Date: Wed, 29 Apr 2015 04:49:27 +0000 (UTC)
> From: Leo <used_to_be_leo@yahoo.com>
> 
> (progn
> (setq f5-key-map (make-sparse-keymap "F5-Key-prefix"))
> (define-key global-map [f5] f5-key-map)
> 
> (define-key f5-key-map [(g)] 'forward-word)
> (define-key f5-key-map [f6] 'forward-word)
> (define-key f5-key-map [kp-4] 'forward-word)
> (define-key f5-key-map [right] 'forward-word)
> )
> 
> Case This key
> # sequence Produces this result
> ---- -------- --------------------
> 1 <f5> <g> moves cursor forward by 1 word
> 2 <f5> <f6> moves cursor forward by 1 word
> 3 <f5> <kp-4> moves cursor forward by 1 word
> 4 <f5> <right> moves cursor forward by 1 word
> 
> Case Define macro by using Result when pressing
> # this key sequence: <f4> again:
> ---- --------------------- --------------------
> 11 <f3> <f5> <g> <f4> moves cursor forward by 1 word
> 12 <f3> <f5> <f6> <f4> nothing
> 13 <f3> <f5> <kp-4> <f4> nothing
> 14 <f3> <f5> <right> <f4> nothing
> 15 <f3> <f5> <f6> <right> <f4> moves cursor forward by 1 word
> 
> Cases #12, #13, and #14 do not work at all. Nothing happens, and
> there is no error message.

If you display the macro after defining it, e.g. with
kmacro-view-macro, or even simply evaluate kmacro-ring, you will see
that only f5 gets recorded there, the following function keys don't.
Only in the "f5 g" case will you see the 'g' in the macro.  That's why
the macro cannot be executed: there's only a prefix key there.

This happens because of the following trick we play in
read_char_minibuf_menu_prompt, which is a function we call whenever a
key we read is a prefix key:

      /* Make believe it's not a keyboard macro in case the help char
         is pressed.  Help characters are not recorded because menu prompting
         is not used on replay.  */
      orig_defn_macro = KVAR (current_kboard, defining_kbd_macro);
      kset_defining_kbd_macro (current_kboard, Qnil);
      do
        obj = read_char (commandflag, Qnil, Qt, 0, NULL);
      while (BUFFERP (obj));
      kset_defining_kbd_macro (current_kboard, orig_defn_macro);

      if (!INTEGERP (obj) || XINT (obj) == -2)
        return obj;

If 'obj' that we read is not a character (as is the case with f6 and
any other function key -- they are symbols), we return here without
recording the key in the macro, as we disabled that recording by
temporarily setting defining_kbd_macro to nil.

The trivial patch below fixes this, but before pushing it, I'd like
Stefan and others to eyeball this, in case there be dragons.

Btw, this has been broken "forever": I see it in Emacs 21.4.

--- src/keyboard.c~0    2015-04-26 07:42:05 +0300
+++ src/keyboard.c      2015-04-29 16:27:42 +0300
@@ -8708,7 +8708,11 @@ read_char_minibuf_menu_prompt (int comma
       kset_defining_kbd_macro (current_kboard, orig_defn_macro);
 
       if (!INTEGERP (obj) || XINT (obj) == -2)
-        return obj;
+       {
+         if (!NILP (KVAR (current_kboard, defining_kbd_macro)))
+           store_kbd_macro_char (obj);
+         return obj;
+       }
 
       if (! EQ (obj, menu_prompt_more_char)
          && (!INTEGERP (menu_prompt_more_char)