www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - Unusual template specialisation behavior

reply Matthias Walter <walter mail.math.uni-magdeburg.de> writes:
Hello guys,

Here's a (minimal) code snippet for my problem:

| struct Expression (R, E) { }
| 
| struct ResolveReference (T)
| {
|     alias T reference_type;
| }
| 
| struct ResolveReference (T : Expression !(R, E))
| {
|     alias T* reference_type;
| }
| 
| void printRefType (T) ()
| {
|     Stdout (ResolveReference !(Expression !(T,
T)).reference_type.stringof).newline;
| }
| 
int main (char[][] args)
| {
|     printRefType !(uint);
|     printRefType !(int);
|     printRefType !(long);
|     printRefType !(char);
|     printRefType !(float);
| 
|     return 0;
| }

The output is:

| Expression!(uint,uint) 
| Expression!(int,int) *
| Expression!(long,long) 
| Expression!(char,char) 
| Expression!(float,float) 

If my understanding of template specialisation is correct, only the second call
instanciates the template "ResolveReference (T : Expression !(R, E))" and the
other ones choose "ResolveReference (T)". Why does this happen only for ints?

My intended behavior would be, that ResolveReference.reference_type is T*, if T
is an Expression and T if not. Thus, I expected, that this code returns 

| Expression!(uint,uint) *
| Expression!(int,int) *
| Expression!(long,long) * 
| Expression!(char,char) *
| Expression!(float,float) *

btw, this is tested on Linux, Tango rev 3152, Digital Mars D Compiler v1.021.
Hopefully, I did not miss a bug-fix.

best regards
Matthias Walter
Feb 08 2008
parent reply Christopher Wright <dhasenan gmail.com> writes:
Matthias Walter wrote:

 Hello guys,
 
 Here's a (minimal) code snippet for my problem:
 
 | struct Expression (R, E) { }
 | 
 | struct ResolveReference (T)
 | {
 |     alias T reference_type;
 | }
 | 
 | struct ResolveReference (T : Expression !(R, E))


The compiler should have caught this, I think. You have R and E, but 
they aren't defined. They default to int (which is why you saw that 
behavior), but not being defined, they should be a compile-time error.

The proper way to do this, unfortunately, involves static if rather than 
overloading. At least if you want your users to supply a single argument 
in all cases.
Feb 08 2008
next sibling parent reply Sergey Gromov <snake.scaly gmail.com> writes:
Christopher Wright <dhasenan gmail.com> wrote:

 Matthias Walter wrote:

 Hello guys,
 
 Here's a (minimal) code snippet for my problem:
 
 | struct Expression (R, E) { }
 | 
 | struct ResolveReference (T)
 | {
 |     alias T reference_type;
 | }
 | 
 | struct ResolveReference (T : Expression !(R, E))


 
 The compiler should have caught this, I think. You have R and E, but 
 they aren't defined. They default to int (which is why you saw that 
 behavior), but not being defined, they should be a compile-time error.
 
 The proper way to do this, unfortunately, involves static if rather than 
 overloading. At least if you want your users to supply a single argument 
 in all cases.


struct ResolveReference (T : Expression !(R, E), R, E)
{
    alias T* reference_type;
}

works just fine:

void main (char[][] args)
{
    printRefType !(uint);
    printRefType !(int);
    printRefType !(long);
    printRefType !(char);
    printRefType !(float);

    Stdout (ResolveReference !(uint).reference_type.stringof).newline;
}

Expression!(uint,uint)*
Expression!(int,int)*
Expression!(long,long)*
Expression!(char,char)*
Expression!(float,float)*
uint

-- 
SnakE
Feb 08 2008
parent Matthias Walter <walter mail.math.uni-magdeburg.de> writes:
Sergey Gromov Wrote:


 Christopher Wright <dhasenan gmail.com> wrote:

 Matthias Walter wrote:

 Hello guys,
 
 Here's a (minimal) code snippet for my problem:
 
 | struct Expression (R, E) { }
 | 
 | struct ResolveReference (T)
 | {
 |     alias T reference_type;
 | }
 | 
 | struct ResolveReference (T : Expression !(R, E))


 
 The compiler should have caught this, I think. You have R and E, but 
 they aren't defined. They default to int (which is why you saw that 
 behavior), but not being defined, they should be a compile-time error.
 
 The proper way to do this, unfortunately, involves static if rather than 
 overloading. At least if you want your users to supply a single argument 
 in all cases.


 
 struct ResolveReference (T : Expression !(R, E), R, E)
 {
     alias T* reference_type;
 }
 
 works just fine:
 
 void main (char[][] args)
 {
     printRefType !(uint);
     printRefType !(int);
     printRefType !(long);
     printRefType !(char);
     printRefType !(float);
 
     Stdout (ResolveReference !(uint).reference_type.stringof).newline;
 }
 
 Expression!(uint,uint)*
 Expression!(int,int)*
 Expression!(long,long)*
 Expression!(char,char)*
 Expression!(float,float)*
 uint
 
 -- 
 SnakE


Thank you both! I guess, this works, because when passing an Expression !(R,
E), the 2nd and 3rd specialisation parameters R and E are deduced and
everything works fine, right? Thanks for your help!

best regards
Matthias Walter
Feb 08 2008
prev sibling parent reply Matthias Walter <walter mail.math.uni-magdeburg.de> writes:
Christopher Wright Wrote:


 Matthias Walter wrote:

 Hello guys,
 
 Here's a (minimal) code snippet for my problem:
 
 | struct Expression (R, E) { }
 | 
 | struct ResolveReference (T)
 | {
 |     alias T reference_type;
 | }
 | 
 | struct ResolveReference (T : Expression !(R, E))


 
 The compiler should have caught this, I think. You have R and E, but 
 they aren't defined. They default to int (which is why you saw that 
 behavior), but not being defined, they should be a compile-time error.
 
 The proper way to do this, unfortunately, involves static if rather than 
 overloading. At least if you want your users to supply a single argument 
 in all cases.


Thanks for that hint. How would such a static if look like?

I currently have 

| static if (is (T == struct)).

This works for distinguishing native datatypes like int from my
Expression-Templates.
But is there a native way to statically check, whether T is some specialisation
like
 "T == Expression !(R, E)" (for all R, E).

best regards
Matthias Walter
Feb 08 2008
parent Christopher Wright <dhasenan gmail.com> writes:
Matthias Walter wrote:

 Christopher Wright Wrote:
 

 Matthias Walter wrote:

 Hello guys,

 Here's a (minimal) code snippet for my problem:

 | struct Expression (R, E) { }
 | 
 | struct ResolveReference (T)
 | {
 |     alias T reference_type;
 | }
 | 
 | struct ResolveReference (T : Expression !(R, E))


 The compiler should have caught this, I think. You have R and E, but 
 they aren't defined. They default to int (which is why you saw that 
 behavior), but not being defined, they should be a compile-time error.

 The proper way to do this, unfortunately, involves static if rather than 
 overloading. At least if you want your users to supply a single argument 
 in all cases.


 
 Thanks for that hint. How would such a static if look like?


template ResolveReference (T) {
    static if (is (T == Expression!(R, E))) {
    // or `static if (is (T == Expression!(R, E), R, E)) {`
       alias T* ResolveReference;
    } else {
       alias T ResolveReference;
    }
}

Except that doesn't work, right now. There's a bug listed for it.
Feb 08 2008