>>>>> In <vzpth7k7iz.fsf@redqueen.bytechase.cx>
>>>>>	Norbert Koch <viteno@xemacs.org> wrote:

> Here we go.  Must have had the fixed version in the path the last time
> around.  Is this any better?

> Thanks,
> norbert.

That's perfect.  Thank you.

> Signaling: (void-variable to-address)
>   (message-wide-reply to-address)

It happened in the " *gnus article copy*" buffer.  Just before
it, to-address was surely bound by `let*' in the summary buffer.

[...]
>   (let* ((group ...) (charset ...) (pgroup group) to-address ...
[...]
>   (gnus-setup-message (cond (yank ...) (article-buffer ...) (...
[...]
>   gnus-post-news(nil "nnml:lfnet" [9709 "exim-filter in Abhae...
[...]
>   gnus-summary-followup((9709) nil)
>   gnus-summary-followup-with-original(nil)
>   call-interactively(gnus-summary-followup-with-original)

I have only one idea for the reason it occurs.  It is that the
to-address buffer-local variable exists only in the summary
buffer.  I don't know which function puts it there.  By
evaluating the following form in the summary buffer when an
error has occurred, you can see if it is true or not:

(local-variable-p 'to-address (current-buffer))

If it returns t, it may be better to grep all your files (except
for the Gnus sources) for "to-address". :p

Though it is just a workaround, here is a patch for the problem: