• Andrej Mitrovic <andrej.mitrovich gmail.com> Aug 17 2010

Andrej Mitrovic <andrej.mitrovich gmail.com> writes:

In TDPL page 251 there's some piece of code and an explanation where we can
force the compiler to do a move on a struct instead of a copy. If I understood
right, then I think there might be a bug here:

import std.algorithm, std.stdio;

struct Widget
{
    private int[] array;
    
    this(uint length)
    {
        array = new int[length];
    }
    
    // postblit constructor
    this(this)
    {
        array = array.dup;
    }
    
    int get(size_t offset)
    {
        return array[offset];
    }
    
    void set(size_t offset, int value)
    {
        array[offset] = value;
    }
}

void kun(Widget w)
{
}

unittest
{
    auto w = Widget(10);
    // Call to move inserted
    kun(move(w));       // w will be moved, 
                                // an empty default-constructed Widget
                                // replaces w's contents
    
    assert(w == Widget.init);   // Passes
}

void main()
{
}

The function kun() does get the moved struct, and not a copy, like intended.
But the assert does not pass. If I change kun to this:

void kun(Widget w)
{
    w.set(2, 40);
}

then in the unittest I will see the changes if I print out the contents of w. 

From what I understand, in this code we want to move a struct and not copy it,
but we don't want "w" in the unittest to refer to the same memory anymore, so w
should get initialized with it's init value. 

But I'm not seeing this behavior in the code. Is this a bug?