bug#38785: 26.3; `y-or-n-p' leaves prompt and response in echo area

bug#38785: 26.3; `y-or-n-p' leaves prompt and response in echo area

[Drew Adams]

`y-or-n-p' puts its prompt in the echo area.

[Aside: That is not ideal for a prompt.  Some other interaction might be
better, such as using a pop-up for the prompt.  In any case, `y-or-n-p'
and similar functions should NOT use the minibuffer.  `read-key' should
continue to act without any use of the minibuffer.  IMHO, it would make
zero sense to use the minibuffer to read a key.]

At the end, `y-or-n-p' puts its prompt in the echo area again, followed
by the user's response (`y' or `n'):

 (unless noninteractive (message "%s%c" prompt (if ret ?y ?n)))

This is a bother, and it can confuse users, especially when there are
additional questions that follow the `y-or-n-p' prompting.

For example, consider a `y-or-n-p' followed by a `map-y-or-n-p'.  When
the latter is finished, the prompt from the preceding `y-or-n-p' gets
restored to the echo area.  That makes little sense, and it can be quite
confusing.

`map-y-or-n-p' does not leave any of its prompts in the echo area.  This
is a good thing, not a bad thing.  Instead, as a good citizen, it does
this at the end, to ensure that it hasn't left any of its echo-area
prompts behind, as litter:

 ;; Clear the last prompt from the minibuffer, and restore the
 ;; previous echo-area message, if any.
 (let ((message-log-max nil))
   (if msg (message "%s" msg) (message "")))

Perhaps that, or similar, is what `y-or-n-p' should do.

Presumably the reason that `y-or-n-p' does what it does is to provide
feedback of the char (`y' or `n') that you typed.  It's true that that's
helpful, but it also causes problems (see above).

Perhaps `y-or-n-p' should do what it does now, but then reset the echo
area after a brief delay (e.g. 1 sec), where "reset" means to do what
`map-y-or-n-p' does: restore any previous echo-area content.  IOW, it's
not bad to echo your response to the `y-or-n-p' question.  But it is bad
not to clean up afterward, leaving the echo area littered with the
prompt.

For example, have `y-or-n-p' bind `msg' to `(current-message)' at the
outset, and then do this at the end:

 (let ((ret (eq answer 'act)))
       (unless noninteractive
         (message "%s%c" prompt (if ret ?y ?n))         ; <==== (1)
         (run-with-timer 1 nil
                         (lambda () 
                           (let ((message-log-max nil)) ; <==== (2)
                             (if msg 
                                 (message "%s" msg)
                               (message ""))))))
       ret)

(1) is what `y-or-n-p' does now, to show you your response.
(2) is what `map-y-or-n-p' does, to clean up after itself.

If something similar to this 1-2 punch is not done, and if the only
available choice is for `y-or-n-p' to do what it does now (#1), or
to just clean up after itself (#2), then my vote is for #2.

In GNU Emacs 26.3 (build 1, x86_64-w64-mingw32)
 of 2019-08-29
Repository revision: 96dd0196c28bc36779584e47fffcca433c9309cd
Windowing system distributor `Microsoft Corp.', version 10.0.17763
Configured using:
 `configure --without-dbus --host=x86_64-w64-mingw32
 --without-compress-install 'CFLAGS=-O2 -static -g3''

bug#38785: 26.3; `y-or-n-p' leaves prompt and response in echo area

[Drew Adams]

> Perhaps `y-or-n-p' should do what it does now, but then reset the echo
> area after a brief delay (e.g. 1 sec), where "reset" means to do what
> `map-y-or-n-p' does: restore any previous echo-area content.  IOW, it's
> not bad to echo your response to the `y-or-n-p' question.  But it is
> bad not to clean up afterward, leaving the echo area littered with the
> prompt.

Actually, that's no good either.  (Damn timers!)

A function such as `y-or-n-p' or `map-y-or-n-p' can't
clean up after a delay, because that will likely wipe
out a subsequent such prompt, by restoring what was
in the echo area before the first one.

E.g. `y-or-n-p' followed by `y-or-n-p' or `map-y-or-n-p'.
The second prompting function would have its prompt
wiped out by the first one restoring what was in the
echo area before the first one started.

I think `y-or-n-p' should probably just use `(sit-for 1)'
after it shows the prompt + answer, and then do what
`map-y-or-n-p' does: clean up after itself:

 (let ((ret  (eq answer 'act)))
   (unless noninteractive
     (message "%s%c" prompt (if ret ?y ?n))
     (sit-for 1)
     (let ((message-log-max  nil))
       (if msg  (message "%s" msg) (message ""))))
   ret)))

bug#38785: 26.3; `y-or-n-p' leaves prompt and response in echo area

[Andreas Schwab]

On Dez 28 2019, Drew Adams wrote:

> I think `y-or-n-p' should probably just use `(sit-for 1)'
> after it shows the prompt + answer,

That would always delay the caller of y-or-n-p by one second.  I'm
pretty sure that would be highly annoying.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

bug#38785: 26.3; `y-or-n-p' leaves prompt and response in echo area

[Eli Zaretskii]

> From: Andreas Schwab <schwab@linux-m68k.org>
> Date: Sun, 29 Dec 2019 09:58:24 +0100
> Cc: 38785@debbugs.gnu.org
> 
> On Dez 28 2019, Drew Adams wrote:
> 
> > I think `y-or-n-p' should probably just use `(sit-for 1)'
> > after it shows the prompt + answer,
> 
> That would always delay the caller of y-or-n-p by one second.  I'm
> pretty sure that would be highly annoying.

Yes.  And besides, this area underwent significant changes since Emacs
26.3, so I suggest to try the latest emacs-27 branch before continuing
the discussion.

bug#38785: 26.3; `y-or-n-p' leaves prompt and response in echo area

[Drew Adams]

> > I think `y-or-n-p' should probably just use
> > `(sit-for 1)' after it shows the prompt + answer,
> 
> That would always delay the caller of y-or-n-p by one
> second.  I'm pretty sure that would be highly annoying.

1. Not always delay.  It's sit-for, not sleep-for,
   after all.  Have you tried it?  I haven't noticed
   a problem with it so far, but there are no doubt
   lots of possible use cases.

2. More annoying than the problem it aims to solve?

3. Alternative solutions are welcome.

The basic problem stems from `read-key' prompting
in the echo area.  That's really not ideal, at
least in some contexts.  (But making it read using
the minibuffer would be an even bigger mistake,
IMO.)

bug#38785: 26.3; `y-or-n-p' leaves prompt and response in echo area

[Andreas Schwab]

On Dez 29 2019, Drew Adams wrote:

>> > I think `y-or-n-p' should probably just use
>> > `(sit-for 1)' after it shows the prompt + answer,
>> 
>> That would always delay the caller of y-or-n-p by one
>> second.  I'm pretty sure that would be highly annoying.
>
> 1. Not always delay.  It's sit-for, not sleep-for,

That doesn't make it any better.  It introduces latency, and latency is
very annoying, prone to induce errors.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

bug#38785: 26.3; `y-or-n-p' leaves prompt and response in echo area

[Drew Adams]

> > > I think `y-or-n-p' should probably just use `(sit-for 1)'
> > > after it shows the prompt + answer,
> >
> > That would always delay the caller of y-or-n-p by one second.  I'm
> > pretty sure that would be highly annoying.
> 
> Yes.  And besides, this area underwent significant changes since Emacs
> 26.3, so I suggest to try the latest emacs-27 branch before continuing
> the discussion.

`read-key', and functions like `y-or-n-p'
that use it or do the same thing, should not
be made to read from the minibuffer.

The fact that such functions have prompted
using the echo area is bad enough.  Making
them read from the minibuffer is even worse.

That's my opinion about (my understanding of)
those "significant changes".  Doing that just
introduces a new (design) bug that should,
and will hopefully someday, be fixed.