emplace was recetly improved to (better) support structs that have immutable members.

Implementation-wise, it was actually "lucky" it worked (code wasn't written to explicitly support it), and generated runtime that does it is sub-optimal (calls to memcpy when an assignment would be enough, calls to tid.postblit that aren't actually necessary.). Details.

More importantly though, support is "flakey" in the sense that "postblit" initialization will work (S to S), but aggregate initialization will fail.

//----
import std.conv;

struct S
{
    immutable int i;
}

void main()
{
    S s = void;
    emplace(&s, S(1)); //Fails 2.063.2; Passes 2.064ALPHA
    emplace(&s, 1); //Fails on both 2.063.2 and 2.064ALPHA
}
//----

So, in 2.064ALPHA, while "emplace(&s, S(1));" will work, "emplace(&s, 1);" will not. This is inconsistent, and emplace should be fixed to support it.