The constness problem

[Lars T. Kyllingstad via Digitalmars-d]

On Saturday, 12 July 2014 at 15:03:13 UTC, Peter Alexander wrote:
> http://pointersgonewild.wordpress.com/2014/07/11/the-constness-problem/
>
> I can relate to this. I don't use classes much in D, but I find 
> it frustrating that you need to reach to 
> std.typecons.Rebindable for what I would consider to be a more 
> common usage of const with classes.

It seems she laments the lack of both head-const in general and 
tail-const for classes.  Head-const isn't really an issue, in my 
opinion, because it is rarely needed and easily obtained with 
existing language features:

     struct HeadConst(T)
     {
         this(T value) { value_ = value; }
         alias get this;
     private:
         @property T get() { return value_; }
         T value_;
     }

     HeadConst!MyClass obj = new MyClass;
     obj = new MyClass; // Error!

The lack of tail-const class references, on the other hand, is 
one of D's most severe shortcomings.

> I really feel that a mutable reference to a const object should 
> be expressible in the core language.

Michel Fortin once made a working pull request for DMD that 
introduced the syntax

     const(MyClass) ref thisIsRebindable = new MyClass;

Personally, I would prefer replacing the current object reference 
syntax with T$ (or T# if that conflicts with the current dollar 
operator), like this:

     Foo$ obj1 = new Foo;        // Rebindable and mutable
     const(Foo)$ obj2 = new Foo; // Rebindable but const
     const(Foo$) obj3 = new Foo; // Fully const

The current 'T' reference type would of course have to be 
accepted and equivalent to the 'T$' reference type for a long 
time before it is eventually deprecated.

Come to think of it, "new T" could return "T$" for non-class 
types too, where T$ would then act like a non-dereferentiable 
pointer on which it is not allowed to do pointer arithmetics.

Lars