The following code:

void main () {
	uint8 test[10] = {1, 2, 3};
	uint8 test2[10] = test;
	for (int count = 0; count < test.length; count++) {
		assert (test[count] == test2[count]);
		print (@"count: $count, value: $(test[count])\n");
	}
}

when compiled with valac 0.30.2 and run, it produces:

count: 0, value: 1
count: 1, value: 2
count: 2, value: 3
count: 3, value: 0
count: 4, value: 0
count: 5, value: 0
count: 6, value: 0
count: 7, value: 0
count: 8, value: 0
count: 9, value: 0

When compiled with valac 0.37.91 and run, it produces:

count: 0, value: 1
count: 1, value: 2
count: 2, value: 3
count: 3, value: 52
count: 4, value: 76
count: 5, value: 127
count: 6, value: 0
count: 7, value: 0
count: 8, value: 184
count: 9, value: 153

So the array is not initialized to zero first, before being initialized to {1, 2, 3}.

With valac 0.30.2 the relevant C code produced is:

void _vala_main (void) {
        guint8 test[10] = {0};
        guint8 _tmp0_[10] = {0};
        guint8 test2[10] = {0};
        _tmp0_[0] = (guint8) 1;
        _tmp0_[1] = (guint8) 2;
        _tmp0_[2] = (guint8) 3;
        memcpy (test, _tmp0_, 10 * sizeof (guint8));
        memcpy (test2, test, 10 * sizeof (guint8));

With valac 0.37.91 the relevant C code is:

void _vala_main (void) {
        guint8 test[10] = {0};
        guint8 _tmp0_[10];
        guint8 test2[10] = {0};
        _tmp0_[0] = (guint8) 1;
        _tmp0_[1] = (guint8) 2;
        _tmp0_[2] = (guint8) 3;
        memcpy (test, _tmp0_, 10 * sizeof (guint8));
        memcpy (test2, test, 10 * sizeof (guint8));

_tmp0_[10] is no longer initialized to zero first with the newer version of Vala. Ironically test[10] is still initialized to zero, so this can't be a performance enhancement?