Re: Zero-length arrays

[Graham Stott <grahams at redhat dot com>]

Richard

Richard Kenner wrote:
> 
> What *precisely*, are the semantics of this feature.  Consider the following:
> 
> struct foo {int a; int x[0]; int y;} xyz = {1, {2, 3, 4}, 5};
> 
> int
> sub1 (struct foo *p)
> {
>   return p->y;
> }
> 
> This compiles with no errors with the current tree and XYZ contains in
> memory the integers 1 through 5 as consecutive words.
> 
> But sub1 would return 2 when passsed &xyz, not the correct value of 5.
> 
> Of course there's no way to support returning the proper value here, so I
> think the initializer should produce an error message.

I believe the zero-length arrays are only valid if they are the last
element of a structure which is not nested within another structure.

So I would agree your example should not compile hence we have a bug :-(

Graham