window-system and batch mode

window-system and batch mode

[Katsumi Yamaoka]

Hi,

Recently Emacs sets `window-system' to t in the batch mode:

emacs -batch -Q -eval '(message "%s" window-system)'
t

Shouldn't it be nil?  If not, what is the right way that a Lisp
program knows that it runs in the batch mode?  Is it `noninteractive',
`(display-graphic-p)' or possibly `initial-window-system'?

I run Gnus in the batch mode using cron every day before I come
to the office for fetching new mails, but it failed nowadays
because Gnus calls `gnus-x-splash' which requires a graphic Emacs
frame[1].  The relevant code is:

(defun gnus-1 (&optional arg dont-connect slave)
[...]
      (cond
       ((featurep 'xemacs)
	(gnus-xmas-splash))
       (window-system
	(gnus-x-splash))))

([1] Although I've fixed in Gnus CVS the real cause, that is the
 form `(frame-parameter nil 'vertical-scroll-bars)' returns nil
 in `gnus-x-splash', it is essentially needless to be called if
 there's no window system.)

Regards,

Re: window-system and batch mode

[Richard Stallman]

    Recently Emacs sets `window-system' to t in the batch mode:

    emacs -batch -Q -eval '(message "%s" window-system)'
    t

    Shouldn't it be nil?

I think that is a bug, and it should be nil.
Would someone please DTRT then ack?

However, the right way to test for batch mode is `noninteractive'.
That is documented in the Lisp Manual.

Re: window-system and batch mode

[Stefan Monnier]

> Recently Emacs sets `window-system' to t in the batch mode:

Yes, when I changed window-system to be terminal-local I had to decide
what value to give to it for the initial "terminal" (which is basically
just stdin/stdout rather than an X11 or tty connection).  I decided to
use t (rather than nil which means "tty") to see what happens and
hopefully catch some bugs.  And indeed it helped me get the
"terminal-local window-system" patch working.

> Shouldn't it be nil?

Yes, we can easily change it back to nil.  I think code which relies on
this is broken, but I don't feel strongly about it.

> If not, what is the right way that a Lisp
> program knows that it runs in the batch mode?  Is it `noninteractive',
> `(display-graphic-p)' or possibly `initial-window-system'?

If we ever get the "frameless" emacs-server working, it will most likely
be using such a terminal, now whether that's interactive or not... the
limit will be a bit more blurry by then.


        Stefan

Re: window-system and batch mode

[Richard Stallman]

    Yes, when I changed window-system to be terminal-local I had to decide
    what value to give to it for the initial "terminal" (which is basically
    just stdin/stdout rather than an X11 or tty connection).  I decided to
    use t (rather than nil which means "tty") to see what happens and
    hopefully catch some bugs.  And indeed it helped me get the
    "terminal-local window-system" patch working.

I am glad it helped catch some bugs, but the correct value for a stdin
is certainly nil.  stdin is not a window system.

Re: window-system and batch mode

[Stefan Monnier]

>     Recently Emacs sets `window-system' to t in the batch mode:
>     emacs -batch -Q -eval '(message "%s" window-system)'
>     t

>     Shouldn't it be nil?

> I think that is a bug, and it should be nil.
> Would someone please DTRT then ack?

Done,


        Stefan

Re: window-system and batch mode

[Katsumi Yamaoka]

>>>>> Stefan Monnier wrote:

>>     Recently Emacs sets `window-system' to t in the batch mode:
>>     emacs -batch -Q -eval '(message "%s" window-system)'
>>     t

>>     Shouldn't it be nil?

>> I think that is a bug, and it should be nil.
>> Would someone please DTRT then ack?

> Done,

Confirmed.  Thanks.

>>>>> Richard Stallman wrote:

> However, the right way to test for batch mode is `noninteractive'.
> That is documented in the Lisp Manual.

I see.  I verified Gnus uses `noninteractive' properly for
`gnus-batch-kill'.   Only `gnus-batch-kill' seems to be a non-
interactive feature that Gnus explicitly supports, so I don't
add something useful for the batch mode to other Gnus functions
for the moment.  FYI, my crontab entry is in the next page.

OTOH, the usages of `window-system' are not all necessarily
proper.  For instance, `mailcap-mime-data' inhibits some X
applications if Emacs runs in tty even if it runs in the local
X window system.  Gnus may assume that nil for `window-system'
means a user uses a teletype in a far place, though.

\f
crontab -l
--8<---------------cut here---------------start------------->8---
15 6 * * * rm -f ~/#.newsrc-dribble# ~/.newsrc-dribble; emacs -batch\
 -u yamaoka -l gnus --eval '(let ((gnus-batch-mode t)) (gnus 1))'\
 -f gnus-group-quit
--8<---------------cut here---------------end--------------->8---

`-u yamaoka' is similar to `-l ~yamaoka/.emacs'.  Binding
`gnus-batch-mode' to t makes Gnus not make query when opening a
primary (nntp, in my case) server fails.  The argument `1'
passed to `gnus' means fetching only new mails (the group levels
of my mail groups are all `1').