I think aliases should be expanded to allow for arguments, and allow constant expressions. The syntax that I used could be added to the language without breaking any existing code.

-- Arguments
Like with functions and classes, the arguments could simply add a template to the alias.
alias T* ptr(T);
would be the same as writing:
template ptr(T){ alias T* ptr; }

And they would be instantiated just like templates:
ptr!(int)


-- Expressions
alias 10 base;
alias (10*20) val;

Combine with arguments:
alias (a*b) mul(a, b);


I'm sure these ideas have been posted before, but I think they could really be useful. For example, I once wrote an encryption algorithm which would encrypt in 128k blocks (131072 bytes). I could have defined an alias to perform the basic math for this:
alias (n * 131072) SizeofBlocks(n);
SizeofBlocks!(8);


To accomplish what alias expressions could do, you would have to write all of this:
template mul(V...){
	const typeof(V[0] * V[1]) mul = V[0] * V[1];
}

Compared to:
alias (a*b) mul(a, b);

The syntax for aliases would be simpler and quicker. Plus, you wouldn't have to guess the base type of the expression.

Xinok wrote:
> To accomplish what alias expressions could do, you would have to write all of this:
> template mul(V...){
> 	const typeof(V[0] * V[1]) mul = V[0] * V[1];
> }
> 
> Compared to:
> alias (a*b) mul(a, b);

That typeof(...) is redundant. Remove that and it should default to automatic type deduction, which would do exactly the same.

Frits van Bommel Wrote:

> Xinok wrote:
> > To accomplish what alias expressions could do, you would have to write all of this:
> > template mul(V...){
> > 	const typeof(V[0] * V[1]) mul = V[0] * V[1];
> > }
> > 
> > Compared to:
> > alias (a*b) mul(a, b);
> 
> That typeof(...) is redundant. Remove that and it should default to automatic type deduction, which would do exactly the same.

Regardless, I think the extension to aliases would be a nice addition to D.

template mul(V...){ const auto mul = V[0] * V[1]; }
alias (a*b) mul(a, b);

One of the problems with the template implementation is it uses a tuple. First off of course is the V[] syntax. Sure you could define aliases or constants for the arguments, but this only further pushes the problem:

template mul(V...){
	// This currently doesn't work due to bugs in the compiler:
	// alias V[0] a; alias V[1] b;

	// D doesn't allow this with the 'auto' type:
	// const auto a = V[0], b = V[1];

	// This is the best you can do:
	const auto a = V[0]; const auto b = V[1];

	// So you can create symbols for the tuple arguments:
	const auto mul = a * b;
}

void main(){
	// Because 'a' and 'b' are members of the template, the compiler can't use the template implicitly:
	// writefln(mul!(15, 30));
	// Defining 'a' and 'b' as private doesn't work either

	// So you have to manually specify the member 'mul':
	writefln(mul!(15, 30).mul);
}


The second problem is tuples aren't self-documenting. The user sees (V...), they'll have no idea what arguments they're supposed to pass. Alias arguments on the other hand can be named, giving the user a small idea what arguments must be provided for the expression.

Xinok wrote:
> Frits van Bommel Wrote:
> 
>> Xinok wrote:
>>> To accomplish what alias expressions could do, you would have to write all of this:
>>> template mul(V...){
>>> 	const typeof(V[0] * V[1]) mul = V[0] * V[1];
>>> }
>>>
>>> Compared to:
>>> alias (a*b) mul(a, b);
>> That typeof(...) is redundant. Remove that and it should default to automatic type deduction, which would do exactly the same.
> 
> Regardless, I think the extension to aliases would be a nice addition to D.
> 
> template mul(V...){ const auto mul = V[0] * V[1]; }
> alias (a*b) mul(a, b);
> 
> One of the problems with the template implementation is it uses a tuple. First off of course is the V[] syntax. Sure you could define aliases or constants for the arguments, but this only further pushes the problem:
> 
> template mul(V...){
> 	// This currently doesn't work due to bugs in the compiler:
> 	// alias V[0] a; alias V[1] b;
> 	
> 	// D doesn't allow this with the 'auto' type:
> 	// const auto a = V[0], b = V[1];
> 	
> 	// This is the best you can do:
> 	const auto a = V[0]; const auto b = V[1];
> 	
> 	// So you can create symbols for the tuple arguments:
> 	const auto mul = a * b;
> }
> 
> void main(){
> 	// Because 'a' and 'b' are members of the template, the compiler can't use the template implicitly:
> 	// writefln(mul!(15, 30));
> 	// Defining 'a' and 'b' as private doesn't work either
> 	
> 	// So you have to manually specify the member 'mul':
> 	writefln(mul!(15, 30).mul);
> }
> 
> 
> The second problem is tuples aren't self-documenting. The user sees (V...), they'll have no idea what arguments they're supposed to pass. Alias arguments on the other hand can be named, giving the user a small idea what arguments must be provided for the expression.

First, I really don't think its a completely bad idea, although I might prefer the template parameters being on the alias keyword, even if it is a change from how its done with classes, structs, etc.  Something like:
alias(a, b) (a * b) mul;

Second, I don't really comprehend what you are trying to do with the template above... it would just be:
# template mul (a, b) { const mul = a * b; }
#
# void main () {
#   writefln(mul!(15, 30));
# }

Which, IMHO, isn't so bad.  (Note that 'auto' in this case isn't needed for type inferance, as any storage class without a specified type defaults to auto.)

-- Chris Nicholson-Sauls

> Second, I don't really comprehend what you are trying to do with the template above... it
> would just be:
> # template mul (a, b) { const mul = a * b; }
> #
> # void main () {
> #   writefln(mul!(15, 30));
> # }

a and b are typenames, not variables. You can't use aliases either since they don't accept variables: template mul(alias a, alias b)

The only thing that will work is a tuple:
template mul(V...){ const mul = V[0] * V[1]; }
And for a complex expression, that isn't very pleasing to the eye.


You were right about not needing 'auto'. I didn't think D allowed that though.

Xinok wrote:
>> Second, I don't really comprehend what you are trying to do with the template above... it would just be:
>> # template mul (a, b) { const mul = a * b; }
>> #
>> # void main () {
>> #   writefln(mul!(15, 30));
>> # }
> 
> a and b are typenames, not variables. You can't use aliases either since they don't accept variables:
> template mul(alias a, alias b)
> 
> The only thing that will work is a tuple:
> template mul(V...){ const mul = V[0] * V[1]; }
> And for a complex expression, that isn't very pleasing to the eye.
> 
> 
> You were right about not needing 'auto'. I didn't think D allowed that though.

Ah okay, didn't realize a and b were supposed to be types.  (What I get for reading just before bed.)

-- Chris Nicholson-Sauls