Re: Set X primary selection with Emacs in xterm

[Duncan Findlay]

On Thu, Jun 2, 2022 at 11:56 PM Eli Zaretskii <eliz@gnu.org> wrote:

> Thanks.  I think we should solve this differently.  I don't think it's
> a good idea to call arbitrary Lisp from input-processing loop in
> keyboard.c, anymore than we already do (which is already too much,
> IMNSHO), especially if we envision advices for that code.
>
> We should instead modify the condition in command_loop_1 to support
> terminals that can set GUI selections.  terminal-parameter is a
> primitive written in C, so command_loop_1 could call it directly (it
> should also pay attention to the defcustom described below).

I considered this, but given that we're making the same decision in
lisp/simple.el (deactivate-mark) using display-selections-p, the
benefits of sharing an implementation seemed compelling.

I see your point about wanting to minimize lisp in the command loop.
Can we just port display-selections-p to C and use it from both
places, or will that break things?

> > The second patch changes  `(display-selections-p)' to return true
> > under xterm with the setSelection feature enabled.
>
> This part is mostly fine, but it should be augmented to resolve the
> issues below.
>
> > I don't know if this second patch can be submitted as is. It may break
> > existing users. tmux, for example, removes the selection indicator
> > from OSC 52 codes, so if emacs writes to both CLIPBOARD and PRIMARY
> > selections, both updates will go to the same buffer on the user's
> > side. I've filed https://github.com/tmux/tmux/issues/3192 with tmux. I
> > haven't tested GNU screen.

FWIW, the tmux bug is now fixed.

> > This patch will also lead to extra data being sent to the user's
> > terminal which they may not need or want. It might be wise to only
> > send OSC 52 codes for primary selection if the user actually has a
> > primary selection buffer, but I'm not sure the best way to do that.
> > I'd appreciate some guidance here, or if somebody more experienced
> > wants to take this on, that'd be most appreciated.
>
> I think TRT here is to provide a defcustom, so that users could
> disable this feature if it causes more trouble than it's worth.  With
> time, perhaps we will collect enough user experience to come up with
> the default value that makes the most sense on most supported systems;
> for now setting the X selection could just be disabled by default.

My initial resistance to a defcustom was because this feature already
requires xterm support for setSelection, which is already somewhat
rare, and it's already controlled by `select-active-regions'. Without
a new defcustom, it can be turned off with:
(add-hook 'terminal-init-xterm-hook (lambda () (setq
select-active-regions nil)))

But I accept the feedback that this is not discoverable, and people
want to avoid surprises, so having a defcustom that's off by default
makes sense. I'll upload a new version of the patch to the tracker
shortly.

>
> > --- a/lisp/frame.el
> > +++ b/lisp/frame.el
> > @@ -2164,6 +2164,9 @@ display-selections-p
> >         (not (null dos-windows-version))))
> >       ((memq frame-type '(x w32 ns pgtk))
> >        t)
> > +     ((and (memq frame-type '(t))
> > +           (eq (terminal-parameter nil 'xterm--set-selection) t))
> > +      t)
>
> This is unnecessarily strict: there should be no need to test
> frame-type, since any frame type could arrange for this parameter when
> it supports selections.

In practice, are there other frame types? Is it reasonable to set
terminal-parameter for other frame types?

Thanks
Duncan