IFTI with value / template value parameter shadowing

[Mathias LANG]

On Sunday, 23 February 2014 at 11:49:26 UTC, Marc Schütz wrote:
> There's no IFTI here. IFTI is about inferring a template 
> argument (= type) from the type of a function argument. Your 
> template argument is a value, not a type, so there's nothing to 
> infer.
>
There is the value. In IFTI_Type, the compiler understands that 
the 'T' in the parameter list is the same as the 'T' in the 
TemplateArgumentList.

For example this code compiles (from bug #4986):
-----
struct Foo(int n, int m) {
    void bar(int m2)(Foo!(m2, n) arg) {}
}

void main( )( ) {
    Foo!(3,2) f;
    Foo!(3,3) g;
    f.bar(g);
}
-----
So the compiler can deduce a value argument from a type, but not 
from a value known at compile time.


> Not sure what you want to achieve. Do you want IFTI_Value(6) to 
> be instantiated as IFTI_Value!6(6)? In this case, just leave 
> the runtime parameter out:

Yeah that works, and actually I agree that passing a value both 
in the parameters and template argument seems silly / odd at 
first sight. But there can be valid reasons to do so. I came 
accross this example while trying to shift a runtime argument to 
compile time while keeping source-level compatibility in a 
library.


>> In addition it looks like template parameter are not 
>> considered while looking if a symbol with shadow another one. 
>> I didn't find anything on the bugtracker but 
>> (this)[https://d.puremagic.com/issues/show_bug.cgi?id=6980], 
>> but it's only related.
>
> I believe it works exactly as intended. The short form for 
> template functions is just syntactic sugar for:
>
> template IFTI_Value(int n) {
>     int IFTI_Value(int n) { return n; }
> }
>
> This means that the function (runtime) parameter n is declared 
> in the inner scope, and is thus expected to shadow the template 
> parameter n in the outer scope.

I have to give it a second read but IIRC the TDPL was pretty much 
saying that shadowing is not legal in D.