Regression in 2.060 - corruption when lambda closes over foreach variable

[Ben Davis]

Hi,

I have some code which has started breaking as of 2.060. I've simplified 
it down to the following:

------------

import std.stdio;

void main() {
        broken();
        reffix();
        noclosefix();
}

void broken() {
        foreach (z; 0..1) {
                writeln(z);     //Corrupt
                () {writeln(z);} ();    //Corrupt (same)
        }
}

//Adding 'ref' fixes it:
void reffix() {
        foreach (ref z; 0..1) {
                writeln(z);     //0
                () {writeln(z);} ();    //0
        }
}

//It only breaks if 'z' is closed into the lambda:
void noclosefix() {
        foreach (z; 0..1) {
                writeln(z);     //0
                int z2=z;
                () {writeln(z2);} ();   //0
        }
}

------------

The sort of corrupt values I see for z are for example
28835840 = 0x01B80000
29949952 = 0x01C90000
38535168 = 0x024C0000
36110336 = 0x02270000
But it's always the same between one writeln and the other.

Also breaks with foreach_reverse.

I'm compiling with no switches at all, just "dmd filename.d".

I'm compiling with debug, without optimisations.

Anyone fancy taking a look?

Thanks,

Ben :)