Re: Emacs's set-frame-size can not work well with gnome-shell?

[Dmitry Gutov]

On 09.03.2020 11:03, martin rudalics wrote:

>  > That, or the patch using the obsolete
>  > gtk_container_set_resize_mode. I'm not 100% sure which is the best
>  > choice (you seem to be favoring the hiding approach, judging by the
>  > next paragraph), but the other one seems to provide a better
>  > experience in the use case I'm looking at. Especially when combined
>  > with (setq x-wait-for-event-timeout 0.0).
>
> The problem is that the gtk_container_set_resize_mode patch might break
> with the next update of mutter or GTK.

I wondered about that. But we'll probably release a new version of Emacs 
before then?

It might also send a stronger message to GTK/Mutter developers, who are 
not so fast to answer a bug report from a fellow GNU project.

> It also freezes Emacs on
> desktops like kwin and xfwm and therefore must be made optional anyway.

That's unfortunate. If there was a way to detect Mutter 
programmatically, maybe we should use that. Even if just to set the 
default value of the new variable you've described below.

> I meanwhile tested some ten window managers and so far only the GNOME
> and Budgie (both mutter based) desktops exhibit the resize problem.

Makes sense.

> So I intend to add an option, say 'x-gtk-resize-child-frames', that
> would be nil by default and have the optional values say 'hide and
> 'resize-mode which would either use the hiding or the container based
> approach to deal with the problem.

I think 'resize-mode' should be the main mode of operation for it. But 
if 'hide' wouldn't complicate the code too much, I can't really argue 
against it.

>  >> It's also the only one that
>  >> can show a scroll bar on a child frame (not necessarily right away but
>  >> after one resize operation at least).  Which hints at some outer-window
>  >> vs edit-window snafu but so far I have no idea where to look.
>  >
>  > FWIW, I'm pretty sure 99.9% of users of child frames out there aren't 
> going to enable scroll bars.
>
> Maybe.  The fact that it doesn't work with mutter still hints at a more
> serious underlying problem.

A more serious problem in Mutter, seems like. But it's not a given, of 
course.