The code that failed 21.4.13 did in effect the following (the names have been changed to protect the innocent, or something like that): (let ((buf1 (get-buffer-create "buf1")) (buf2 (get-buffer-create "buf2"))) (set-buffer buf1) (insert "This's not some text\n") (set-buffer-modified-p nil) (toggle-read-only 1) (goto-char (point-min)) (re-search-forward "not " nil t) (set-buffer buf2) (insert "Please don't overwrite me, please!\n") (goto-char (point-min)) (re-search-forward "foo" nil t) (replace-match "")) The second replace-match succeeded and that appears to be fixed in 21.4 CVS. Whew. Leaving aside the question of whether the last search match data needs to save the buffer, I can possibly see the need for that, though it breaks FSF compatibility. What is the logic behind not clearing it after a failed search (other than that's the way Stallman wishes it done)? Is there any possible use for stale match data after a failed search? Somehow, I suspect not. -sb (Please overwrite me, please!) _______________________________________________ XEmacs-Beta mailing list XEmacs-Beta(a)xemacs.org http://calypso.tux.org/cgi-bin/mailman/listinfo/xemacs-beta