bug#12972: [PATCH] Avoid regression in mailcap-view-file similar to Bug#44824

[Maxim Nikulin <manikulin@gmail.com>]

On 03/07/2021 18:56, Eli Zaretskii wrote:
>> From: Maxim Nikulin Date: Sat, 3 Jul 2021 18:29:30 +0700
>>
>> I am giving up with this issue.
> 
> That's too bad.  I see no reason to give up, and I urge you to
> reconsider, please.

Sorry, but the space of your assumptions and maybe confusions has too 
high number of dimension to realize what you actually mean and what you 
consider as a problem that should be fixed. So any further steps are 
impossible.

> Your patch unconditionally assumes that every handler will immediately
> exit, and that it doesn't care about the connection type with the
> parent Emacs process, but that is not guaranteed to be true.

There is no intention of such assumption and it *works* even for 
handlers that does not exit immediately.

I admit that I wrongly added ":noquery t", for some reason I believed 
that it allows to choose whether processes are allowed to exist longer 
than emacs or it is preferred to kill them with emacs. Actually 
asynchronous processes are killed always and the option manages the 
query only. This option should be dropped to restore compatibility with 
previous variant.

I have not found a way to detach asynchronous process from emacs. 
Surprisingly it is possible for synchronous processes but it is 
impossible to detect failure (thus to allow a user to figure out what 
has happened)

     (process-file-shell-command command nil 0 nil)

So process API in emacs is a kind of a short blanket.

Accidentally I have created an example of program that is incompatible 
with 'pipe asynchronous processes in emacs

#!/bin/sh
exec 1>&-
exec 2>&-
sleep 30

(let ((command "cpu-stress-test")
       (process-connection-type nil))
   (start-process-shell-command command nil command))

And emacs (at least 26.3) consumes 100% CPU for the specified amount of 
time. I do not see any reason to do so since the program does not do 
anything ugly. I have not found a way to explicitly force emacs to close 
pipes. That is why I consider it as an outstanding bug. Emacs must 
properly handle closed pipes.

So `process-file-shell-command' ... 0 is better than current 
`start-process-shell-command' but it does not allow to add error handling.

So besides that I still have no guess what problem you suspect, now I 
know that emacs may become mad in response to purely innocent action of 
a child process.