On 05/11/2017 12:39 PM, Andre Pany wrote:
in this example, both asserts fails. Is my assumption right, that UDA on
alias have no effect? If yes, I would like to see a compiler warning.

But anyway, I do not understand why the second assertion fails. Are UDAs
on arrays not allowed?

import std.traits: hasUDA;

enum Flattened;

struct Foo
{
    int bar;
}

@Flattened alias FooList = Foo[];

struct Baz
{
    FooList fooList1;
    @Flattened FooList[] fooList2;
}

void main()
{
    Baz baz;
    static assert(hasUDA!(baz.fooList1, "Flattened")); // => false
    static assert(hasUDA!(baz.fooList2, "Flattened")); // => false
}

1) You have to test against `Flattened`, not `"Flattened"`. A string is a valid UDA, but you're not using the string on the declarations.

When you fix this, the second assert passes.

2) `Baz.fooList1` doesn't have any attributes. Attributes apply to declarations. If it's valid, the attribute on `FooList` applies only to `FooList`. It doesn't transfer to `Baz.fooList1`.

If anything, you could assert that `hasUDA!(FooList, Flattened)` holds. Maybe you could, if it compiled.

3) Why does `hasUDA!(FooList, Flattened)` fail to compile?

The error message reads: "template instance hasUDA!(Foo[], Flattened) does not match template declaration hasUDA(alias symbol, alias attribute)".

We see that `FooList` has been replaced by `Foo[]`. It's clear then why the instantiation fails: `Foo[]` isn't a symbol.

Unfortunately, the spec is a bit muddy on this topic. On the one hand it says that "AliasDeclarations create a symbol", but it also says that "Aliased types are semantically identical to the types they are aliased to" [1].

In practice, the compiler doesn't seem to create a symbol. The alias identifier is simply replaced with the aliased thing, and you can't use the alias identifier as a symbol.

That means, you might be able to add an attribute to `FooList`, but you can't get back to it, because whenever you use `FooList` it's always replaced by `Foo[]`. And `Foo[]` doesn't have the attribute, of course.

I agree that it would probably make sense to disallow putting attributes on aliases. You can also mark aliases `const`, `static`, `pure`, etc. And they all have no effect.


[1] http://dlang.org/spec/declaration.html#AliasDeclaration

Reply via email to