bug#71337: 30.0.50; `electric-pair-mode' and custom keybinding

[Eli Zaretskii <eliz@gnu.org>]

On June 4, 2024 11:08:51 AM GMT+03:00, "João Távora" <joaotavora@gmail.com> wrote:
> On Tue, Jun 4, 2024 at 8:37 AM João Távora <joaotavora@gmail.com> wrote:
> 
> I've had a look.  it looks like the problems is e-p-mode's assumption
> that last-command-event is the thing to be inserted.  The fact that
> it isn't here (somehow an innocent 92 is now a monstrous 8388643),
> triggers the problem (8388643 isn't a representation of a character,
> apparently).
> 
> But according to the docstring of post-self-insert-hook,
> the assumption seems sane, and I probably coded for it.
> 
>   post-self-insert-hook is a variable defined in `src/cmds.c'.
> 
>   ...
> 
>   The hook can access the inserted character via `last-command-event'.
>   ...
> 
> I don't think the patch is fully correct.  I think Stefan is the right
> person to call here.  I've had an even briefer look at cmds.c and I
> don't understand how that hook's promise is honoured.


It looks like Arash made the mistake of being the first one, ever, of invoking self-insert-command from Lisp with 2nd arg non-nil, and turning on electric-pair-mode on top of that.  When self-insert-command is called with 2 args, it uses the 2nd arg as the character to insert, but it does NOT overwrite last-command-event with that character.  So post-self-insert-hook sees the wrong event and rightfully barfs.

Which means the patch proposed by Stephen is not TRT, because it means electric-pair-mode will ignore the inserted backslashes.

I think internal_self_insert should overwrite last-command-event or something, if we want to support this kind of scenario.  Stefan, WDYT?