Re: Selection changes

[David De La Harpe Golden <david@harpegolden.net>]

On 14/07/10 19:08, Chong Yidong wrote:

> I believe that this change should be pretty much seamless, but let me
> know if there is any problems.
>

'fraid the no-zero-length-region change to deactivate-mark, while 
well-intentioned, is wrong - by that stage emacs may have already taken 
ownership of the selection, and pointed it at the buffer.

e.g. try C-SPC C-SPC, move point a few chars, middle click, middle 
click, middle click, Yeuch! Or try making the region nonzero sized then 
zero sized again.

So please remove from deactivate-mark the
(not (eq (region-beginning) (region-end)))
for now - not having it is IMO the lesser of two evils, it's _way_ too 
easy to apparently randomize (well, it's not actually random, but users 
aren't going to appreciate rhat) primary with it there.  At least IMO.

Really, emacs needs to take ownership of the primary selection and set 
it to the buffer /the first time in each region activation that the 
active region becomes nonzero sized/.

Remember when the emacs-level selection val is set to a buffer object, 
the buffer is _lazily_ queried for its point-mark span if/when other X 
clients ask for the selection, so if it manages to stay set to the 
buffer past deactivation of the region you'll have a selection depending 
on where the mark and point are at that time.  Hence freezing off the 
string as the selection in deactivate-mark.

I'm out of "emacs time" for today for actual implementation of a 
proposal for a fix - I'll have time at the weekend if you haven't got to 
it by then...