Re: [Vala] A seemingly impossible segfault.

[Darren Warner <vala-list dazwin com>]

There is a bug in vala, but also one in vfcgi. I've just pushed fix here 
 (be careful though - it goes as far as changing the command line 
options, possibly not want you want):

http://bazaar.launchpad.net/~launchpad-dazwin/libvfcgi/fixes/revision/23

The bug in vala (I think) is that setting a value declared as 
'construct' (as per 'max_threads' in gio-2.0.vapi) should produce an 
error instead of a warning (otherwise, the invalid code as I posted 
below is the result).

-Darren

Michael Trausch wrote:
> On Mon, Dec 14, 2009 at 1:35 PM, Darren Warner <darren dazwin com> wrote:
> > Hi Michael,
> >
> > I think this is the same problem I was experiencing (though I get it with
> > vala 0.7.6 as well, so I'm not sure why this was previously working for
> > you).
>
> Well, it hasn't worked since I added that line, on this project...
>
> > The problem is that bad code being generated by vala - 'self' is being used
> > before it is being initialized. The following simplified code will reproduce
> > the problem:
>
> D'oh!  Now that I look at the generated code more closely, I see that.
>  So I suppose that means that there is a bug in Vala (anyone know if
> it has been fixed yet?  I suppose I can go build trunk and see…).
> Crap, that means my desire to depend on Vala 0.7.6 as a minimum is
> out.  Oh, well.
>
> > public class Bar : GLib.Object {
> >    private ThreadedSocketService m_service;
> >    private uint16 m_max_connections;
> >
> >    public Bar() {
> >        m_service = new ThreadedSocketService(10);
> >        m_service.max_threads = m_max_connections;
> >    }
> > }
> >
> > Compiling with 'valac -C bar.vala --pkg=gio-2.0' will produce the following:
> >
> > Bar* bar_construct (GType object_type) {
> >        GParameter * __params;
> >        GParameter * __params_it;
> >        GThreadedSocketService* _tmp0_;
> >        Bar * self;
> >        __params = g_new0 (GParameter, 1);
> >        __params_it = __params;
> >        self->priv->m_service = (_tmp0_ = (GThreadedSocketService*)
> > g_threaded_socket_service_new (10), _g_object_unref0
> > (self->priv->m_service), _tmp0_);
> >        ...
> >        self = g_object_newv (object_type, __params_it - __params, __params);
> >
> >
> > I don't know what it is about ThreadedSocketService, but I cannot reproduce
> > the problem if I use a simple 'Foo' class instead, even if it is derived
> > from GLib.Object. Bar does need to be constructed from GLib.Object - the bug
> > is related to some kind of readahead optimization maybe?
> >
> > I was going to raise a bug about this, but wanted to check if you were
> > having the same problem as well - seems you are.
>
> It would seem so.  Do you want to file the bug report, or shall I?
> Since the short example is something you already have, and the problem
> is still in trunk… I'd guess you probably want to?
>
>    --- Mike