Auteur: Steve Litt Date: À: dng Sujet: Re: [DNG] dbus [was: Re: logging uses of machine-id]
On Tue, 12 Mar 2019 13:35:10 +0100
Antony Stone <Antony.Stone@???> wrote:
> On Tuesday 12 March 2019 at 13:07:08, aitor_czr wrote:
>
> > Hi,
> >
> > On 12/3/19 9:25, KatolaZ wrote:
> > > Again, this is pretty pointless: just look for reverse-deps on
> > > libdbus and you'll find the answers you are looking for.
> > >
> > > /var/lib/dbus/machine-id is read by anyting that opens a dbus
> > > channel and/or sends/receives a message through dbus. It is not
> > > read by the process itself, rather by libdbus (please have a look
> > > at dbus' code).
> >
> > The next systemd...
>
> If that's a serious comment, what is evil/wrong with (lib)dbus?
* It's a single point of failure
* It can transmit megabyte long messages. Bad idea.
* In the past I've had dbus instances consume 99% cpu.
* By acting as a messaging turntable, knowing who caused what gets much
harder. Witness the fact that normal apps don't error on machine-id,
that happens from dbus.
* dbus is getting harder and harder to avoid.
* dbus is a critical component of systemd.
>
> I ask because I think if we can define what we specifically disagree
> with about systemd and dbus (and anything else which might qualify),
> this would be a powerful message to put on the Devuan web page to
> explain precisely why it exists and why people should consider
> migrating to it.
Systemd and dbus are both massively entangled monoliths. For that
matter, so is KDE, Gnome, and the kernel. Kernel can't help it: It does
a lot, and performance mandates that a lot of that stuff stays in the
kernel.
KDE is such a massively entangled monolith that in 2013 I banished all
KDE programs and libraries from my computer. I could likely do the same
with Gnome if I wanted.
And therein lies the reason systemd is such a special case in evil.
Once it's on your computer, you can't just kick it off.
I suggest Devuan not take a stand on dbus, KDE, Gnome, or other
software where we have a choice. The reason we're against systemd is
once it's on your computer, to get rid of it you need to change
distros, or hope that Debian and Gentoo keep giving us the "gift" of
init choice.
>
> If we can identify things other than systemd which we consider to be
> bad for people (provided there's a common justification for why
> they're bad) I think this is far better than just saying "we don't
> like systemd" as a specific element of modern Linux distros.
Well, from my perspective, anything lacking modularity (massive
entanglement) is something I don't want. But there are practical
priorities. Systemd is much, much worse, and I'm willing to unite with
KDE fans in order to fight systemd. If it's shown that Gnome can run
without systemd, I'll unite with Gnome enthusiasts too.
The perfect is the enemy of the good, so I'll not demand modularity
purity of criticizing stuff like KDE and the current dbus.
> > Recently i packaged wpa removing its dependency on libdbus, but i
> > did it for lubuntu (merely a trial) because i was testing the
> > live-sdk on it; so, this packaging is missing in the repository of
> > gnuinos. Most of the network managers depend strongly on dbus, such
> > as conman, wicd, etc.