bug#9581: 24.0.50; dbus-unregister-object fails if service is nil

[Michael Albinus <michael.albinus@gmx.de>]

Julien Danjou <julien@danjou.info> writes:

> When using `dbus-register-signal', this happens:
> 1. the dbus_bus_add_match() function is called to add a match on the bus
> 2. the (match callback) pair is recorded into
>    `dbus-registered-objets-table' 
>
> This makes things work. When a signal happens, something is looking into
> `dbus-registered-objets-table' and call the callback function.
>
> But to stop listening for a signal, the function to use is
> `dbus-unregister-object', and it is doing this:
> 1. remove the (match callback) pair from `dbus-registered-objets-table'
> 2. call ReleaseName on the service we were listening
>
> While I agree on point 1., the point 2. is totally irrelevant in such a
> case. There's no need to do such a thing, the name has never been
> requested with RequestName before.

It's simply an error. We are speaking abut the generalized
`dbus-unregister-object', which is used for both signals and
methods. ReleaseName shall be called only in case a *method* has been
registered; I'll fix this.

> I think that:
> - step 2 should be removed or another function should be created which
>   does not send a ReleaseName

Nope. See above.

> - dbus_bus_remove_match() should be used to remove the watch from the
>   bus, which would be a lot cleaner.

Good point. Registering a signal shall also keep the match string in
dbus-registered-objects-table (it doesn't yet). Then we could use this
string to send RemoveMatch.

I'll prepare a patch for this (hopefully in time before starting the
pretest).

Best regards, Michael.