Re: impending gdbus merge

[David Zeuthen <zeuthen gmail com>]

Hey,

On Wed, May 12, 2010 at 9:51 AM, Christian Persch <chpe gnome org> wrote:
> Hi;
>
> I have just a few remarks/questions after having ported a couple of
> things to gdbus, and having looked at the API:

Cool, thanks for doing this!

> - g_bus_own_name has a _on_connection variant that directly takes a
>  GDBusConnection*; maybe g_bus_watch_name and g_bus_watch_proxy
>  shoule also have such a variant.

Sure, if it turns out we need such variants we can always add them

(IIRC, the reason I added g_bus_own_name_on_connection() was simply to
make writing test cases easier...)

> - timeouts in g_dbus_connection_call[_sync] etc are in ms and default
>  to 30000 ms. Does it really matter if the timeout is exact here, or
>  just 30s-ish ? I'd suggest to make these timeout parameters take
>  seconds instead, and use the sloppy glib timeouts. (If there really is
>  a use case for exact timeouts, maybe a call flag like
>  G_DBUS_CALL_PRECISE_TIMEOUT could be added that makes these timeouts
>  interpreted as ms and use precise glib timeouts.)

I don't think it matters much - I mean, timeouts are not common at all
so using sloopy timeouts is kinda like optimizing for the 0.1% case.
And since libdbus (and other bindings) already uses milliseconds I'd
like to keep it the same for compat reasons.

> - should g_bus_unwatch_name/proxy call the name/proxy_disappeared
>  callbacks, if the name/proxy_appeared callbacks have been called
>  before? It appears they don't. (I checked the watch_name one; I
>  assume watch_proxy is the same?) It would be nice not to have to
>  manually clean up after calling unwatch, when I also do the same in
>  the disappeared-callback.

It actually used to be like that. This is why we changed it

    Don't guarantee that on_name_lost() will be called after g_bus_unown()

    This guarantee made it impossible to use g_bus_own_name() from within
    a GObject e.g. by called g_bus_own_name() in constructed() and
    g_bus_unown_name() in finalize() because on_name_lost() was delivered
    in an idle.

With the way things work now, you are guaranteed to *never* get
callbacks if you call g_bus_unown_name() from the same thread from
where g_bus_own_name() was called....

Thanks,
David