Problem with a convoluted templated struct

[Ali Çehreli]

On Wed, 29 Jun 2011 04:06:02 +0200, Kiith-Sa wrote:

> Hello.
> 
> I am trying to implement a templated struct containing a data member
> created by std.variant.Algebraic instantiated with the struct's template
> parameters.
> 
> The problem is, I want to template the struct with e.g. arrays (lists,
> etc.) of its own type. I don't know any way to do this without ugly
> string mixin code.
> 
> I'm not sure if I can explain this clearly, so here's some code:
> 
> 
> 
> import std.variant;
> 
> struct Node(Types ...)
> {
>     Algebraic!(Types) value;
>     
>     //other members, etc...
> }
> 
> unittest
> {
>     //this works
>     Node!(string, int) node1;
> 
>     //this is what I want, but I can't do it, since //Node must be
>     templated
>     Node!(Node[], string, int) node1;
> }
> 
> 
> //Now, I can get around this with a string mixin: //(just one argument
> in this example, but it // could be variadic with more complex code)
> 
> 
> struct Node2(string t)
> {
>     mixin("private alias " ~ t ~ " Type;"); Algebraic!(Type) value;
> }
> 
> unittest
> {
>     //works, but ugly
>     Node2!("Node2[]") node;

That works because the name Node2 represents the complete type of the 
templated struct in the struct definition.

> }
> 
> 
> 
> Is there any way to do this without string mixins? (in case this is
> proposed: I specifically need structs, not classes. I know I could do
> this with templated derived classes storing an array/list/whatever with
> parent class type.
> 
> 
> Thanks for any help.

Providing the container type as an alias template parameter seems to work:

import std.container;
import std.variant;

struct MyAlgebraic(Types ...)
{}

struct MyList(T)
{}

struct Node(alias ContainerType, Types ...)
{
    ContainerType!Node container;
    MyAlgebraic!(Types) value;
}

unittest
{
    Node!(Array, string, int) node1;
    Node!(MyList, double, char) node2;
}

void main()
{}

But I hit other compilation problems in Algebraic, which I believe to be 
const-correctness issues.

Also consider looking at Phobos ranges. Instead of templatizing on the 
container type, templatizing on the range type may be better.

Ali