Re: custom-set-variables fails to set variable

Re: custom-set-variables fails to set variable

[Richard M. Stallman]

    Why does define-minor-mode need to defvar the mode hook?  There is no
    need to defvar hooks.  `add-hook' can perfectly handle unbound hook
    variables.  I propose that we revert this change in
    define-derived-mode.

I agree, if we can't find a better solution.

Another possible solution is to autoload the defcustoms
for these particular variables.  Would that work?

A more automatic solution could be: define-derived-mode
could automatically produce an autoload definition
of the defvar.  That would solve the problem, right?
It is not normal for update-file-autoloads to generate
an autoload with no autoload cookie, but we could do that.

Re: custom-set-variables fails to set variable

[Luc Teirlinck]

       Why does define-minor-mode need to defvar the mode hook?  There is no
       need to defvar hooks.  `add-hook' can perfectly handle unbound hook
       variables.  I propose that we revert this change in
       define-derived-mode.

   I agree, if we can't find a better solution.

   Another possible solution is to autoload the defcustoms
   for these particular variables.  Would that work?

Yes, but this is inevitably going to be forgotten for some defcustoms.

   A more automatic solution could be: define-derived-mode
   could automatically produce an autoload definition
   of the defvar.  That would solve the problem, right?

Yes, but both the manual and the automated solutions have the
potential for problems.  Somebody supplying a defcustom may want a
tailor-made docstring for the defcustom, mentioning typical uses for
the hook and the defvar's docstring could overwrite that.  In general,
it does not seem very good to have two competing different definitions
for a variable.  This solution might also unnecessarily bloat loaddefs.

Remember that all of this came about because the mode's docstring
contains:

    Entry to this mode calls the value of `lisp-mode-hook'
    if that value is non-nil.

And `C-h v' on lisp-mode-hook did not work.

Maybe we could just change that to:

    Entry to this mode calls the value of `lisp-mode-hook'
    if that value is bound and non-nil.  `add-hook' automatically
    binds a hook variable if it was previously unbound.

That would also help people understand why `C-h v foo-hook' does not
always work for hooks.  Many hooks that are not mode hooks have no
defvar.

Sincerely,

Luc.

Re: custom-set-variables fails to set variable

[Luc Teirlinck]

>From my earlier message:

   Unless I am confused, this also has the disadvantage that it is not
   going to work for uses of define-minor-mode in third party packages.

I meant `define-derived-mode', not `define-minor-mode', obviously.

Sincerely,

Luc.

Re: custom-set-variables fails to set variable

[Juri Linkov]

> Remember that all of this came about because the mode's docstring
> contains:
>
>     Entry to this mode calls the value of `lisp-mode-hook'
>     if that value is non-nil.
>
> And `C-h v' on lisp-mode-hook did not work.

Please note also that even when `C-h v' works correctly on a hook variable
(like it works now), an attempt to follow the link to the source code
(i.e. typing RET on the file name link in the Help buffer) fails.
That is because `find-function-search-for-symbol' tries to find a
definition of the variable whose name contains the suffix "-hook",
but it is defined in the macro without the suffix "-hook" (e.g.
`mh-letter-mode-hook' is defined as `define-derived-mode mh-letter-mode').

It is possible to specify the name of the definition explicitly by
setting the symbol's property `definition-name'.  `define-global-minor-mode'
already does this, so `define-derived-mode' could do the same.  This
means putting a line like

    (if (not (boundp ',map)) (put ',map 'definition-name ',child))

before every `defvar' in the body of `define-derived-mode'.

Checking the boundness is necessary for the case when the variable is
already defined before evaluating the body of `define-derived-mode'.

There are four variables defvar'ed currently in `define-derived-mode':
`-hook', `-map', `-syntax-table' and `-abbrev-table'.  The condition
above works for the first three variables, but not for `-abbrev-table'
which requires special treatment for the case when an abbrev table
is defined in ~/.abbrev_defs.

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

Re: custom-set-variables fails to set variable

[Luc Teirlinck]

Richard Stallman wrote:

   A more automatic solution could be: define-derived-mode
   could automatically produce an autoload definition
   of the defvar.  That would solve the problem, right?
   It is not normal for update-file-autoloads to generate
   an autoload with no autoload cookie, but we could do that.

Unless I am confused, this also has the disadvantage that it is not
going to work for uses of define-minor-mode in third party packages.

Sincerely,

Luc.

Re: custom-set-variables fails to set variable

[Richard M. Stallman]

       A more automatic solution could be: define-derived-mode
       could automatically produce an autoload definition
       of the defvar.  That would solve the problem, right?
       It is not normal for update-file-autoloads to generate
       an autoload with no autoload cookie, but we could do that.

    Unless I am confused, this also has the disadvantage that it is not
    going to work for uses of define-derived-mode in third party packages.

You are right.  So that approach won't work.

Meanwhile, another point occurs to me.
Why is the defvar that comes from the define-derived-mode
executed before the defcustom?  Could that be solved
by putting the defcustom earlier in the file?

Another idea: replace the defvar with a put call that will put on a
variable-documentation property.

    Yes, but both the manual and the automated solutions have the
    potential for problems.  Somebody supplying a defcustom may want a
    tailor-made docstring for the defcustom, mentioning typical uses for
    the hook and the defvar's docstring could overwrite that.

We don't want the automatically generated doc string to override
an explicit one, so perhaps the define-derived-mode should generate

  (unless (get VAR 'variable-documentation)
    (put VAR 'variable-documentation STRING))

Re: custom-set-variables fails to set variable

[Luc Teirlinck]

Richard Stallman wrote:

   Meanwhile, another point occurs to me.
   Why is the defvar that comes from the define-derived-mode
   executed before the defcustom?  Could that be solved
   by putting the defcustom earlier in the file?

Yes, but with the current code, the defvar's docstring would overwrite
the defcustom's.  Also, I would rather not rely on all Elisp authors
to be perfectly aware of this issue.

   We don't want the automatically generated doc string to override
   an explicit one, so perhaps the define-derived-mode should generate

     (unless (get VAR 'variable-documentation)
       (put VAR 'variable-documentation STRING))

I believe that this is a much more solid alternative.  The patch below
implements this.  I can install if it looks OK.

But remember that the original confusion which let to the defvar was
not that outline-mode-hook had no docs.  I believe that the OP was
perfectly aware of the fact that outline-mode-hook was run at the end
of outline-mode.  What really confused him was the fact that
outline-mode was unbound.  So I propose to expand the docstring to
explain that this is no problem.

After my patch `C-h v paragraph-indent-text-mode-hook' results in:

    paragraph-indent-text-mode-hook is void as a variable.


    Documentation:
    Hook run when entering Parindent mode.
    No problems result if this hook is not bound.
    `add-hook' automatically binds it.  (This is true for all hook variables.)

Patch:

===File ~/derived-diff======================================
*** derived.el	06 Aug 2005 13:23:00 -0500	1.12
--- derived.el	12 Nov 2005 17:31:33 -0600	
***************
*** 194,200 ****
  		     parent child docstring syntax abbrev))
  
      `(progn
!        (defvar ,hook nil ,(format "Hook run when entering %s mode." name))
         (defvar ,map (make-sparse-keymap))
         ,(if declare-syntax
  	    `(defvar ,syntax (make-syntax-table)))
--- 194,205 ----
  		     parent child docstring syntax abbrev))
  
      `(progn
!        (unless (get  ',hook 'variable-documentation)
! 	 (put ',hook 'variable-documentation
! 	      ,(format "Hook run when entering %s mode.
! No problems result if this hook is not bound.
! `add-hook' automatically binds it.  (This is true for all hook variables.)"
! 		       name)))
         (defvar ,map (make-sparse-keymap))
         ,(if declare-syntax
  	    `(defvar ,syntax (make-syntax-table)))
============================================================

Re: custom-set-variables fails to set variable

[Richard M. Stallman]

Please install your patch.