Re: [RFC] Install introspection files v2

[Dan Williams <dcbw redhat com>]

On Mon, 2008-05-05 at 10:13 +0200, Helmut Schaa wrote:
> Am Freitag, 2. Mai 2008 17:52:41 schrieb Dan Williams:
> > On Thu, 2008-04-24 at 15:18 +0200, Helmut Schaa wrote:
> > > Hi,
> > >
> > > Attached is a new version of the patch (I merged both patches together)
> > > against r3591. Therefore the introspection files should be up-to-date.
> >
> > So I apparently lied; we may not (yet) be able to split out the files
> > with multiple interfaces.  Is that going to be a problem?  We'll need
> > updates to dbus-glib-binding-tool to handle multiple input files, since
> > you can only set the interface information once with dbus-glib, you
> > can't add to it AFAIK.
> 
> It is not necessary to split out files with multiple interfaces. It would only 
> be cleaner if every file contains only one interface.
> 
> Nevertheless I was able to build NM with the splitted file. Are there any 
> other problems I did not notice?

I think the symptom would be NM just not responding to some dbus methods
or emitting some signals.  The calls to
dbus_g_object_type_install_info() install the methods and singals that
the object will respond to, and the type information passed in as the
second argument there comes directly from the files processed by
dbus-glib-binding-tool (the -glue.h files).  When the file gets split,
only part of the information previously in the -glue.h file will be
added to the object.

I did notice that your patch split out the files, but didn't actually
remove the additional interfaces from the existing .xml files that you'd
split out.

> > If that's still OK for you, I'm fine with committing the patch.
> 
> Sure.
> 
> > Also, please note that the introspection XML files are really an
> > implementation detail of dbus-glib, and as such, if GObject
> > introspection or some better method of hooking up D-Bus and GObject, the
> > introspection XML files in their current form will probably go away.
> > We'd have to figure out another way to generate the API docs at that
> > time, which might also include a method to get the introspection XML
> > back out of NM directly with D-Bus Introspect() calls or something.
> 
> The introspection files are needed at compile time to create the needed 
> proxy-classes in KNM (or any other frontend). Getting the XML through a D-Bus 
> call is not an option here.

What I mean is that it might still be possible to generate the .xml
files at build time from _NM_ using introspection XML, like dbus
apparently does right now.  Shouldn't matter though; we'll cross that
bridge when we come to it.

Dan