Let this() figure out T implicitly?

[Timon Gehr]

On 02/18/2012 03:22 AM, Ali Çehreli wrote:
> On 02/17/2012 05:59 PM, Timon Gehr wrote:
>  > On 02/18/2012 12:04 AM, Ali Çehreli wrote:
>  >> On 02/17/2012 09:08 AM, Steven Schveighoffer wrote:
>  >>
>  >> > What you are asking for is IFTI (Implicit Function Template
>  >> > Instantiation) on constructors, and is perfectly possible, but not
>  >> > implemented:
>  >> >
>  >> > http://d.puremagic.com/issues/show_bug.cgi?id=6082
>  >>
>  >> What was the resolution for the case when the constructor is a
> template?
>  >> I think that's why IFTI doesn't work for structs and classes in C++.
>  >> Although, I can't get the following to compile in D anyway (that old
> and
>  >> annoying error message again! :p):
>  >>
>  >> Error: no property 'opCall' for type 'deneme.B'
>  >>
>  >> import std.conv;
>  >>
>  >> class B
>  >> {
>  >> string s;
>  >>
>  >> this(T)(T t) // <-- Is the struct a template
>  >> // or the constructor a template?
>  >
>  > The constructor it is, and B is a class, not a struct.
>
> Ah! Thanks!
>
> So today the following works; and not B, but the constructor is a template:
>
> import std.conv;
>
> struct B
> {
>      string s;
>
>      this(T)(T t)       // <-- constructor is a template
>      {
>          s = to!string(t);
>      }
> }
>
> void main()
> {
>      auto b0 = B(42);       // construct with int
>      assert(b0.s == "42");
>
>      auto b1 = B(1.5);      // construct with double
>      assert(b1.s == "1.5");
> }
>
> With the proposed feature, the struct would be a template.
>

With the proposed feature still the constructor would be a template. The 
proposed feature is just that IFTI for constructors should be able to 
deduce the parameters for the enclosing struct/class template 
additionally to just its own parameters.


>  >> {
>  >> s = to!string(t);
>  >> }
>  >> }
>  >>
>  >> void main()
>  >> {
>  >> auto b0 = B(42);
>  >> }
>  >>
>  >> I wasn't around when the static opCall() was designed but it is
> probably
>  >> the very first thing that bugged me about D. :) I want B(42) to be
>  >> object construction, not opCall(), which is not even defined.
>  >>
>  >> Ali
>  >>
>  >
>  > Why? What useful semantics would that have for classes?
>
> I am not sure but the point is, if function templates provide IFTI, then
> because of being functions the constructors could provide IFTI as well,
> as it does for the struct B above.
>
> As an aside, I don't know why it is not the same with classes. I hope I
> am not again making an error in the following code. I have replaced
> 'struct' with 'class' and inserted two 'new' keywords:
>
> import std.conv;
>
> class B
> {
>      string s;
>
>      this(T)(T t)
>      {
>          s = to!string(t);
>      }
> }
>
> void main()
> {
>      auto b0 = new B(42);   // line 61459
>      assert(b0.s == "42");
>
>      auto b1 = new B(1.5);  // line 61462
>      assert(b1.s == "1.5");
> }
>
> The compiler says:
>
> deneme.d(61459): Error: no constructor for B
> deneme.d(61462): Error: no constructor for B
> deneme.d(61462): Warning: statement is not reachable
>
> Ali
>

This seems to be a bug.