On Wed, Apr 09, 2008 at 06:34:05PM +0100, Andrew Haley wrote:
>
> But it's *not* always easily accessible. In this case,
>
> int bar(int p[n])
Well, n must be referring a symbol that is in scope, so
it must be bound to a value at run-time. Otherwise, if its undefined,
the sizeof should also be undefined...
>
> n is somewhere in global scope, maybe in another translation unit,
> and as I point out below you'd have to copy it at the time the array
> was created.
What's wrong with taking a snapshot of "n" when the function is
entered? After all, that's what p[n] specifies in the signature?
>
> So, you're saying it *should* create a hidden copy of the size
> parameter for use by sizeof? While this would work, it's not very
> C.
Hhm, not strictly following C99 standard, sure, but following common
sense, yes. That's what seems to be happening in this valid C99
snippet:
int main() {
int s=10,a[s];
s=11;
assert(sizeof a == 40);
}
I do understand that our idle rumblings will not change the C99
standard, but, frankly speaking, its very disappointing having to
write things such as:
void foo(int s,int a[s])
{
/* work-around the C99 always returning
* sizeof(int*) instead of s*sizeof(int)
*/
int _a[s];
s=11;
for (int i=0; i<sizeof _a/sizeof _a[0]; i++);
}
This one also generates slightly less optimal code...
Maybe it's a good canditate for a GCC extension?
>
> Andrew.
>
Pjotr
