While reading the source code of std.traits I came across a bunch of variadic templates enforcing the length of their parameter type tuple to be of 1.

Is there any semantic difference between
> template ReturnType(func...) if (func.length == 1 && isCallable!func)
and
> template ReturnType(alias func) if (isCallable!func)

If not, is there any reason why this syntax is used ?

Guillaume
--
The list of templates from std.traits ( https://github.com/D-Programming-Language/phobos/blob/master/std/traits.d ) matching this pattern :

template ReturnType(func...)
 if (func.length == 1 && isCallable!func)

template ParameterTypeTuple(func...)
 if (func.length == 1 && isCallable!func)

template ParameterStorageClassTuple(func...)
 if (func.length == 1 && isCallable!func)

template functionAttributes(func...)
 if (func.length == 1 && isCallable!func)

template functionLinkage(func...)
 if (func.length == 1 && isCallable!func)

template variadicFunctionStyle(func...)
 if (func.length == 1 && isCallable!func)

template FunctionTypeOf(func...)
 if (func.length == 1 && isCallable!func)

template isFunctionPointer(T...)
 if (T.length == 1)

template isDelegate(T...)
 if(T.length == 1)

template isSomeFunction(T...)
 if (T.length == 1)

template isCallable(T...)
 if (T.length == 1)

template isAbstractFunction(method...)
 if (method.length == 1)

template mangledName(sth...)
 if (sth.length == 1)

On 06/22/2012 11:50 PM, Guillaume Chatelet wrote:
> While reading the source code of std.traits I came across a bunch of
> variadic templates enforcing the length of their parameter type tuple to
> be of 1.
>
> Is there any semantic difference between
>> template ReturnType(func...) if (func.length == 1&&  isCallable!func)
> and
>> template ReturnType(alias func) if (isCallable!func)
>

Yes there is, but I have argued that this is a bug in the past. With DMD, the (func...) accepts built-in types, while the (alias func) does
not.


> If not, is there any reason why this syntax is used ?
>
> Guillaume
> --

On 06/23/12 01:05, Timon Gehr wrote:
> 
> Yes there is, but I have argued that this is a bug in the past. With
> DMD, the (func...) accepts built-in types, while the (alias func) does
> not.

Ok got it, the following won't compile I agree.

template isInt(alias T) {
	enum bool isInt = is(T==int);
}

void main(){
	static assert(isInt!int);
}

So I understand the following functions have to go the variadic way
because you want an answer at compile time and not a compiler error
template isFunctionPointer(T...)
template isDelegate(T...)
template isSomeFunction(T...)
template isCallable(T...)
template isAbstractFunction(method...)
template mangledName(sth...)

Nevertheless the following templates are all requiring 'func.length == 1 && isCallable!func'

template ReturnType(func...)
template ParameterTypeTuple(func...)
template ParameterStorageClassTuple(func...)
template functionAttributes(func...)
template functionLinkage(func...)
template variadicFunctionStyle(func...)
template FunctionTypeOf(func...)

So they won't instantiate if you pass a built-in type anyway. Using an
'alias' instead of a 'func...' would make it clear that :
* you can't pass a built-in type in
* the template is requiring exactly one argument

As a user, I find it odd that ReturnType takes a TypeTuple rather than
an alias. 'func...' basically means 'I will accept a tuple of functions'
which is then restricted in the if clause to the case where func.length==1.

--
Guillaume

On 2012-06-23 01:05, Timon Gehr wrote:

> Yes there is, but I have argued that this is a bug in the past. With
> DMD, the (func...) accepts built-in types, while the (alias func) does
> not.

Would it be possible to overload? Something like this:

template Foo (T) {}

template Foo (alias a) {}

-- 
/Jacob Carlborg

I'm reposting this because I'm not sure it made it to the NG ( it doesn't show up in the web interface )

On 06/23/12 01:05, Timon Gehr wrote:
> Yes there is, but I have argued that this is a bug in the past. With
> DMD, the (func...) accepts built-in types, while the (alias func) does
> not.

Ok got it, the following won't compile I agree.

template isInt(alias T) {
	enum bool isInt = is(T==int);
}

void main(){
	static assert(isInt!int);
}

So I understand the following functions have to go the variadic way
because you want an answer at compile time and not a compiler error
template isFunctionPointer(T...)
template isDelegate(T...)
template isSomeFunction(T...)
template isCallable(T...)
template isAbstractFunction(method...)
template mangledName(sth...)

Nevertheless the following templates are all requiring 'func.length == 1 && isCallable!func'

template ReturnType(func...)
template ParameterTypeTuple(func...)
template ParameterStorageClassTuple(func...)
template functionAttributes(func...)
template functionLinkage(func...)
template variadicFunctionStyle(func...)
template FunctionTypeOf(func...)

So they won't instantiate if you pass a built-in type anyway. Using an
'alias' instead of a 'func...' would make it clear that :
* you can't pass a built-in type in
* the template is requiring exactly one argument

As a user, I find it odd that ReturnType takes a TypeTuple rather than
an alias. 'func...' basically means 'I will accept a tuple of functions'
which is then restricted in the if clause to the case where func.length==1.

--
Guillaume

On 06/23/2012 12:52 PM, Jacob Carlborg wrote:
> On 2012-06-23 01:05, Timon Gehr wrote:
>
>> Yes there is, but I have argued that this is a bug in the past. With
>> DMD, the (func...) accepts built-in types, while the (alias func) does
>> not.
>
> Would it be possible to overload? Something like this:
>
> template Foo (T) {}
>
> template Foo (alias a) {}
>

alias should just accept built in types.

Le 23 juin 2012 12:53, "Jacob Carlborg" <doob@me.com> a écrit :
>
> On 2012-06-23 01:05, Timon Gehr wrote:
>
>> Yes there is, but I have argued that this is a bug in the past. With
>> DMD, the (func...) accepts built-in types, while the (alias func) does
>> not.
>
>
> Would it be possible to overload? Something like this:
>
> template Foo (T) {}
>
> template Foo (alias a) {}

I think it is. IIRC, I did something like this.

But then the trouble is with user-defined types: they are both a type and a symbol, so they may fire both templates.