[patch] remove DBusGPendingCall

[Havoc Pennington]

On Tue, 2005-07-05 at 13:37 -0400, Colin Walters wrote:
> As discussed before, the DBusGPendingCall object's lifecycle is really
> tied to that of the proxy, so this patch removes it entirely from the
> API.  Now the begin_call API looks like:
> 
> guint             dbus_g_proxy_begin_call            (DBusGProxy        *proxy,
>                                                       const char        *method,
> 						      DBusGProxyCallNotify notify,
> 						      gpointer           data,
> 						      GDestroyNotify     destroy,
>                                                       GType              first_arg_type,
>                                                       ...);
> 

I was with you until I noticed the guint return value, which to me means
the change really is:

 - have the proxy own the only refcount on the pending call (aka call 
   id) rather than requiring the caller to own one, so no intermediate
   variable is required - the good part
 - use a guint instead of a typed pointer - I'm not sure about this one

or am I missing something as usual?

> When I was looking over the GNOME Service Manager, I noticed Ray had
> written a function very similar to this, which encourages me that this
> is the right idea.  I think it is a nicer API, and more GLib-like.

My basic take on guint IDs in GLib is that they are crackrock - more
recent APIs support this view with e.g. GSource* replacing the old ID
API.

guints have at least three downsides:
 - lack of typesafety
 - inefficiency due to the hash table vs. 
   just pointer deref
 - weird GLib-specific oddity

> Some other changes in this patch:
> 
> o Updated all callers of begin_call (most of these were actually changed
>   to just invoke dbus_g_proxy_call, since all they did was pair
>   begin_call and end_call)
> o Now the proxy manager has its own special proxy for the bus, in order
>   to reduce redundant code

This reminded me that the glib bindings are supposed to work in direct
app-to-app mode (no bus daemon) in theory, and I have no idea if they do
in practice. 

> o Disabled async client side wrappers for now, sorry Ross!  Will fix 
>   once this patch goes in.
> o Used GType qdata to map object info instead of global hash
> 

Havoc