Create using-alias-types.md

[CyrusNajmabadi]

No description provided.

[jcouv]

Does type include top-level nullability annotation?
I assume this allows built-in types using Id = int;. Is that right?

[CyrusNajmabadi]

Does type include top-level nullability annotation?

Yes. It would.

I assume this allows built-in types using Id = int;. Is that right?

yes, it would.

[jcouv]

Yes. It would [include top-level nullability annotation].

From our offline discussion, this raises a question for how the three annotations (?, ! and ~/oblivious) interact in the declaration versus the use-site. For example, what happens if you put ? in the declaration and the use-site?

[cston]

See also #4452.

[CyrusNajmabadi]

See also #4452.

This is very intentionally not trying to support the case of using X<T> = ... While i think that is valuable for us to look into, it's orthogonal and much larger in scope (esp. on the impl side). Whereas the above proposal is actually quite trivial to support and hits a lot of pain cases we have today.

[333fred]

Spec looks good to me. Do we have a championed issue?

[tannergooding]

@CyrusNajmabadi, would this also work for other aliases?

That is, could I do the following?

using HANDLE = void*;
using HWND = HANDLE;

I was wanting to add a feature to the ClangSharp P/Invoke generator that generates global using directives for typedefs and I've found the following three issues that are blocking:

• Can't use keywords, which is particularly relevant since nint and IntPtr behave differently
• Can't use pointers (important for typedefs to pointers)
• Can't use other defined aliases (important for typedefs to typedefs)

[CyrusNajmabadi]

That is, could I do the following?

using HANDLE = void*;
using HWND = HANDLE;

The first is completely allowed. The second would not be allowed as a using directive cannot reference another using directive in the same block. However, this is ok:

using HANDLE = void*;
namespace X
{
  using HWND = HANDLE;
}

or

using HANDLE = void*;
using HWND = void*;

Can't use keywords, which is particularly relevant since nint and IntPtr behave differently

Keywords would be fine.

Can't use pointers (important for typedefs to pointers)

You will be able to use pointers.

Can't use other defined aliases (important for typedefs to typedefs)

If in the same scope this will continue to not work and you'll still have to explicitly write out the entire type signature again.

[tannergooding]

Keywords would be fine.
You will be able to use pointers.

Awesome.

If in the same scope this will continue to not work and you'll still have to explicitly write out the entire type signature again.

This is a bit unfortunate but also not overly complex to handle, thankfully.

[CyrusNajmabadi]

This is a bit unfortunate but also not overly complex to handle, thankfully.

Understood. Note that i'm not opposed to work here. But i would consider it more of a separate feature.

[tannergooding]

Yep definitely understand.

Its basically just traversing the typedefs to the "concrete" type for ClangSharp, I expect its a bit more complex for Roslyn to handle 😄