newCTFE report July

[Stefan Koch]

Hi Guys, last month I did the work make this snippet compile and 
execute correctly:

static immutable four = [1, 2, 3, 4];
int fn(int idx = 2)
{
     int fn2(const int* x)
     {
       return x[idx];
     }

     return fn2(&four[0]) + *(&four[0]);
}

static assert(fn() == 4);


There where two major problems with this one.

1. There was the circular initialization-detected which would 
object to referencing `&four` twice in the same expression.

2. there was the problem of the IndexExp being unaware of being 
inside an AddrExp which would cause it to first evaluate four[0] 
to 1, and then trying to take the address of literal one (which 
is zero as the marker for an invalid address).

Fixing this involved adding a special case to how &x[y] is 
handled.
This solution does complicate the internals of newCTFE a bit, and 
I hope to simplify it once other blocking problems are out of the 
way.