bug#64394: [PATCH] Fix `async-shell-command-display-buffer' display

[martin rudalics]

>> I'm probably missing something, but how can display-buffer fail to
>> support any action function, such as display-buffer-no-window?
>>
>> Martin, what am I missing here?

We may have to ask Juri, he conceived the "allow-no-window" concept.

> I was also confused.  Based on the documentation for
> `display-buffer-no-window', it seems that callers are supposed to
> explicitly pass an `(allow-no-window . t)' cons pair when calling
> `display-buffer' as a signal that they can correctly handle a return
> value of nil.  If it's absent, `display-buffer-no-window' seems to err
> on the side of caution, assume the caller can't handle nil, displays the
> window anyway, and returns it like all the other display functions.

I think that's the idea, yes.

> Technically it seems that you can add `(allow-no-window . t)' to
> `display-buffer-alist' to always force the buffer never to appear, but
> that doesn't seem at all like its intended use.

Maybe "force" is too strong here.  You can "force" it by adding an
'allow-no-window' entry to the alist _and_ a 'display-buffer-no-window'
action in a position that precedes any other display buffer action.

>> That's fine, but those are your preferences.  I'd feel uncomfortable
>> with forcing them on everyone, if we already have a way of tailoring
>> this behavior by user customizations.
>
> I might not have been clear with what I meant here, sorry; I mean that
> in my own personal config, when I run `mpv', its output appears in a
> buffer named `*shell:mpv*' instead of `*Async Shell Command*', and I
> have an explicit entry for it in `display-buffer-alist' so that it
> doesn't appear via `display-buffer-no-window'.  This was functioning
> well, except the moment I set `async-shell-command-display-buffer' to
> nil, the buffer displayed itself the moment mpv began to write to
> stdout.
>
> A minimally reproable example in `emacs -Q':
>
> --8<---------------cut here---------------start------------->8---
> (setq display-buffer-alist
>        '(("\\*Async Shell Command\\*"
>     (display-buffer-no-window))))
> (setq async-shell-command-display-buffer nil)
> --8<---------------cut here---------------end--------------->8---
>
> `M-& echo foo RET' will unexpectedly show the `*Async Shell Command*'
> buffer.

I suppose (Juri will correct me) that this snippet in 'shell-command'

                (if async-shell-command-display-buffer
                    ;; Display buffer immediately.
                    (display-buffer buffer '(nil (allow-no-window . t))) <<<<<
                  ;; Defer displaying buffer until first process output.
                  ;; Use disposable named advice so that the buffer is
                  ;; displayed at most once per process lifetime.
                  (let ((nonce (make-symbol "nonce")))
                    (add-function :before (process-filter proc)
                                  (lambda (proc _string)
                                    (let ((buf (process-buffer proc)))
                                      (when (buffer-live-p buf)
                                        (remove-function (process-filter proc)
                                                         nonce)
                                        (display-buffer buf)))) <<<<<
                                  `((name . ,nonce)))))))

adding an 'allow-no-window' entry if and only if
'async-shell-command-display-buffer' is non-nil is responsible for the
behavior Eliza sees.  I have no idea whether adding such an entry in the
second case could cause problems.  We could give
'async-shell-command-display-buffer' a third value, say 'allow-no-window
and, if a user has set it to that value, have 'shell-command' add an
'allow-no-window' entry in the second case too.

martin