On 10/10/2022 1:53 AM, Eli Zaretskii wrote:
> It is okay not to prompt when there's no need.  That wasn't the part
> on which I commented, AFAIU.
> 
> You said:
> 
>> On the other hand, I think it *would* be useful to prompt if you're in
>> the last client, but there are other non-client frames.
> 
> I'm saying that no, it will not be useful to prompt in that case: if
> the only frames left are non-client frames, we should not prompt,
> because we don't prompt when there are no client frames to begin with.
> Apologies if I misunderstood the use case.

Ah, I see. In that part of my original message, I wanted to say that 
this was a case where I thought the old code made sense, since it 
prompts the user then. However, I also thought that the prompt message 
could be more informative.

Here's why I think prompting then makes sense: when you're in an 
emacsclient frame and there are other non-client frames (i.e. ones 
"owned" by the main Emacs process), that looks very similar to the user 
as when you have a second emacsclient running. In other words, I think 
it would be good for this function to prompt when doing this:

   $ emacs -Q --daemon
   $ emacsclient foo.txt
   $ emacsclient -c bar.txt

   ;; From either frame:
   M-x save-buffers-kill-emacs

Or this:

   $ emacs -Q --daemon
   $ emacsclient foo.txt
   $ emacsclient --no-wait -c bar.txt

   ;; From either frame:
   M-x save-buffers-kill-emacs

Or this:

   $ emacs foo.txt
   M-x server-start
   $ emacsclient -c bar.txt

   ;; From either frame:
   M-x save-buffers-kill-emacs

In all of those cases, the user would have 2 (visible) frames, possibly 
on different displays/terminals. Since they're visually similar, I think 
it makes sense to have similar prompts.

Next, if the user ran 'save-buffers-kill-emacs' a second time in any of 
those examples, it *wouldn't* prompt, since then all the remaining 
frames are associated with a single emacsclient (or no client at all).

Since the use case for this is non-obvious (it actually took me several 
iterations to identify all the possible cases), I updated my patch with 
clearer (to me) code, plus comments explaining the reasoning.