bug#72426: 29.2.50; comint-pager doesn't affect async-shell-command

[Spencer Baugh <sbaugh@janestreet.com>]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Spencer Baugh <sbaugh@janestreet.com>
>> Cc: Jim Porter <jporterbugs@gmail.com>,  72426@debbugs.gnu.org
>> Date: Wed, 07 Aug 2024 11:05:47 -0400
>> 
>> Eli Zaretskii <eliz@gnu.org> writes:
>> >> Date: Tue, 6 Aug 2024 19:36:15 -0700
>> >> Cc: 72426@debbugs.gnu.org
>> >> From: Jim Porter <jporterbugs@gmail.com>
>> >> 
>> >> Just to be extra-sure, I tried using 'async-shell-command' to run "git 
>> >> log" with PAGER unset, and sure enough it tried to use "less" for 
>> >> paging, which didn't go very well. "PAGER=cat" was much better.
>> >
>> > Did you try with process-connection-type nil?
>> 
>> With process-connection-type nil, git will never run a pager (since like
>> many programs it checks whether stdout is a terminal before doing so).
>> 
>> So both:
>> 
>> (let ((process-connection-type nil)
>>       (process-environment (cons '("PAGER" "less") process-environment)))
>>   (async-shell-command "git log"))
>> 
>> and
>> 
>> (let ((process-connection-type nil)
>>       (process-environment (cons "PAGER" process-environment)))
>>   (async-shell-command "git log"))
>> 
>> behave identically.
>> 
>> Setting PAGER=cat is only necessary for process-connection-type=t.
>
> As expected.  The problem is that a Lisp program could let-bind this
> variable around a call to async-shell-command (or some other similar
> API), in which case a setting in startup.el will not catch that.  But
> maybe we don't care, since a program whose stdout is not a console
> device will ignore PAGER anyway.

Yep - I think it's fine for the same reason it's fine to have TERM set
by default, even though it's ignored by programs whose stdout is a pipe.