boxing, struct opAssign and constructors

[Sergey Gromov]

Sat, 14 Feb 2009 15:25:23 +0200, gena wrote:

> struct Value {
>    Value opAssign(double arg){...}
> }
> 
> Value v1, v2;
> v1 = v2;
> 
> With D2 (at least 014 and 023) I get compilation error:
> function Value.opAssign (double arg) does not match parameter types (Value)

The D2 specification seems inconsistent.  The operator overloading
states,

| The assignment operator = can be overloaded if the lvalue is a struct
| aggregate, and opAssign is a member function of that aggregate.  The
| assignment operator cannot be overloaded for rvalues that can be
| implicitly cast to the lvalue type.

http://www.digitalmars.com/d/2.0/operatoroverloading.html#Assignment

However, the Structs documentation says otherwise:

| While the compiler will generate a default opAssign as needed, a
| user-defined one can be supplied. The user-defined one must still
| implement the equivalent semantics, but can be more efficient.

http://www.digitalmars.com/d/2.0/struct.html#AssignOverload

Even considering the latter, the specs are still promising to provide a
default opAssign so that v1 = v2 should still work.

I think it's a bug, either in the documentation, in compiler
implementation, or both.

> -------------------------------
> having the above struct definition,
> 
> Value v = 10
> 
> gives compilation error:
> Error: cannot cast int to Value[]
> 
> why no line number in error?
> Where does [] assumption come from?
> 
> whereas
> 
> Value v = 10.0
> gives
> test.d(25): Error: cannot cast double to Value
> 
> and finally
> 
> Value v; v = 10;
> Value v; v = 10.0;
> 
> do compile and work the same.
> 
> NOTE: there is NO opAssign for int, only for double and Value.

I have no idea what's going on here.  Note though that struct specs
mentioned above say that copy-construction is automatically taken care
about, and only assignment semantics can be overloaded to provide for
more efficient implementations.

> -------------------------------
> I couldn't find other way of calling struct constructors except using 
> new operator, shouldn't there be a way to do the same on stack? Say
> 
> struct Value {
>    this(double arg) {...}
> }
> 
> Value v(10)
> AND/OR maybe the above
> Value v = 10
> to be interpreted as calling the constructor?
> 
> right now only following initializer seems to work:
> Value v; v = 10;
> But It's ugly and I tend to question the efficiency of the produced code

The canonical way seems to be

auto v = Value(10);