Re: custom-set-variables fails to set variable

[Luc Teirlinck]

   I'm still a bit confused though. For example, mh-show-mode-hook is
   defined implicitly by the definition of mh-show-mode in mh-utils.el
   which requires mh-customize.el which contains the defcustom of
   mh-show-mode-hook. Shouldn't that defcustom have won against the
   defvar in the mh-show-mode?

I should look at the details to figure that out.  But do not forget
that add-hook also will bind an undefined hook, just as defvar does.
Custom and hooks quite simply do not work very well together.  Per
designed Custom under the assumption that there are two disjoint
classes of variables: internal variables for use by programs and
Custom variables for user customization which programs should never
touch.  Emacs does not satisfy that assumption and this creates
problems.  These problems are the worst for hooks, which should be
manipulated with add-hook and remove-hook, not with setq-default, as
Custom does.  I described several nasty problems related to Custom and
hooks earlier.  This is a yet another one.  We are planning for a fix,
but decided it was too involved to do before the release.

(add-hook 'foo-hook 'essential-function) custom-set-variables:

The custom-set-variables undoes the add-hook.

This is a problem I described earlier.  It could create problems if a
user upgrades to an Emacs version in which essential-function is new
and essential.  It is a constant potential pitfall for CVS users.

But now:

custom-set-variables (add-hook 'foo-hook 'essential-function) defcustom:

Now the add-hook means that the user's customization is ignored.
This is yet another problem related to Custom and hooks.

custom-set-variables defcustom (add-hook 'foo-hook 'essential-function)

will work as intended, but I see no way to always guarantee this order.

   I'd still be curious to hear the opinions on whether one should use
   the ;;;###autoload cookie on the defcustoms or not.

It could solve the newly pointed out problem with Custom and hooks,
but would not solve the other ones, like, for instance, the first one
I pointed out above (there are others).  Some people who do not have a
lot of memory available might not like it.

Sincerely,

Luc.