On 09.09.2021 06:07, Richard Stallman wrote:
> [[[ To any NSA and FBI agents reading my email: please consider    ]]]
> [[[ whether defending the US Constitution against all enemies,     ]]]
> [[[ foreign or domestic, requires you to follow Snowden's example. ]]]
> 
>    > Others replied later that users generally expect 'redo' to require one
>    > more modifier than 'undo'. In graphical Emacs, it can work out to having
>    > 'C-/' call 'undo', then one can additionally hold down the Shift key to
>    > the right of '/', calling 'redo' with the resulting 'C-?'.
> 
> If users are happy with that, I won't argue against it.  It doesn't have
> any other problem.
> 
> We still have the problem of ttys.  We can't use C-/ or C-? on a tty
> because each of them is the same character as DEL.  So we use C-_ instead
> for undo.
> 
> Because there is no difference on a tty between C-_ and C--.
> there is no way to disguinuish between Sh-C-- and C--.

I previously suggested M-_ (working fine on my tty), but it does make 
the transition more awkward.

> However, C-M-- could be used for redo on ttys.
> 
> Does that make everyone happy?  If so, maybe we're ready to
> introduce that binding.

That's pretty elegant. On tty, the user would press C-- to undo and 
C-M-- to redo. One-modifier difference.

Though in my testing I had to bind C-M-_ instead. That's what my tty 
Emacs translated the combination to. And that also resolved the conflict 
with the existing 'C-M--' binding for 'negative-argument'.

> Initially we should continue the current behavior of undo
> to avoid an incompatible change, and have an option to enablke
> the undo-redo behavior.  Once Emacs 28 is out, we could find out
> what the users think and whether we old fogeys can adjust to undo-redo.
> Then we could decide whether to make that the default.

Sounds good.

I wonder if we should add some mechanism to detect that the user had 
called undo-redo in the current session, and then one of his/her undo-s 
is going to perform a "redoing" undo (and suggest to disable the 
behavior, by changing some user option that will turn their 'undo' into 
'undo-only').

Or maybe the attached patch is enough: long-time users will use 'undo' 
like they always did, and the undo-redo users aren't generally going to 
notice that 'undo' sometimes goes too far.

We should also modify cua-mode to use undo-redo, but that can go in after.