[webkit-gtk] WebKit2 GTK+ API

[Alexander Larsson]

----- Original Message -----
> El Thu, 15 Sep 2011 08:56:25 -0400 (EDT)
> Alexander Larsson <alexl at redhat.com> escribió:
> >
> >
> > ----- Original Message -----
> > > On Thu, Sep 15, 2011 at 11:50:22AM +0200, Carlos Garcia Campos
> > > wrote:
> > > >
> > > > [...]
> > > >
> > > > Since it's a big change, I would like to share some general
> > > > ideas
> > > > I have about the new API so that we can discuss them.
> > > >
> > > >  - Try to keep webkit1 API when possible to make porting apps
> > > >  easier.
> > > >  - Use async API following the glib/gio pattern for operations
> > > >  started
> > > >    by the user like loading a page, finding text, etc. that has
> > > >    an
> > > >    end.
> > >
> > > I've talked to Carlos and just wanted to do some clarifications
> > > about what this would mean. Developers will have to use an API
> > > where they add
> > > the callbacks for the function instead of connecting to signals,
> > > for
> > > instance, load uri was:
> > >
> > > webkit_web_view_load_uri (WebKitWebView *webView,
> > > const gchar *uri);
> > >
> > > And now would be:
> > >
> > > webkit_web_view_load_uri (WebKitWebView *web_view,
> > > const gchar *uri,
> > > GCancellable *cancellable,
> > > WebKitLoadProgressCallback progress_callback,
> > > gpointer progress_callback_data,
> > > GAsyncReadyCallback callback,
> > > gpointer user_data);
> > >
> >
> > I'm not sure passing the progress callback like this is the best
> > approach. It means e.g. that you have to reconnect the whole
> > progress
> > UI handling on each load, in say a webbrowser. Its a nice API if all
> > your app does is load a single page, but not right if you want to
> > always show the load status of the view in some other widget.
> >
> > One possible way to make the API "nicer" is to return a
> > WebKitLoadOperation object and have progress signals on that.
> 
> what do you mean by return? if load_uri() returns the operation, it
> might be too late to connect the signals and you would still have to
> connect the signals for every load operation. Or do you mean something
> like g_volume_mount() api that takes a mount operation as parameter?

Yeah, i didn't think about it much, but obviously returning something will not work if the data loading is happening on some thread while we're returning from load_uri. In that case you have to pre-setup some object with where you already connected to the signals.