bug#24394: 25.1.50; (find-file "/sudo::") ignores async-shell-command-buffer settings

[Michael Albinus]

Tino Calancha <tino.calancha@gmail.com> writes:

Hi Tino,

> emacs -Q -eval "(setq async-shell-command-buffer 'new-buffer)"
>
> I) M-&: i=0; while true; do echo $i; i=$((i+1)); sleep 1;done RET
> ;; Repeating the command will not prompt for confirmation
> II) M-&: M-p RET ; create buffer "*Async Shell Command*<2>".
> III) M-:(let ((dir (expand-file-name default-directory)))
>           (find-file (concat "/sudo::") dir)) RET
> IV) M-&: M-p RET
> ;; Received prompt to kill buffer "*Async Shell Command*".
>
> In IV) the machine is the same as before and the original user
> still is the owner of the Emacs session: i would expect IV) create
> the buffer "*Async Shell Command*<3>" without any prompt.
> Even if i set `async-shell-command-buffer' to 'new-buffer in
> /root/.emacs i receive the prompt.

You are right, Tramp's handler for `shell-command' ignores
`async-shell-command-buffer'. The reason for this is the following
comment in `tramp-handle-shell-command':

--8<---------------cut here---------------start------------->8---
    ;; Check whether there is another process running.  Tramp does not
    ;; support 2 (asynchronous) processes in parallel.
--8<---------------cut here---------------end--------------->8---

Well, this comment is more than 8 years old, and it is not true anymore
(I've just tested). I don't remember when this was fixed, but so what ...

However, I'm kind of reluctant to fix this in
`tramp-handle-shell-command'. The respective code in `shell-command'
spans over ~40 lines, and I don't believe Tramp shall simply copy those
lines (and other details not handled in Tramp yet). It's even
questionable that Tramp shall offer an own handler for `shell-command'.

The reason why Tramp does this is the use of `shell-file-name' and
`shell-command-switch'. They keep host local values, for remote
connections other values are needed. It is a long standing request, that
Tramp shall offer connection local variables, which carry different
values for different remote hosts. If we would have such a mechanism,
`shell-command' could use `process-file' and `start-file-process', and it
would not need to call a file name handler anymore.

And this error would go away.

Comments?

Best regards, Michael.