Re: [Chicken-users] Objective-C interface (sort of...)

[Thomas Chust]

Am 26.11.2005, 14:30 Uhr, schrieb felix winkelmann <address@hidden>:

> On 11/25/05, Thomas Chust <address@hidden> wrote:
> > Building NSInvocation instances is even easier than doing everything by
> > hand -- for example the parsing of ObjC parameter type encoding strings
> > is much less work, because the NSMethodSignature objects at least split
> > them
> > correctly for you. You just have to create the object, set the  
> > parameters,
> > execute the call and retrieve the return value -- like in this  
> > pseudocode
> > snippet:
> >    SEL sel = sel_registerName("someSelector");
> >    NSMethodSignature *sig = [someObject methodSignatureForSelector:sel];
> >    NSInvocation *inv = [NSInvocation invocationWithMethodSignature:sig];
> >    [inv setTarget:someObject];
> >    [inv setSelector:sel];
> >    for (int i = 2; i < [sig numberOfArguments]; i++)
> >      [inv setArgument:convertArgumentFromScheme(i-2, [sig
> > getArgumentTypeAtIndex:i])
> >               atIndex:i];
> >    [inv invoke];
> >    void *rbuf = alloca([sig methodReturnLength]);
> >    [inv getReturnValue:rbuf];
> >    convertReturnValueToScheme([sig methodReturnType], rbuf);
> >    [inv release];
> >    [sig release];
>
> Yes, this looks fine. Can one assume the buffer for arguments/return  
> values
> uses a "natural" representation?
> [...]

Hi,

As far as I know (because it always worked like that :), the parameter or
return value buffer can just be a pointer to a C variable of the required
type. I haven't tried this with structure types passed by value, though,
where alignment issues could arise.

cu,
Thomas