GNU bug report logs - #68272
[PATCH] Fix -1 leaking from C to lisp in 'read-event' etc.

Previous Next

Full log

Message #38 received at 68272 <at> debbugs.gnu.org (full text, mbox):

From: Stefan Monnier <monnier <at> iro.umontreal.ca>
To: Tim Ruffing <crypto <at> timruffing.de>
Cc: Eli Zaretskii <eliz <at> gnu.org>, 68272 <at> debbugs.gnu.org
Subject: Re: bug#68272: [PATCH] Fix -1 leaking from C to lisp in
 'read-event' etc.
Date: Tue, 05 Mar 2024 11:45:41 -0500

> Hi, this is an updated patch set.

Looks really nice, thank you.
Comments/nitpicks below.

Eli Zaretskii [2024-03-05 15:10:39] wrote:
> IOW, what about callers that actually _want_ to know when the macro
> ends prematurely?

I couldn't find any, really.  Of course, we could export
`at_end_of_macro_p` to ELisp, but I don't see any need for it.

[ Also: define "prematurely".  My impression is that the callers of
  `read-char` are generally not in a position to know what is premature
  and what isn't because it tends to depend on the users' intentions.   ]

> +/* Whether the execution of a macro has reached its end.
> +   It makes only sense to call this when while executing a macro.  */
               ^^^^                    ^^^^
                 \----------------------^

[ And same for the other copy of that coment.  ]

> +/* Return true if there are any pending requeued events (command events
> +   or events to be processed by input methods).  */

I think I'd say "other levels of the input processing stages" instead of
"input methods", so as to conceptually include any other
"unread_*_events" we may end up with.

> -      /* If not, we should actually read a character.  */
> +      /* If we're at the end of a macro, exit it by returning 0,
> +	 unless there are unread events pending.  */
> +      else if (!NILP (Vexecuting_kbd_macro)
> +	  && at_end_of_macro_p ()
> +	  && !requeued_events_pending_p ())
> +	{
> +	  t = 0;
> +	  /* The Microsoft C compiler can't handle the goto that
> +	     would go here.  */
> +	  dummyflag = true;
> +	  break;
> +	}

This "Microsoft C compiler" business dates back to 1994 (commit
bc536d847736f466727453ca6aa7c07aef6fce46).
I think it's safe to clean it up now :-)

> index 62129be1629..98290e7e276 100644
> --- a/src/macros.c
> +++ b/src/macros.c
> @@ -314,6 +314,29 @@ DEFUN ("execute-kbd-macro", Fexecute_kbd_macro, Sexecute_kbd_macro, 1, 3, 0,
>  		      Vreal_this_command));
>    record_unwind_protect (pop_kbd_macro, tem);
>  
> +  /* The following loop starts the execution of the macro.  Individual
> +     characters from the macro are read by read_char, which takes care
> +     of incrementing executing_kbd_macro_index.  The end of the
> +     macro is handled as follows:
> +     - read_key_sequence asks at_end_of_macro_p whether the end of
> +     (one iteration of the macro) has been reached.  If so, it returns
> +     the magic value 0 to command_loop_1.
> +     - command_loop_1 returns Qnil to command_loop_2.
> +     - command_loop_2 returns Qnil to this function
> +       (but only the returning is relevant, not the actual value).

Could you complete the sequence to the point where we clear
Vexecuting_kbd_macro?

> +     If read_char happens to be called at the end of the macro, but
> +     before read_key_sequence could handle the end (e.g., because lisp
> +     code calls 'read-event', 'read-char', and 'read-char-exclusive'),
> +     read_char will simply continue reading other available input
> +     (Bug#68272).

Could you clarify here what happens w.r.t the value of
Vexecuting_kbd_macro (AFAICT, we "remain `at_end_of_macro_p`").

> +     Note that this is similar (in observable behavior) to a simpler
> +     implementation of keyboard macros in which this function simply
> +     pushed all characters of the macro into the incoming event queue
> +     and returned immediately.  Maybe this is the implementation that
> +     we ideally would like to have, but switching to it will require
> +     a larger code change.  */

It might be worth mentioning that the main difference is the
availability of `executing-kbd-macro` to let ELisp code behave
differently when called via a kmacro than via "live input".
Which also kind of justifies why `read-key-sequence` wants to
detect the end: if a kmacro ends in the middle of a key sequence, then
it's triggered both my kmacro and by live input.
[ Of course, we could handle it in the command loop instead:
  check and compare the set of pending kmacro events before and after we
  call `read-key-sequence`.  ]


        Stefan

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.