Re: [Chicken-users] FFI: wrapping method that returns object

[Felix Winkelmann]

Thomas Chust wrote:
> Sergey Khorev wrote:
>
> > > Your wrapper is the correct way of doing this.
> >
> >
> >
> > Then I have a question ;)) I have created such wrapper, but got
> > strange access violation in chicken.dll during C_c_pointer_or_null.
> > My knowledge of Chicken internals are (temporarily) insufficient to
> > understand what is wrong. It seems that ___abstract is such a reason.
> > Windows, MSVC6 SP5 and SP6.
> > [...]
>
>
> Hello,
>
> apparently the problem is not platform specific as I observed the same 
> behaviour on my Linux box.

Indeed.

> I used the dissector unit to examine the object generated by the call to 
> makeT and found out, that it's pointer slot is set to the unspecified 
> value, which instantly explains the symptoms your program is suffering! 
> When you declare T normally (not ___abstract) as
>     #>?
>     class T
>     {
>     public:
>       T(int x_);
>     };
>     [...]
> though, everything works fine and the pointer slot is occupied. On the 
> other hand, replacing the above snippet by
>     #>?
>     ___abstract class T
>     {
>     public:
>       T(int x_);
>     };
>     [...]
> doesn't break anything. I have the strong feeling, that CHICKEN's ffi 
> parser just ignores the ___abstract declaration and somehow get's 
> confused about the missing constructor declaration in your code -- but 
> I'm only guessing...

The FFI parser omits generating a intilializer method for abstract
classes. The patch I submitted provides a default initializer for
this case.


cheers,
felix