'in' storage class

[torhu]

Funog wrote:
> Robert Fraser Wrote:
> 
>> Funog Wrote:
>> 
>> > Nathan Reed Wrote:
>> > 
>> > > Funog wrote:
>> > > > void testFinal(final int[] foo)
>> > > > {
>> > > >     foo[0] = 10;        //OK
>> > > >     foo = new int[20];  //ERROR
>> > > > }
>> > > > void testConst(const int[] foo)
>> > > > {
>> > > >     foo[0] = 10;        //ERROR
>> > > >     foo = new int[20];  //ERROR
>> > > > }
>> > > > void testFinalConst(final const int[] foo)
>> > > > {
>> > > >     foo[0] = 10;        //ERROR
>> > > >     foo = new int[20];  //ERROR
>> > > > }
>> > > > 
>> > > > 
>> > > > So what is the difference between 'const' and 'final const' ?
>> > > > 
>> > > 
>> > > I suppose there's no practical difference in that case since you're 
>> > > using the transitive const.  Note that final const(int)[] foo is 
>> > > different from const(int)[] foo though.
>> > > 
>> > > Thanks,
>> > > Nathan Reed
>> > 
>> > 
>> > 
>> > I agree... But then, what is the point of having 'in' equivalent to 'final const scope' rather than just 'const scope' ?
>> > 
>> 
>> The difference is easier for me to grasp when talking about class references than arrays.
>> 
>> class Foo { int bar; }
>> 
>> final Foo baz;
>> baz.bar = 5; // Okay
>> baz = quux; // Illegal
>> 
>> const Foo quux; // Equivilent, I _think_ to const(Foo) quux;
>> quux.bar = 5; // Illegal
>> quux = baz; //Okay
>> 
> 
> 
> It isn't... Both lines are illegal. quux = baz is legal for const(Foo) quux.
> If nobody can tell the difference between "const scope" and "in" ( = final const scope ), should we report it as a bug ? xD
> 

I tried this:

---
void f(const int* p)
{
	int y = 9;
	
	*p = 7;	 // line 5
	p = &y;  // line 6
}

void main()
{
	int x = 5;

	f(&x);
}
---
dmd -run inargs
inargs.d(5): Error: *p is not mutable
inargs.d(6): variable inargs.f.p cannot modify final/const/invariant 
variable 'p'


Adding 'final' doesn't change anything.  So it would seem that this 
feature isn't finalized yet.  'scope' doesn't seem to work either.  If I 
use 'final const(int)*' instead, I get the same errors.

Barring optimizations, a function's arguments are not known a 
compile-time.  So I don't know if giving them the const or invariant 
storage class makes much sense.  Maybe the idea was that const in a 
parameter list would always mean const as a type constructor.  For other 
declarations, 'final const(int)*' allows run-time (by a constructor) 
initialization, while 'const int*' does not.  So maybe the latter one 
should be disallowed in parameter lists.