I'm currently trying to upgrade from DMD 2.008 to 2.010. There's one problem I could not find a work-around. There are a bunch of structs which have constants defined inside:

  struct S {
    static const S some_const1 = {0};
    static const S some_const2 = {1};

    int member;
  }

This works fine, until there is a second struct, which uses on of these constants in an equivalent initializer:

  struct T {
    static const T dependent_constant = {S.some_const1};
    // <- error: cannot convert const(S) to S

    S member;
  }

I've tried to use an enum instead, but that causes a segfault in the compiler:

  struct S {
    enum S some_const1 = {0}; // compiler segfault without error message
  }

A possible solution would be to put the constants outside of the struct. However, the actual structs look more like:

  template(S, int N) Vector {
    static const S zero = {...};
    static const S one = {...};
  }
  alias Vector!(int, 2) Vec2i;

  // in code used as Vec2i.one

Since the types used for S are not known, putting the constants outside of the struct is not possible.

I think the semantics should be changed to allow assignment of const values to non-const variables, as long as the specific struct does not contain a reference  /pointer of some form (directly or indirectly).

Also manifest constants inside structs using enum would be nice.

Does anybody have another idea how to sidestep this issue for now?
I think I'm running out of ideas after also having tried a bunch of other things.

Finally found a dirty hack that works:

T ct_strip_const_helper(T)( const(T)[] t ){ return (cast(T[])t)[0]; }
T ct_strip_const(T)( const(T) t ){ return ct_strip_const_helper([t]); }

struct S {
  static const S ess_const = {1};

  int member;
}


struct T {
  static const T tee_const = {ct_strip_const(S.ess_const)};

  S member;
}

Sönke Ludwig wrote:
> Finally found a dirty hack that works:
> 
> T ct_strip_const_helper(T)( const(T)[] t ){ return (cast(T[])t)[0]; }
> T ct_strip_const(T)( const(T) t ){ return ct_strip_const_helper([t]); }
> 
> struct S {
>   static const S ess_const = {1};
> 
>   int member;
> }
> 
> 
> struct T {
>   static const T tee_const = {ct_strip_const(S.ess_const)};
> 
>   S member;
> }

Eww! Here's to hoping the enum-in-structs thing gets fixed.

Both of these should go into bugzilla: http://d.puremagic.com/issues/

Walter Bright wrote:
> Both of these should go into bugzilla: http://d.puremagic.com/issues/

Done.


Also I just noticed that the crash only applies to the specific case where the same struct is used as the enum type, in which it is declared:

struct S { enum S s_const = ...; }

Sönke Ludwig wrote:
> Done.

Thanks.