--no-desktop broken?

--no-desktop broken?

[Alfred M. Szmidt]

According to --help, --no-desktop should be a valid option to emacs.
But when one does, `emacs --no-desktop', it errors out and complains
that it is a unknown option.

Happy hacking (sorry for not supplying a patch).

Re: --no-desktop broken?

[Eli Zaretskii]

> From: "Alfred M\. Szmidt" <ams@gnu.org>
> Date: Sat, 21 Jan 2006 01:48:41 +0100
> 
> According to --help, --no-desktop should be a valid option to emacs.
> But when one does, `emacs --no-desktop', it errors out and complains
> that it is a unknown option.

Thanks for reporting this.

It turns out that --no-desktop would only be processed if desktop.el
was loaded in ~/.emacs or in some other init file.  I fixed this for
the case, like yours, when desktop.el was not loaded.  Please see that
the error is gone now.

Re: --no-desktop broken?

[Lars Hansen]

Alfred M. Szmidt wrote:

>According to --help, --no-desktop should be a valid option to emacs.
>But when one does, `emacs --no-desktop', it errors out and complains
>that it is a unknown option.
>  
>
The --no-desktop option is handled in desktop.el, so if desktop.el is
not loaded, the option is unknown.
If you put
   (desktop-save-mode 1)
in your .emacs, you get no error.

I agree that it is confusing that --no-desktop is documented in --help
but seems to not work. Maybe --help should not list options that are not
loaded by default.

Is there some standard or policy here?

Re: --no-desktop broken?

[Ken Raeburn]

On Jan 21, 2006, at 04:38, Lars Hansen wrote:
> I agree that it is confusing that --no-desktop is documented in --help
> but seems to not work. Maybe --help should not list options that  
> are not
> loaded by default.

In the long term (i.e., maybe too much to change right now), perhaps  
we could have arbitrary command-line options handled via Lisp code,  
with a table or something indicating the file to be loaded or  
function to be invoked to process the option, and some mechanism akin  
to loaddefs.el for collecting the options and their help messages  
from various Lisp files and making them available to be printed out  
if --help is given.  Aside from actually invoking some Lisp code when  
--help is given, I think this could mostly be done in new Lisp code.

One approach, off the top of my head, might be to designate symbols  
named "command-line-arg:foo" as handlers for command-line argument "-- 
foo", "--foo=...", and "--no-foo" (pass the function t, "...", or nil  
respectively); then the existing loaddefs would have all the info we  
need to make the list of options.  If the startup code didn't  
recognize an option but the corresponding symbol was defined with a  
function value, that function object would be invoked, and would  
perhaps have the ability to gobble up additional arguments that might  
be needed for the option.  Some new Lisp function invoked via --help  
would walk the obarray atoms looking for such name patterns, make  
sure the autoloads were performed, and then look for 'command-line- 
arg-help-string in the property list of each symbol.  It could even  
load the user's .emacs after displaying the standard Emacs packages'  
options, and see if changes to the load path or autoloads provided  
any additional options, and display them as well.

That wouldn't handle pattern-based argument process, though, like  
turning "--disable-rmail" into "(put 'rmail 'disabled t)" or short "- 
x" style names, or other random things people might want to do, but  
it's a starting point.

Just a thought... maybe this is too much effort to put into the Emacs  
command line... :-)

Ken

Re: --no-desktop broken?

[Lars Hansen]

Ken Raeburn wrote:

> In the long term (i.e., maybe too much to change right now), perhaps 
> we could have arbitrary command-line options handled via Lisp code, 
> with a table or something indicating the file to be loaded or 
> function to be invoked to process the option, and some mechanism akin 
> to loaddefs.el for collecting the options and their help messages 
> from various Lisp files and making them available to be printed out 
> if --help is given.  Aside from actually invoking some Lisp code when 
> --help is given, I think this could mostly be done in new Lisp code.  

I agree that some kind of table approach may be considered at some later
time. But not now ;-)

Re: --no-desktop broken?

[Juri Linkov]

>> In the long term (i.e., maybe too much to change right now), perhaps
>> we could have arbitrary command-line options handled via Lisp code,
>> with a table or something indicating the file to be loaded or
>> function to be invoked to process the option, and some mechanism akin
>> to loaddefs.el for collecting the options and their help messages
>> from various Lisp files and making them available to be printed out
>> if --help is given.  Aside from actually invoking some Lisp code when
>> --help is given, I think this could mostly be done in new Lisp code.
>
> I agree that some kind of table approach may be considered at some later
> time. But not now ;-)

There is already such table approach implemented in Emacs long ago,
and I think desktop.el should use it instead of handling --no-desktop
in startup.el.

There is a special variable `command-switch-alist' which is intended
exactly for this case, i.e. you could add to desktop.el:

;;;###autoload
(add-hook 'command-switch-alist
          '("--no-desktop" . (lambda (argi) (load "desktop"))))

However, loading desktop.el is not a good solution.  Perhaps better would be:

;;;###autoload
(add-hook 'command-switch-alist
          '("--no-desktop" .
            (lambda (argi)
             (message "\"--no-desktop\" ignored because the Desktop package is not loaded"))))

Still not ideal.  Another solution is to add an autoload cookie before
`add-hook' adding a lambda to `after-init-hook' in desktop.el:

;;;###autoload
(add-hook
  'after-init-hook
  '(lambda ()
    (let ((key "--no-desktop"))
      (when (member key command-line-args)
        (setq command-line-args (delete key command-line-args))
        (setq desktop-save-mode nil)))
    (when desktop-save-mode (desktop-read))))

This would work as well.  Or maybe the lambda from `after-init-hook'
should be duplicated in command-switch-alist?

--
Juri Linkov
http://www.jurta.org/emacs/

Re: --no-desktop broken?

[Eli Zaretskii]

> From: Juri Linkov <juri@jurta.org>
> Date: Mon, 23 Jan 2006 03:51:16 +0200
> Cc: ams@gnu.org, raeburn@raeburn.org, emacs-devel@gnu.org
> 
> There is already such table approach implemented in Emacs long ago,
> and I think desktop.el should use it instead of handling --no-desktop
> in startup.el.

??? Did you look at what desktop.el does with --no-desktop?  It has
this code:

  ;; We cannot use `command-switch-alist' to process "--no-desktop" because these
  ;; functions are processed after `after-init-hook'.
  (add-hook
    'after-init-hook
    '(lambda ()
      (let ((key "--no-desktop"))
	(when (member key command-line-args)
	  (setq command-line-args (delete key command-line-args))
	  (setq desktop-save-mode nil)))
      (when desktop-save-mode (desktop-read))))

> There is a special variable `command-switch-alist' which is intended
> exactly for this case

The above comment explains why this is not a good idea.

> ;;;###autoload
> (add-hook 'command-switch-alist
>           '("--no-desktop" . (lambda (argi) (load "desktop"))))
> 
> However, loading desktop.el is not a good solution.  Perhaps better would be:
> 
> ;;;###autoload
> (add-hook 'command-switch-alist
>           '("--no-desktop" .
>             (lambda (argi)
>              (message "\"--no-desktop\" ignored because the Desktop package is not loaded"))))
> 
> Still not ideal.  Another solution is to add an autoload cookie before
> `add-hook' adding a lambda to `after-init-hook' in desktop.el:
> 
> ;;;###autoload
> (add-hook
>   'after-init-hook
>   '(lambda ()
>     (let ((key "--no-desktop"))
>       (when (member key command-line-args)
>         (setq command-line-args (delete key command-line-args))
>         (setq desktop-save-mode nil)))
>     (when desktop-save-mode (desktop-read))))
> 
> This would work as well.  Or maybe the lambda from `after-init-hook'
> should be duplicated in command-switch-alist?

All of these solutions will load desktop.el (and what's worse, load it
unconditionally, I think), which I think is not a good design.  What
startup.el does now is simply ignore the "--no-desktop" switch (we
could remove the informative message if people think it's too
annoying).  But it avoids loading desktop.el, which I think is the
right way of handling such cases.

Re: --no-desktop broken?

[Juri Linkov]

>   ;; We cannot use `command-switch-alist' to process "--no-desktop" because these
>   ;; functions are processed after `after-init-hook'.
>
> The above comment explains why this is not a good idea.

Yes, I'm already aware of this comment.  That's why I haven't proposed to use
`command-switch-alist' directly to process "--no-desktop".

> All of these solutions will load desktop.el (and what's worse, load it
> unconditionally, I think), which I think is not a good design.

Only the first solution loads desktop.el, and I agree that it is not
a good solution.  The remaining two solutions don't load desktop.el.
They use the autoload cookie.  This means that at the time startup.el
starts processing command switches, it already has "--no-desktop"
in `command-switch-alist' without loading desktop.el.  The effect is
exactly the same as adding "--no-desktop" to the local variable
`longopts' in `command-line-1'.  I think not using package-specific
switches in startup.el gives better modularity.

-- 
Juri Linkov
http://www.jurta.org/emacs/

Re: --no-desktop broken?

[Lars Hansen]

[-- Attachment #1: Type: text/plain, Size: 782 bytes --]


>Only the first solution loads desktop.el, and I agree that it is not
>a good solution.  The remaining two solutions don't load desktop.el.
>They use the autoload cookie.  This means that at the time startup.el
>starts processing command switches, it already has "--no-desktop"
>in `command-switch-alist' without loading desktop.el.  The effect is
>exactly the same as adding "--no-desktop" to the local variable
>`longopts' in `command-line-1'.  I think not using package-specific
>switches in startup.el gives better modularity.
>
I like your last solution. It works right and has better modularity.
Also I think the '"--no-desktop" ignored' message is more confusing than
helpful.
So I suggest the attached patch. If there are no objections, I will
install in a couple of days.

[-- Attachment #2: patch.diff --]
[-- Type: text/x-patch, Size: 2072 bytes --]

*** /home/lh/cvsroot/emacs/lisp/desktop.el.~1.97.~	2006-01-23 08:13:25.000000000 +0100
--- /home/lh/cvsroot/emacs/lisp/desktop.el	2006-01-23 08:28:12.766853621 +0100
***************
*** 1156,1161 ****
--- 1156,1162 ----
  ;; after finishing loading the init file.
  ;; We cannot use `command-switch-alist' to process "--no-desktop" because these
  ;; functions are processed after `after-init-hook'.
+ ;;;###autoload
  (add-hook
    'after-init-hook
    '(lambda ()
*** /home/lh/cvsroot/emacs/lisp/startup.el.orig	2006-01-23 08:13:25.714942861 +0100
--- /home/lh/cvsroot/emacs/lisp/startup.el	2006-01-23 08:23:20.391560955 +0100
***************
*** 1632,1638 ****
            (longopts
             (append '(("--funcall") ("--load") ("--insert") ("--kill")
                       ("--directory") ("--eval") ("--execute") ("--no-splash")
!                      ("--find-file") ("--visit") ("--file") ("--no-desktop"))
                     (mapcar (lambda (elt)
                               (list (concat "-" (car elt))))
                             command-switch-alist)))
--- 1632,1638 ----
            (longopts
             (append '(("--funcall") ("--load") ("--insert") ("--kill")
                       ("--directory") ("--eval") ("--execute") ("--no-splash")
!                      ("--find-file") ("--visit") ("--file"))
                     (mapcar (lambda (elt)
                               (list (concat "-" (car elt))))
                             command-switch-alist)))
***************
*** 1732,1744 ****
                  ((equal argi "-kill")
                   (kill-emacs t))
  
- 		;; This is for when they use --no-desktop with -q, or
- 		;; don't load Desktop in their .emacs.  If desktop.el
- 		;; _is_ loaded, it will handle this switch, and we
- 		;; won't see it by the time we get here.
- 		((equal argi "-no-desktop")
- 		 (message "\"--no-desktop\" ignored because the Desktop package is not loaded"))
- 
                  ((string-match "^\\+[0-9]+\\'" argi)
                   (setq line (string-to-number argi)))
  
--- 1732,1737 ----

[-- Attachment #3: Type: text/plain, Size: 142 bytes --]

_______________________________________________
Emacs-devel mailing list
Emacs-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/emacs-devel

Re: --no-desktop broken?

[Eli Zaretskii]

> Date: Mon, 23 Jan 2006 09:24:08 +0100
> From: Lars Hansen <larsh@soem.dk>
> CC: Eli Zaretskii <eliz@gnu.org>,  ams@gnu.org,  raeburn@raeburn.org, 
>  emacs-devel@gnu.org
> 
> Also I think the '"--no-desktop" ignored' message is more confusing than
> helpful.

You are entitled to your opinion, but before you take the message out,
let me first explain why I put it there: it's because if a user says,
e.g., "emacs -q --no-desktop", she doesn't really understand what
"--no-desktop" does.  So I thought it would be good to point out that
the option is not supposed to do anything in that situation.  If we
silently ignore the option, the user might become bewildered as to why
Emacs didn't do whatever she thought it should do in this case.

Re: --no-desktop broken?

[Eli Zaretskii]

> From: Juri Linkov <juri@jurta.org>
> Date: Mon, 23 Jan 2006 08:18:31 +0200
> Cc: larsh@soem.dk, ams@gnu.org, raeburn@raeburn.org, emacs-devel@gnu.org
> 
> Only the first solution loads desktop.el, and I agree that it is not
> a good solution.  The remaining two solutions don't load desktop.el.
> They use the autoload cookie.

Are you sure?  I didn't have time to try your proposal, but the hook
it uses references 2 desktop.el symbols which are both autoloaded from
desktop.el.

Even if today it won't load desktop.el at startup, some future change
in that hook might cause it to be loaded, especially if we don't even
say in a comment there that the hook is supposed to ignore
"--no-desktop" without loading the package.

Re: --no-desktop broken?

[Lars Hansen]

>If we
>silently ignore the option, the user might become bewildered as to why
>Emacs didn't do whatever she thought it should do in this case.
>  
>
Well, I beleive that the message might make the user think that there
was some kind of problem; that Emacs was not abel to do what the
--no-desktop is supposed to do. And that's not the case.

But maybe it is a matter of taste.

Re: --no-desktop broken?

[Eli Zaretskii]

> Date: Tue, 24 Jan 2006 00:43:41 +0100
> From: Lars Hansen <larsh@soem.dk>
> CC: juri@jurta.org,  ams@gnu.org,  raeburn@raeburn.org, 
>  emacs-devel@gnu.org
> 
> Well, I beleive that the message might make the user think that there
> was some kind of problem

There _was_ a problem: the user typed --no-desktop in a situation
where it makes no sense.

Re: --no-desktop broken?

[Lars Hansen]

>There _was_ a problem: the user typed --no-desktop in a situation
>where it makes no sense.
>  
>
I don't see it as a problem to turn something off that was already
turned off.
But if we give a message in such cases, we should be consistent and also
give a message when desktop _is_ loaded but desktop-save-mode is turned off.