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

[Dmitry Gutov]

On 18.01.2020 11:32, martin rudalics wrote:
>  > Is there anything more I can do to facilitate fixing this?
>
> We'd first need a somewhat wider audience here.  Are all the colleagues
> running into these problems on GNOME or tunneling into it, do they all
> use a HiDPI display?

The reports are here:

https://github.com/tumashu/company-posframe/issues/2

Mutter is singled out as a cause, HiDPI or not.

> Could one of them try to change the WM for
> experimenting so we can at least determine that this is a mutter-only
> problem? Maybe
> there's also someone still running Metacity who could tell us whether
> the same problems exist there already.

Andrey Orst said that he tried it with Mate desktop (using a fork of 
Metacity), and didn't see this problem here.

> Could we find at least one guru able to build mutter so she or
> he can debug it - I don't have the faintest idea how to do that.

I've posted that question.

> Here with Xfce on Debian I'm using a minibuffer child frame that moves
> along with the selected window, resizes exactly like the normal
> minibuffer window and disappears whenever it's not active.  So child
> frames work reliably here.

Well... guess I'm happy it works at least for some users.

> OTOH, as we've seen with Bug#38452, mutter also seems to have problems
> processing calls for normal Emacs frames as requested.  So we really
> should try to contact Florian Müllner (address@hidden) - maybe he
> could give us at least some hints on how to proceed or what to test.
> But this means that at least two or three people using Emacs on GNOME
> shell should participate, so this doesn't look like some sort of stray
> issue, occasionally encountered by one or two users as with
> https://gitlab.gnome.org/GNOME/mutter/issues/840.

Should we start commenting on that issue?

Maybe you could start with a more comprehensive explanation there. Using 
terms that Mutter developers would understand more easily (meaning, not 
Lisp code).

Florian seems to be subscribed to that issue, BTW.