General questions/conjectures about DBUS etc.

[Andrzej]

On 03/24/2012 06:29 AM, Chris Green wrote:
>
> (running Calibre in particular)

Which Calibre? (just curious)

> I've noticed with thunar and calibre that running them on my laptop but
> using my desktop machine as the display it feels as if they're sort of
> "half and half", some of what they're doing relates to the laptop on
> which they're running but some relates to my desktop machine where they
> are displaying.

It's not just thunar. It is a general problem and Xfce is not in the 
position to change this trend. Which makes the rest of this email 
off-topic, but well...

This behavior is exactly what you would expect when mixing DBUS, audio 
and X together. X only forwards your display, keyboard and mouse. Audio 
and DBUS messages go through completely different channels, which, 
unless you set them up explicitly, usually end up pointing to the remote 
(X client) machine.

There are also other problems, modern applications usually use some X 
extensions that require shared memory (render, composite etc), perform 
poorly or look ugly over network, NFS is getting rare these days (what's 
the point of having network transparency if each app sees different 
data?) etc. In short: network transparency only works with legacy X 
applications in a legacy X environment (with NFS/NIS, ethernet etc.).

You can try to hack things a bit. For DBUS you can play with 
DBUS_SESSION_BUS_ADDRESS variable (first you will have to enable 
networking in your local DBUS session daemon). To tell the truth, I've 
never tried it. I found that accessing a remote machine over VNC/RDP 
gives much better results - there is no network transparency but at 
least it is made obvious (not half-remote, half-local as in X). Often 
the performance is even better than X because these protocols are less 
sensitive to network latency and the bandwidth has improved a lot. Not 
to say about detach/reattach capability, which is a must today.

> Does this sound possible/sensible?  Are modern GUI apps tested in a
> multi-machine X environment?

Worse, it is *designed* not to work well in such environments. The 
disparity between network latency and performance of local systems 
became so large that people started pushing local-only solutions. Can't 
really blame them for that - X was good in '90 when graphics performance 
was limited by drawing operations, network latency was smaller than now, 
network and memory bandwidth narrow, and network installations we 
centralized. None of this is true now.

</offtopic>

Andrzej