Re: applications that fail to build with new gnustep from svn

[David Chisnall]

On 6 Apr 2011, at 10:25, Riccardo Mottola wrote:

>> If that code works for you, you really should update your compiler :-)
> Well, I get no warning at all ranging from gcc 2.95 to gcc 4.5.2...


I suspect this is because gcc doesn't really understand Objective-C types, it 
just follows the StepStone approach of mapping classes to structures and adding 
a few extra checks.  Unless you happen to hit one of the specific cases where 
someone on the GCC team thought about checking, then they're just treated as C 
types.  In this case, I think it will be passing each value through as a 
structure, so trying to write a complete copy of the object out onto the stack. 
 This will, of course, break horribly in the presence of subclassing, which is 
why the language disallows it.  

I'm a bit surprised though, as this ought to be caught by the code that handles 
dereferencing pointers (which should check if it's an Objective-C type, and 
reject it if it is).

In clang, the Objective-C types are represented in the type system in the AST, 
right up to code generation. This means that any part of the code generation 
knows if it's dealing with Objective-C objects or pointers to C structures (or 
C++ classes).

David

-- Sent from my IBM 1620