bug#13697: Reopening

[Eli Zaretskii]

> From: Trey Ethan Harris <treyharris@gmail.com>
> Date: Wed, 3 Apr 2019 17:01:38 -0400
> 
> Many people seem to be attempting to run `emacs --daemon` as part of their 
> system on login startup--mostly
> as a way to improve emacsclient initialization time when Emacs is first 
> needed--but they run into trouble
> because systemd or other startup system or the GUI login process is 
> non-interactive. If Emacs attempts to
> interact in any way at startup time, then the daemon startup hangs.
> 
> There is no way AFAIK to respond to such an interaction--the Emacs server 
> hasn't started yet, so
> emacsclient can't be used.
> 
> The issue can exacerbate system issues, especially after abnormal exit. If 
> the user uses the empty-string form
> of ALTERATE_EDITOR, they may unwittingly start up multiple Emacs daemons that 
> will all hang at the same
> point in startup. If they so configure their startup system with a watchdog 
> functionality, they may repeatedly kill
> and restart the Emacs daemon, doing nothing but drawing useless CPU and IO.
> 
> One can workaround the issue in a piecemeal fashion by overriding or advising 
> functions in startup packages
> that may interact (for instance, advising `desktop-restore-file-buffer` so 
> that it will not block if the daemon is
> being restarted from a crash and a desktop lockfile still exists). However, 
> this is not a general solution--one
> can never know for certain from where an interaction might arise.
> 
> Of course Emacs itself can't supply a 100% solution--the user could always 
> put something in initialization
> which blocks on interaction, even if it's specific Elisp to do so or a call 
> out to a shell program.
> 
> But Emacs _could,_ I think, provide a 99% solution:
> 
> 1. Intercept interaction functions like `yes-or-no-p` and password prompts 
> running in asynchronous threads,
> allowing them to block their thread
> 2. Automatically convert interaction in other threads into a warning failure
> 3. Continue execution of other threads so that the server can start
> 4. Upon first connection with an emacsclient, display the warnings 
> (presumably by popping up *Warnings*) so
> that the user is aware of what may have failed in initialization due to 
> inability to interact
> 5. Take the user through the queued blocked threads so they can complete the 
> threaded interactions.
> 
> I have tried to do this in a pure user-side Elisp way, but I don't think it's 
> possible without Emacs' assistance.

The problem in desktop.el was fixed in Emacs 26.1, and it didn't
require any complications with threads.  What other problems like that
are there?  Please report each such problem as a separate bug, and
please let the Emacs developers decide whether they need a single
unified solution, like the one proposed in this bug report, or
different solutions.  And in any case, just detecting that Emacs
cannot interact does not yet tell what to do in each such case (and
I'm not sure a single solution will fit all).