Re: Side Windows

[Eli Zaretskii <eliz@gnu.org>]

> Date: Sat, 24 Sep 2016 21:06:07 +0200
> From: martin rudalics <rudalics@gmx.at>
> CC: emacs-devel@gnu.org
> 
>  > A question: would it make sense to invert the meaning of the 'slot'
>  > parameter when the main window shows a buffer whose
>  > bidi-paragraph-direction is right-to-left?  The idea is that when the
>  > main window shows R2L text, the entire geometry of the window
>  > arrangement should switch direction, at least optionally, otherwise
>  > applications that want that will have to include tedious code to
>  > recompute the slots on the fly.
> 
> Suppose the main window shows a left-to-right buffer and we have two
> bottom side windows.  Switching to a right-to-left buffer in the main
> window would now mean to exchange the buffers shown in the side windows.
> We would also have to renumber these windows, resize them maybe,
> exchange their parameters and some of those properties stored in window
> configurations.  And we would have to adjust all overlays with a
> `window' property in those windows.  It's certainly possible to do all
> that but it's not entirely trivial.

What's the alternative? that the application programmer does all that
in their application code?  What are the chances of them getting this
right?

> Personally, I think that such layout decisions should be made before the
> first side window is created.

That might be impossible in the general case, because the same window
arrangement could be used for displaying buffers of different
directions.  One example is an email client that shows messages in the
main window.

> I doubt that you change the menubar layout whenever you switch
> `bidi-paragraph-direction'.

Some applications out there actually do that.  Emacs doesn't, but only
because I deliberately decided it wasn't TRT (there's a FIXME comment
about that in the sources).  Menu bars are different, because they are
not really associated with any particular window, they are associated
with a frame.

> But in practice I never work with bidirectional text so I can't
> really tell.

I think we should at least allow for such a behavior as an option.

Thanks.