bug#29349: read_key_sequence is only partially recursive. This is a bug.

[Alan Mackenzie]

Hello, Emacs.

On Sat, Nov 18, 2017 at 09:38:43 +0000, Alan Mackenzie wrote:
> In branch emacs-26.

> I came across this bug whilst working on bug #29272 ("C-h k C-mouse-3"
> followed by menu selection asks for more keys).

> In a Linux tty using the GPM mouse package, doing read_key_sequence (the
> C function in keyboard.c), when a menu action is activated,
> read_key_sequence calls itself recursively to handle all the mouse
> manipulation.

> Unfortunately, the variable raw_keybuf_count is initialised to 0 in
> r_k_s.  This includes in the recursive call.  This variable indexes the
> global buffer raw_keybuf, which accumulates the raw events which make up
> the key sequence.

> The result of this is that the events in the recursive call overwrite
> the stored events of the outer r_k_s call, leaving a mess.

> r_k_s is static in keyboard.c and is called from three functions:
> command_loop_1, read_menu_command (the one that gives the trouble), and
> read_key_sequence_vs.

> So I propose as a solution that raw_keybuf_count be initialised to zero
> in two of these three functions, but not in read_menu_command (and no
> longer in read_key_sequence).  I've tried this, and it seems to work.
> It has the disadvantage of being ugly, and it makes read_menu_command
> only callable as a subfunction of r_k_s.

> Has anybody any thoughts on this?

Here is how I propose to solve this:

(i) In keyboard.c, the static variables raw_keybuf and raw_keybuf_count
will become pointers.  They will initially point to a static buffer and
a static integer.  For safety's sake, they will be reinitialised to
those static variables in command_loop_1(), just before the invocation
of read_key_sequence().

(ii) read_key_sequence() will get a Lisp_Object buffer and a count
variable as local variables.  Around the call to read_char(),
raw_keybuf{,_count} will be set to point to these locals, so that should
read_char encounter a menu, its events will be stored in the local copy
of the Lisp_Object buffer.

(iii) On return from read_char, if any events are in the local buffer,
they will be appended to the buffer in the enclosing scope.  The global
pointers raw_keybuf{,_count} will be restored to their previous values.

In short, raw_keybuf and raw_keybuf_count will be "bound" to local
variables around the call to read_char().

Comments?

-- 
Alan Mackenzie (Nuremberg, Germany).