RedBlackTree thin wrapper initialization

[Rene Zwanenburg via Digitalmars-d]

On Wednesday, 28 May 2014 at 07:13:37 UTC, BlackEdder wrote:
> I'm trying to write a thin wrapper around redblacktree, but it 
> seems every object of the class shares the same copy of 
> redblacktree. Am I doing something wrong or is this a bug.
>
> Minimal code example:
>
> import std.array;
> import std.container;
> import std.stdio;
>
> class A {
>     auto tree = new RedBlackTree!string();
> }
>
> unittest {
>     auto a = new A();
>     a.tree.insert( "a" );
>     auto b = new A();
>     writeln( "Should be empty, but is: ", b.tree.array );
>     writeln( "Should be empty, but has length: ", b.tree.length 
> );
> }
>
> Which results in the following output:
> $ rdmd -unittest rbt.d
> Should be empty, but is: ["a"]
> Should be empty, but has length: 1
>
>
> I'm using dmd 2.0.65.

The problem is the initialization of A.tree. Short answer, move 
it to a constructor and it should work.

Now what I think is going on: default values for member fields 
must be known at compile time. This is why structs can do 
something like

struct S
{
   int i = 5;
}

while they're not allowed to define a default constructor.

So it appears the tree is instantiated at compile time and stored 
somewhere, and the reference to that instance is used as the 
default value for A.tree. I didn't know this was possible to do 
with reference type members, and perhaps it should be disallowed. 
This is highly counter intuitive behavior esp. for people coming 
from Java.