www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Getting around forward references in templates?

reply Tristam MacDonald <swiftcoder gmail.com> writes:
So I have a setup like the following (simplified) example:

module A;

import B;

struct A(T)
{
	T x, y;
	
	void set(B o) {x = cast(T)o.a; y = cast(T)o.b;}
}

alias A!(float) Af;

//=========================

module B;

import A;

struct B
{
	float a, b;
	
	static B opCall(float x, float y) {B t; t.a = x; t.b = y; return t;}
	
	void set(Af o) {a = o.x; b = o.y;}
}

//=============================

import A;
import B;

int main()
{
	B b = B(1.0, 2.0);
	
	Af a;
	
	a.set(b);
	
	return 0;
}

//------------------------------

Now the compiler says:

B.d:12: Error: forward reference to 'A!(float)'
B.d:12: Error: Af is used as a type
B.d:12: Error: cannot have parameter of type void
B.d:12: Error: forward reference to 'A!(float)'
B.d:12: Error: Af is used as a type
B.d:12: Error: cannot have parameter of type void

Obviously this is a contrived example, but it is not uncommon to have a module
containing a template which must import itself in a circular manner. Is there
any way (easy or not) to get around this, other than passing a pointer to
float* into A.set(B)?
May 02 2007
parent reply Tristam MacDonald <swiftcoder gmail.com> writes:
So am I correct that it is impossible to forward reference a template
instantiation? And if so, why?
May 04 2007
parent reply "Stewart Gordon" <smjg_1998 yahoo.com> writes:
"Tristam MacDonald" <swiftcoder gmail.com> wrote in message 
news:f1f6ac$2uvc$1 digitalmars.com...

 So am I correct that it is impossible to forward reference a template 
 instantiation? And if so, why?


Because the compiler has bugs in it.

http://d.puremagic.com/issues/show_bug.cgi?id=810

Generally, if you ever see the phrase "forward reference" in a compiler 
message, chances are it's a bug.

http://d.puremagic.com/issues/show_bug.cgi?id=340

Stewart.
May 07 2007
parent Tristam MacDonald <swiftcoder gmail.com> writes:
Interesting... So since I am using gdc, both compilers exhibit this bug. More
to the point, has anyone come up with ways of dealing with this? From what I
have tried, no source file containing a template specialisation can be imported
by the file containing the template itself - even if the first file doesn't use
the template.

Worth noting is that if I put both structs from my original example in the same
file, the forward references work fine! So why are imports different?

A.d:

module A;

struct As(T)
{
	T x, y;
	
	void set(Bs o) {x = cast(T)o.a; y = cast(T)o.b;}
}

alias As!(float) Af;

struct Bs
{
	float a, b;
	
	static Bs opCall(float x, float y) {Bs t; t.a = x; t.b = y; return t;}
	
	void set(Af o) {a = o.x; b = o.y;}
}
 // No errors and compiles fine???

Stewart Gordon Wrote:

 "Tristam MacDonald" <swiftcoder gmail.com> wrote in message 
 news:f1f6ac$2uvc$1 digitalmars.com...

 So am I correct that it is impossible to forward reference a template 
 instantiation? And if so, why?


 
 Because the compiler has bugs in it.
 
 http://d.puremagic.com/issues/show_bug.cgi?id=810
 
 Generally, if you ever see the phrase "forward reference" in a compiler 
 message, chances are it's a bug.
 
 http://d.puremagic.com/issues/show_bug.cgi?id=340
 
 Stewart.
May 07 2007