Geoff writes:

   Hi David,

   I've found the source of the problem I was having. It seems to be due
   to some interference with the notmuch-mua-reply function and
   gnus-alias. In particular, there was a problem with the part of
   notmuch-mua-reply that inserted the newly setup message contents.

   Towards the end of the notmuch-mua-reply function, there is the following code:

       ;; insert the message body - but put it in front of the signature
       ;; if one is present
       (goto-char (point-max))
   *    (if (re-search-backward message-signature-separator nil t)
   *	  (forward-line -1)
	 (goto-char (point-max)))
       (insert body)
       (push-mark))
     (set-buffer-modified-p nil)

   The problem is with the two lines I've marked with a *. Suppose the
   main message reply buffer already contains a signature and looks like
   this

   [Headers]
   --text follows this line
   --
   [Signature text]

   The above goes goes to the end of buffer with (point-max), then
   searches back to the beginning of the signature separator, which is
   fine. But then it moves one line up to the beginning of '--text
   follows this line--" and then inserts the newly created message body,
   which means that it inserts it before the '--text follows this
   line--'.

   I've solved the problem (I think) by adjust the above code as follows:

       (if (re-search-backward message-signature-separator nil t)
	     (progn
	       (beginning-of-line)
	       (newline))
	 (goto-char (point-max)))

   Hope this helps!

   Best,

   Geoffrey