Re: attribute data structure rewrite

[Gabriel Dos Reis <gdr@cs.tamu.edu>]

"Giovanni Bajo" <giovannibajo@libero.it> writes:

| Gabriel Dos Reis wrote:
| 
| > We used to accept it because we did not implement offsetof correctly.
| > For example, the following variation should be rejected (in C89 an C++
| > mode)
| >
| >    struct A { char foo[10]; };
| >    void bar(void) {
| >      int i;
| >      for (int i = 0; i < 10; ++i)
| >         {
| >            char ary[offsetof(struct A, foo[i] + 1);
| >         }
| >    }
| 
| I don't understand your example because I don't think the +1 there is valid.
| Unless you meant:
| 
| char ary[offsetof(struct A, foo[i])+1];

Yes, this is what I meant.  
Sorry for the typo, and sending a testcase without compiling :-)

| Anyway, what about this in C99? 

Even in strict C99 mode, it will not be valid.  One could think of GNU
extensions that compute offsetof at run-time, but that is something
we've debated much in the past.  I see the move to the new
implementation as a desire to keep away from that.  
RTH might want to weight in.

| We don't need a constant expression in this
| context, and even if offsetof is always a constant expression we could still
| accept it as an extension. 

I agree that in C99, one does not need a constant expression to make
it work, but the offsetof-subexpression still will not be valid.

| In other words, we could transform it to
| offsetof(struct A, foo)+i*sizeof(foo[0]) in contexts that don't require a
| constant expression.

That is a possibility.  I'm not sure how serious the issue is.

-- Gaby