bug#71049: async-shell-command ends with "Process *Async Shell Command* finished" when remote "direct-async-process"

[Michael Albinus]

Dmitry Gutov <dmitry@gutov.dev> writes:

Hi Dmitry,

> Previously mentioned in bug#70901.
>
> When a Tramp connection is configured as "direct-async-process",
>
> 1. If the buffer *Async Shell Command* does not exist, invoking M-& in
> a remote buffer from that connection makes it end with
>
>   Process *Async Shell Command* finished
>
> The command can be simple, like 'ls' or 'echo 123'. I also see this
> added to *Messages*:
>
>   Tramp: Inserting
>   ‘/ssh:dgutov@fencepost.gnu.org:/home/d/dgutov/.tramp_history’...done
>
> Which seems odd (tramp history is re-read every time like this, for
> some purpose?).
>
> 2. If the buffer such named already exists when command is invoked,
> then this doesn't happen (the output seems correct). But *Messages*
> says
>
>   -l: finished.
>
> ...which is a big puzzling as well.
>
> Also, scenario 1 (when the buffer doesn't exist yet) takes longer than
> 2, but that might be a side-effect of implementation external to
> Tramp.

Thanks for the bug report, I can reproduce it. Honestly, direct async
processes are a little bit under-implemented (by decision, they should
be fast). I'll check what could be done w/o loosing performance.

According to reading .tramp_history: this is performed in
comint-read-input-ring, called from shell-mode. comint-input-ring-file-name
is set in shell-mode, I don't see a trivial solution to suppress
this. Likely, we must extend shell-mode for this case.

Best regards, Michael.