Issue 11706 - Add a TypedefType trait to extract the underlying type of a std.typecons.Typedef

[ghost]

https://issues.dlang.org/show_bug.cgi?id=11706

Thanks to some compiler improvements this can be entirely implemented outside the Typedef definition.

[monarchdodra]

I'm unsure about this change to OriginalType. I think code that uses it assumes enum now, and have forgotten about typedefs. Sure, it originally supported typedef, but that has been deprecated and removed. I'm unsure

In particular, an enum is implicitly castable to the OriginalType. In contrast, Typedef is NOT. By design. The entire idea is to create a brand new type that isn't the original type.

I do agree a trait should be available, but I'm not sure it should be packaged into OriginalType.

... Well, that's my "Shields Up!" reaction anyways.

[monarchdodra]

I didn't quite check the failing unittest, but the fact they are failing seems like an indication this isn't exactly a trivial change.

[ghost]

Yeah, I don't quite understand what's going on with the failures. This code works fine in a simplified test-case. I'll investigate.

[ghost]

I do agree a trait should be available, but I'm not sure it should be packaged into OriginalType.

What's your suggestion for a name. Maybe TypedefType or something?

[ghost]

Updated by implementing TypedefType rather than tweaking OriginalType, which caused issues for some reason.

[monarchdodra]

Thanks to some compiler improvements this can be entirely implemented outside the Typedef definition.

Nice. Very nice. We need to do the same with Rebindable. I hate that a lot of functions in traits have to import std.typecons just because maybe the current object is a Rebindable...

[monarchdodra]

Shouldn't this accept any type, and be a no-op for non-Typedefs? It makes generic code much easier. I've "abused" this for OriginalType, that's for sure. It drastically reduces the "condtional bloat":

alias TT = TypedefType!T; //Is T a typedef? Don't care.
static if (is(TT == class)
    ...
else if (is(TT == struct)
   ...
...
}

[ghost]

no-op for non-Typedefs

No-op? You mean alias itself to the passed-in type?

[ghost]

Updated with no-op support, with an even cleaner implementation.

[monarchdodra]

Should I be concerned about:

alias T = TypedefType!(Typedef!int, 75, int, 5, ubyte);

Does that even behave according to spec?

[ghost]

Does that even behave according to spec?

I'll fix this up. But it looks like TemplateArgsOf will also have to be fixed.

[ghost]

Filed as Issue 12651. It might be a compiler bug, CC'ed Kenji.

[DmitryOlshansky]

The linked bug is fixed, so ... is it good to go?

[mihails-strasuns]

Should match exact Typedef template parameter list.

[ghost]

Oh you mean = Args rather than = Args[0]? I can't remember why I used [0] here, should be fixed.

[mihails-strasuns]

I mean that Typedef takes fixed amount of template arguments, not any variadic one.

[ghost]

Ah I thought I had to use the ... syntax.

[mihails-strasuns]

No, it is generic pattern matching thing:

struct ZZZ (T1, T2) { }

alias Z = ZZZ!(int, double);

static if (is(Z == ZZZ!(int, T), T))
{
    pragma(msg, T);
}

Anything that can be in template parameter list can go after comma

[mihails-strasuns]

Ping @AndrejMitrovic

[ghost]

Updated: Let's see if it passes the tester.

[mihails-strasuns]

Please add test case for typedef with custom initializer (i.e. Typedef!(int, 42))

[ghost]

Updated.

[mihails-strasuns]

Unrelated test failure?

[ghost]

Yeah I think so.

[mihails-strasuns]

Ok, I'll mark it for merging, will see if it eventually gets to a green state on its own.

[mihails-strasuns]

Auto-merge toggled on