bug#21731: 24.5; Bug with `transient-mark-mode' wrt `cua-mode'

[Eli Barzilay <eli@barzilay.org>]

This is a bug that I ran into with 24.5.1:

* Run emacs -Q

* Use shifted arrow keys to make some transient selection (on by
  default, so no need to do any customizations).

* Use C-x b to switch to another buffer

* Use unshifted arrows to move in this buffer

* C-x b back into the first buffer

* Now unshifted arrow movements continue to change the selection, as
  if it was made with C-SPC

FWIW#1, it looks like this bug existed for a while -- I see it with
24.3.1 too.  The reason I didn't see it then is that (IIUC) `cua-mode'
did its own thing (correctly), and it looks like this was dropped and
the functionality from simple.el is used instead.

This is why I qualified this as "wrt cua-mode": the buggy behavior
might be considered correct for people who have used it before and
expect it to happen.  (I doubt this, but that's subjective.)

The reason for this was easy to track: `transient-mark-mode' is a
*global* variable.  After shifted movement, it gets a value of
(only . t), and shifted movements turn it back to t even if they
happen in a different buffer.  (I don't see off-hand any quick way to
fix it, but if anyone has an idea, I'd be happy to know about it.)

The fix should involve some buffer-local version of
`transient-mark-mode', but since it's also used globally as a mode,
this might require a new variable.

If the current behavior is intentional, then this becomes harder to
fix, since it would require having both behaviors available.  It would
still be a bug since it changes functionality for people who use
cua-mode -- which would enable this imaginary extension.

-- 
                    ((x=>x(x))(x=>x(x)))                   Eli Barzilay:
                    http://barzilay.org/                   Maze is Life!