The behavior of appended kills is described in

http://www.gnu.org/software/emacs/manual/html_node/emacs/Appending-Kills.html

The idea is that when consecutive kills happen, the new text is either
appended or prepended to the kill ring entry, depending on the ordering
of the mark/point. This works as described for commands such as C-k,
M-d, etc. However, the behavior for C-w and M-w is surprising.

Suppose my buffer contains

123

with the point at the '1'. I then use point navigation commands, C-M-w
and C-w to kill 1 then 2 then 3:

C-SPC C-f C-w C-SPC C-f C-M-w C-w C-SPC C-f C-M-w C-w

I would expect the kill-ring to then contain "123". Instead, it contains
"321". Similarly I get "321" if I do this backwards, i.e. starting with
the point at the '3':

C-SPC C-f C-w C-SPC C-b C-M-w C-w C-SPC C-b C-M-w C-w C-y


If I perform similar actions, but with M-w instead of C-w, I get "123"
for the forward case and "321" for the backward case. I would expect
"123" in all cases.


The current behavior for commands after C-M-w is:

C-w: prepend if point>mark; append otherwise
M-w: always append new kill


I think the behavior should either

1. Match what commands like M-d do; "123" would then result in all 4 of
   the cases described above

or

2. Always append. This is what the manual node at the top of this report
   says. The 4 cases described above would then produce "123", "321",
   "123", "321" respectively.


I'm attaching a patch that implements behavior 1 and another that
implements behavior 2. I have a mild preference for behavior 1, but both
are better than what emacs does currently, I think.

Historically, the behavior has been "always-append". The current
behavior was established in 2006 by
http://git.savannah.gnu.org/cgit/emacs.git/commit/?id=b7690594

I think the intent of that commit was to implement behavior 1, but I'm
not sure.

dima