bug#67526: 30.0.50; Regression: gnus-group-prev-group fails to move to first group

[Simon Leinen <simon.leinen@switch.ch>]

> I do note that the issue goes away when I revert a bit of your
> change, as follows.  I haven't found out *why* that is necessary to
> get the old behavior back...

OK, I think I know now.  It was a mistake to add the "(if backward
(bobp) ...)" branch.  Probably the intent was to restore symmetry for
the backward-moving case.  But the pre-existing (eobp) test will
succeed only when the cursor is *beyond* the last newsgroup line -
because the end of the buffer is not at the end of the last newsgroup
line, but on an empty line *after* that last newsgroup line.  Testing
for (bobp) in the backwards case is not equivalent, because (at least
in my case) the beginning of the buffer position is very much *on* a
valid/eligible newsgroup line.  Maybe this is not the case for Eric,
because he uses topic mode? Anyway in my limited testing, the old
logic works fine for that case too.

So I suggest to revert that part of Eric's change, as per the patch
I'm resending here with a correction - I had written ENDP instead of
EOBP, sorry about the confusion.

diff --git a/lisp/gnus/gnus-group.el b/lisp/gnus/gnus-group.el
index 9a1a6f9b27d..f2258026eca 100644
--- a/lisp/gnus/gnus-group.el
+++ b/lisp/gnus/gnus-group.el
@@ -1758,7 +1758,7 @@ gnus-group-search-forward
     (unless first-too
       (forward-line way))
     (while (and
-	    (not (if backward (bobp) (eobp)))
+	    (not (eobp))
 	    (not (setq
 		  found
 		  (and

If you're worried - and I haven't thought through whether this should
be of any concern - that the (eobp) check should not be done in the
backwards case, then a suitable alternate clause could be

          (or backward (not (eobp)))
-- 
Simon.