roadmap

[Michael Biebl]

2008/8/11 Havoc Pennington <hp at pobox.com>:
> Hi,

Hi Havoc,

> On Mon, Aug 11, 2008 at 12:59 PM, Michael Biebl <mbiebl at gmail.com> wrote:
>> That's not correct afaik. E.g networkmanager 0.6 has logic to
>> reconnect to the system bus (and it works ok). I know other services,
>> like powersaved, which also can cope with a system dbus restart just
>> fine.
>>
>> I also know, that Sjoerd fixed a couple of GNOME programs, so they can
>> cope with a session bus going away (e.g. gnome-keyring). And from my
>> day-to-day usage I think that programs like gnome-power-manager work
>> just fine after dbus restarts. Actually most programs do, only
>> nautilus/gnome-vfs is an unfortunate exception.
>>
>> So, it's very well possible, it just requires additional work.
>
> For the system bus it is possible, though note that to rely on it we
> have to know that *all* services are not only supposed to survive
> restart, but in fact have been tested and do survive restart. That is
> a fairly high bar. Relying on this is definitely not easier than
> adding feature negotiation.
>
> For the session bus, surviving restart doesn't make any sense and is
> actively wrong, generally speaking. The DBUS_SESSION_BUS_ADDRESS env
> variable can't be modified post-login, so there is no way a new
> session bus will appear; we can't "restart" the session bus.
>
> Perhaps more importantly, we've defined that the lifetime of a user
> login session is the lifetime of the session bus and X server, and
> that the session bus and X server should have the same lifetime. Thus,
> applications and services should be exiting when either the session
> bus or X server exits. Sysadmins get very angry if bits of a user
> session are leftover post-logout. It's almost always a bug if anything
> does this. (The exceptions are few, and would be things that
> inherently must stay active after logout, an ssh session example is
> "screen" or "nohup", I can't think of a GUI session example offhand.)
>
> Even if some stuff survives session bus restart, most stuff *should*
> be exiting if the session bus does, so we can't restart it without
> logging people out.
>

Sorry, I was a bit misleading regarding the session bus.
What I meant to say is, that applications in the desktop session, like
g-p-m, can cope with *system* bus restarts, i.e. if they use services
like HAL or NM.

Your remarks regarding the session bus are of course correct.

Cheers,
Michael


-- 
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?