Re: [gtk-list] Re: types checking signal/callback functions revisited

[Marius Vollmer <mvo zagadka ping de>]

Owen Taylor <otaylor@gtk.org> writes:

> Bil Wendling <wendling@ganymede.isdn.uiuc.edu> writes:
> 
> > I had a similar situation come up in a project I'm working on right now.
> > Basically, using the pointer to the function
> > 
> > 	void callback(GtkWidget *, GtkWhatever *)
> > 
> > when the program expects you to use
> > 
> > 	void callback(GtkWidget *, gpointer)
> > 
> > is not a Good Thing (tm). They are two different types and there is no
> > guarantee that the compiler will do the right thing with them. In other
> > words, this may break on some platforms in undefined ways.
> 
> I'm pretty sure you are wrong about this - at least for all
> platforms worth considering in the context of GTK+. The C standard
> requires (if I remember correctly) that any pointer type will
> survive a round-trip cast to a different pointer type.

I wouldn't be surprised if the C standard only guarantees this if the
compiler actually sees the cast.  This is not the case when you invoke
a function thru a pointer that the compiler thinks is pointing to a
function with a different signature.  That is

    void callback (GtkWidget *w, gpointer p)
    {
      GtkWhatever *we = p;  // visible cast from void* to GtkWhatever*

while

    void callback (GtkWidget *w, GtkWhatever *p);

    ...

    gpointer p;
    ((void (*)(GtkWidget *w, gpointer))callback) (w, p);  // invisible