Templates and Specialisation error

[Simon]

Hi,

I'm having a problem with templates and specialisation. I'm using it to
overload the same function so it can return different things. I can't see
what I'm doing wrong, although my compiler promises me I am! Here follows an
example of my code, and i've included the error VC is giving me if that
helps. Can anyone point me in the right direction?
Cheers,
Simon ;o)
--
template <class A, class B>
A foo(B x);

template <>
int foo(int x) {
return x + 1;
}

template <>
float foo(int x) {
return (float)(x) + 1.0f;
}

int main () {
int a = 1;
int b = foo<int,int>(a);
float c = foo<float,int>(a);
return 1;
}

VC6 is giving...

c:\test\main.cpp(13) : error C2556: 'float __cdecl foo(int)' : overloaded
function differs only by return type from 'int __cdecl foo(int)'
c:\test\main.cpp(8) : see declaration of 'foo'
c:\test\main.cpp(13) : error C2371: 'foo' : redefinition; different basic
types
c:\test\main.cpp(8) : see declaration of 'foo'
c:\test\main.cpp(21) : error C2893: Failed to specialize function template
'A __cdecl foo(B)'
With the following template arguments:
'int'
'int'
Error executing cl.exe.

[Mike Wahler]

"Simon" <so*********@no.no> wrote in message
news:bk**********@titan.btinternet.com...

Hi,

I'm having a problem with templates and specialisation. I'm using it to
overload the same function so it can return different things. I can't see
what I'm doing wrong, although my compiler promises me I am! Here follows an example of my code, and i've included the error VC is giving me if that
helps. Can anyone point me in the right direction?
Cheers,
Simon ;o)
--
template <class A, class B>
A foo(B x);

template <>
int foo(int x) {
return x + 1;
}

template <>
float foo(int x) {
Whoops! You cannot overload on return type alone.
See below.
return (float)(x) + 1.0f;
}

int main () {
int a = 1;
int b = foo<int,int>(a);
float c = foo<float,int>(a);
return 1;
}

VC6 is giving...

c:\test\main.cpp(13) : error C2556: 'float __cdecl foo(int)' : overloaded
function differs only by return type from 'int __cdecl foo(int)'
This is an accurate diagnosis. Functinos cannot be
overloaded on return type only. The parameter types
and/or count must differ.
c:\test\main.cpp(8) : see declaration of 'foo'
c:\test\main.cpp(13) : error C2371: 'foo' : redefinition; different basic
types
You're trying to define a function more than
once, with a different return type. Not allowed.
The parameter list must differ. Return type does
not affect overloading.
c:\test\main.cpp(8) : see declaration of 'foo'
c:\test\main.cpp(21) : error C2893: Failed to specialize function template
'A __cdecl foo(B)'
With the following template arguments:
'int'
'int'
Error executing cl.exe.

-Mike

[Gianni Mariani]

Simon wrote:

Hi,

I'm having a problem with templates and specialisation. I'm using it to
overload the same function so it can return different things. I can't see
what I'm doing wrong, although my compiler promises me I am! Here follows an
example of my code, and i've included the error VC is giving me if that
helps. Can anyone point me in the right direction?
Cheers,
Simon ;o)

VC6 is of a vintage that has alot of problems with templates.

Get VC7.1

[WW]

Gianni Mariani wrote:

Simon wrote:

Hi,

I'm having a problem with templates and specialisation. I'm using it
to overload the same function so it can return different things. I
can't see what I'm doing wrong, although my compiler promises me I
am! Here follows an example of my code, and i've included the error
VC is giving me if that helps. Can anyone point me in the right
direction?
Cheers,
Simon ;o)

VC6 is of a vintage that has alot of problems with templates.

Get VC7.1

Other than that can you tell what rule does it get wrong in this case? Or
just guessing?

--
WW aka Attila

[Mike Wahler]

"Gianni Mariani" <gi*******@mariani.ws> wrote in message
news:bk********@dispatch.concentric.net...

Simon wrote:

Hi,

I'm having a problem with templates and specialisation. I'm using it to
overload the same function so it can return different things. I can't see what I'm doing wrong, although my compiler promises me I am! Here follows an example of my code, and i've included the error VC is giving me if that
helps. Can anyone point me in the right direction?
Cheers,
Simon ;o)
VC6 is of a vintage that has alot of problems with templates.

Not in this case. The code is invalid.


Get VC7.1

That won't help (unless they broke it so it
accepts the invalid code -- which isn't much of
a solution. :-))

-Mike

[WW]

Mike Wahler wrote:

VC6 is of a vintage that has alot of problems with templates.

Not in this case. The code is invalid.

Get VC7.1

That won't help (unless they broke it so it
accepts the invalid code -- which isn't much of
a solution. :-))

I have copy pasted the code into Comeau online and it did not complain. Did
I do something wrong? IMHO those might not be overloads (at least AIR they
aren't) but full specializations. I might of course be wrong. That is why
I have asked for standard text.

--
WW aka Attila

[Gianni Mariani]

Mike Wahler wrote:

"Gianni Mariani" <gi*******@mariani.ws> wrote in message
news:bk********@dispatch.concentric.net...

....

VC6 is of a vintage that has alot of problems with templates.

Not in this case. The code is invalid.

Then gcc 3.3.1 is broken. It compiled it without complaint.

So who is right ?

[WW]

Gianni Mariani wrote:

Mike Wahler wrote:

"Gianni Mariani" <gi*******@mariani.ws> wrote in message
news:bk********@dispatch.concentric.net...

...

VC6 is of a vintage that has alot of problems with templates.

Not in this case. The code is invalid.

Then gcc 3.3.1 is broken. It compiled it without complaint.

So who is right ?

The standard. That is why I have asked for it. IMO the guessing you did is
very dangerous. You have stated something and you do not really know if it
is true or not. :-(

I _think_ that if you fully specialize a template it does _not_ become an
overload. Just think about name mangling (which is an implementation
detail, but helps understanding). The mangled name of a function template
specialization will contain all the names (or identifiers or whatevers) of
the types used to specialize it. Including the dependent return type. Now
IIRC MS 6 does it worng. It does not put into the mangled name types which
are not part of the argument list. MS has tried there to mix overload
resolution with templates. Which did not work:

template <class T> void f() {
#1
}
f<int>();
f<std::string>();

Both will call f<int> or f<std::string>, depending on the current phase of
the moon and some other factors.

So it is highly likely that in this case MS is wrong and all the others are
right. But... which part of the standard does it say so? Or is this a grey
area? (I doubt it).

--
WW aka Attila

[Jerry Coffin]

In article <bk**********@titan.btinternet.com>, so*********@no.no
says...

Hi,

I'm having a problem with templates and specialisation. I'm using it to
overload the same function so it can return different things.
You're not doing overloading -- you're doing explicit specialization.
If you tried to do this via overloading, it wouldn't be allowed. You
can't create overloads that differ only in return type, but you CAN
create explicit specializations that do so (though when you do so, the
compiler generally will NOT be able to deduce the type of the parameter
specifying the function's return type).
I can't see
what I'm doing wrong, although my compiler promises me I am! Here follows an
example of my code, and i've included the error VC is giving me if that
helps. Can anyone point me in the right direction?

I believe the code is well-formed, and _should_ be accepted by a
properly functioning compiler. I'm not surprised that quite a few
compilers reject it though, and quite frankly, I'd tend to do the same.
IMO, if you use this, confusion is inevitable while utility is
questionable.

--
Later,
Jerry.

The universe is a figment of its own imagination.

[Simon]

> > I'm having a problem with templates and specialisation. I'm using it to

overload the same function so it can return different things.
You're not doing overloading -- you're doing explicit specialization.
If you tried to do this via overloading, it wouldn't be allowed. You
can't create overloads that differ only in return type, but you CAN
create explicit specializations that do so.

Yeah, sorry. I shouldn't really have used the word overload i guess. I'm
trying to use templates for exactly the reason you suggest, it just is meant
to feel like overloading. e.g. cast<template specialisation to>(overloaded
types from)
(though when you do so, the
compiler generally will NOT be able to deduce the type of the parameter
specifying the function's return type).

I don't understand this comment. Could you expand.

I can't see
what I'm doing wrong, although my compiler promises me I am! Here follows an example of my code, and i've included the error VC is giving me if that
helps. Can anyone point me in the right direction?

I believe the code is well-formed, and _should_ be accepted by a
properly functioning compiler.

So can we confirm that the original code was correct. There seem to be a
number of people saying it is incorrect, and if so I want to understand
where my misunderstanding of C++/templates lies.

Thanks for all the help so far,
Simon ;o)

[tom_usenet]

On Thu, 25 Sep 2003 22:53:22 GMT, "Mike Wahler"
<mk******@mkwahler.net> wrote:

"Gianni Mariani" <gi*******@mariani.ws> wrote in message
news:bk********@dispatch.concentric.net...

Simon wrote:

> Hi,
>
> I'm having a problem with templates and specialisation. I'm using it to
> overload the same function so it can return different things. I can'tsee > what I'm doing wrong, although my compiler promises me I am! Herefollows an > example of my code, and i've included the error VC is giving me if that
> helps. Can anyone point me in the right direction?
> Cheers,
> Simon ;o)

VC6 is of a vintage that has alot of problems with templates.

Not in this case. The code is invalid.

It is perfectly valid. It uses complete specialization of function
templates, not overloading as you stated in your other post.

Primary template:
template <class A, class B>
A foo(B x);

Specialization foo<int, int>:
template <>
int foo(int x) {
return x + 1;
}

Specialization foo<float, int>:
template <>
float foo(int x) {
return (float)(x) + 1.0f;
}

The specializations would be better written as:

template <>
int foo<int, int>(int x);
and
template <>
float foo<float, int>(int x);

to make it clearer which primary template they are specializing
(although it is obvious in the original code).

Tom

[Jerry Coffin]

In article <bl**********@hercules.btinternet.com>, so*********@no.no
says...

[ ... ]

(though when you do so, the
compiler generally will NOT be able to deduce the type of the parameter
specifying the function's return type).

I don't understand this comment. Could you expand.

Sure -- typically with a templated function, the compiler can figure out
the type(s) to instantiate the function over from the types of the
arguments. E.g. with something like:

template<class T>
T increment(T const &t) {
return t+1;
}

I can simply write:

x = increment(2);

and the compiler figures out that since 2 is an int that it needs to
instantiate increment over type int.

With your function the compiler can't do that, so you always have to
explicitly specify the template parameter for the return type.

[ ... ]

I believe the code is well-formed, and _should_ be accepted by a
properly functioning compiler.

So can we confirm that the original code was correct. There seem to be a
number of people saying it is incorrect, and if so I want to understand
where my misunderstanding of C++/templates lies.

Well, I don't know whether it qualifies as a confirmation or not, but
looking at things in the bright light of the morning, I'm still of the
same opinion -- I think the code is well-formed.

OTOH, the "bright light of morning" means the sun is shining in my eyes
right now and I can't see the screen as well as usual at the moment, so
that may not mean all that much. <G>

--
Later,
Jerry.

The universe is a figment of its own imagination.

[Simon]

> > > I believe the code is well-formed, and _should_ be accepted by a

properly functioning compiler.

So can we confirm that the original code was correct. There seem to be a
number of people saying it is incorrect, and if so I want to understand
where my misunderstanding of C++/templates lies.

Well, I don't know whether it qualifies as a confirmation or not, but
looking at things in the bright light of the morning, I'm still of the
same opinion -- I think the code is well-formed.

OTOH, the "bright light of morning" means the sun is shining in my eyes
right now and I can't see the screen as well as usual at the moment, so
that may not mean all that much. <G>

I tried it in g++ and it worked fine, so I think this is a closed case.
MSVC6 seems very broken with regard to templates! I've had it doing some
other wierd stuff with them too.

Thanks for all your help ;o)