import std.stdio;
class A
{
void f(){ writeln("A.f"); }
}
class B: A
{
alias f = A.f;
mixin X!();
}
mixin template X()
{
override void f(){ writeln("B.f"); super.f(); }
}
class C: B
{
override void f(){ writeln("C.f"); super.f(); }
}
void main()
{
(new C).f();
}
The above code prints the following:
C.f
A.f
Instead the following would be expected:
C.f
B.f
A.f
This affects DMD as a library, because ParseTimeTransitiveVisitor uses a mixin with overloads for visit.
Comment #1 by destructionator — 2022-03-12T12:52:55Z
The presence of that alias complicates things. A template mixin work by name and thus doesn't bring in something that's already there. So with the mixin, `f` is already there, meaning the mixin template's one is overridden by that. That `B.f` function is never part of the virtual tree at all, so `super` here isn't wrong.
I expect the alias is there in the real thing because of overloading arguments in the real thing right?
What you actually want to do is to alias BOTH sets together, not just one:
```
class B: A
{
alias f = A.f;
alias f = x.f;
mixin X!() x;
}
```
This combines the overloads of A with the overloads of X, thus allowing the override to take effect.
This should probably be closed as not a bug, since it is working as designed...
Comment #2 by tim.dlang — 2022-03-12T13:39:11Z
(In reply to Adam D. Ruppe from comment #1)
> I expect the alias is there in the real thing because of overloading
> arguments in the real thing right?
Yes, the class ParseTimeTransitiveVisitor overrides some overloads using a template mixins and uses an alias for other overloads: https://github.com/dlang/dmd/blob/cbba5f41a32cfed7f22a213d537f8e2dee0b92f7/src/dmd/transitivevisitor.d#L17-L21
>
> What you actually want to do is to alias BOTH sets together, not just one:
>
> ```
> class B: A
> {
> alias f = A.f;
> alias f = x.f;
> mixin X!() x;
> }
> ```
>
> This combines the overloads of A with the overloads of X, thus allowing the
> override to take effect.
Thanks, this works.
>
> This should probably be closed as not a bug, since it is working as
> designed...
I have now seen, that the same problem was already fixed with the same workaround for SemanticTimeTransitiveVisitor: https://github.com/dlang/dmd/pull/7635
I will close this as a duplicate of issue 18132, which was referenced in the pull request.
Comment #3 by tim.dlang — 2022-03-12T13:39:34Z
*** This issue has been marked as a duplicate of issue 18132 ***