• Atila Neves (16/16) Dec 22 2018 Fun times:
• Rubn (27/43) Dec 22 2018 Per https://p0nce.github.io/d-idioms/

Atila Neves <atila.neves gmail.com> writes:

Fun times:

---------------
void main() {
     Foo foo;
     takesRef(foo.create.i);  // oops
}

struct Foo {
     int i;
     Foo create() {
         return Foo();
     }
}

void takesRef(ref int x) { }
---------------

Even more fun with auto ref:

https://issues.dlang.org/show_bug.cgi?id=19507

Rubn <where is.this> writes:

On Saturday, 22 December 2018 at 16:26:31 UTC, Atila Neves wrote:
 Fun times:

 ---------------
 void main() {
     Foo foo;
     takesRef(foo.create.i);  // oops
 }

 struct Foo {
     int i;
     Foo create() {
         return Foo();
     }
 }

 void takesRef(ref int x) { }
 ---------------

 Even more fun with auto ref:

 https://issues.dlang.org/show_bug.cgi?id=19507

Per https://p0nce.github.io/d-idioms/

mixin template RvalueRef()
{
     alias T = typeof(this); // typeof(this) get us the type we're 
in
     static assert (is(T == struct));

      nogc  safe
     ref T byRef() pure nothrow return
     {
         return this;
     }
}

struct A {

    mixin RvalueRef;
}


void foo( ref A a );

foo( A().byRef ); // ok

The whole argument against 1016 is that if it were implemented 
you couldn't distinguish between rvalues and lvalues. But that is 
already the case, same with your example.

This code just adds a unneeded bloat, including the extra mixin 
for boilerplate code:

     void foo( ref Vector2 );

     foo( (a + b * c).byRef );

Why can't we just do:

     foo( a + b * c );