Re: message-sort-headers-1: Symbol's value as variable is void: sort-fold-case

[Katsumi Yamaoka <yamaoka@jpl.org>]

>>>>> In <87hcydkfdl.fsf@baldur.nicundtas.de> Tassilo Horn wrote:

> since a few days I sometimes get the message in the subject when I press
> RET on an article in *Summary*. The *Article* buffer then shows all
> headers without fontification.

> Even a `M-: (require 'sort)' doesn't fix it, although it's defined there
> (with `defcustom'). This seems to happen only if I start Gnus right
> after I started emacs...

> ,----[ C-h v sort-fold-case RET ]
>| sort-fold-case is a variable defined in `sort.el'.
>| It is void as a variable.
> `----

> A `(defvar sort-fold-case nil)' fixes it.

> If I start emacs and do `M-: (require 'sort)' before I start Gnus, it's
> properly set.

> ,----[ C-h v sort-fold-case RET ]
>| sort-fold-case is a variable defined in `sort.el'.
>| Its value is nil
> `----

You seem to load a program in which `sort-fold-case' is bound to
some value by `let' and uses some sort function to be autoloaded.
For instance, I can reproduce a similar error by `emacs -Q' and
the following form:

(with-temp-buffer
  (let ((sort-fold-case t))
    (sort-lines nil (point-min) (point-max)))
  (sort-lines nil (point-min) (point-max)))

The first `sort-lines' autoloads sort.elc which provides the
`sort-fold-case' variable, however the variable disappears after
exiting the `let' form.  After performing this, the feature
`sort' is provided, so the form `(require 'sort)' does nothing.
Such a code should be written like the following so that the
`sort-fold-case' variable may exist before it is bound by `let'.

  (require 'sort)
  (let ((sort-fold-case t))
    (sort-lines nil (point-min) (point-max)))

There is no culprit in the Gnus sources.  I guess you will
probably be able to find it by performing grep.  For example:

grep '(sort-fold-case' ~/.emacs ~/.gnus.el
grep '(sort-fold-case' /.../site-lisp/*.el