On Sunday, 15 March 2015 at 16:53:34 UTC, Marc Schütz wrote:
On Sunday, 15 March 2015 at 16:44:14 UTC, Ali Çehreli wrote:
On 03/15/2015 08:47 AM, "Marc =?UTF-8?B?U2Now7x0eiI=?=
<schue...@gmx.net>" wrote:
> Should this work?
>
> struct V(string s) {
> }
>
> struct S(int U) {
> V!"xyz" x;
> alias x this;
> }
>
> void main() {
> S!10 a;
> static assert(is(a : V!Args, Args...));
> }
>
> With DMD Git master, the static assert() fails. Should it? Am
I doing
> something wrong? How can I test whether something is
implicitly
> convertible to any instance of a particular template?
There is no way other than checking for compile-time duck
typing (see the implementations of isInputRange and others).
One reason is that the compiler does not have the concept of
"an instance of a template". Templates are for code generation
and only the end-result (i.e. S!10) lives as a concept when
compiling.
The code contained a small mistake, I forgot a `typeof()`:
// static assert(is(a : V!Args, Args...));
// should be:
static assert(is(typeof(a) : V!Args, Args...));
This still fails, but it works when I change it to:
static assert(is(typeof(a) : S!Args, Args...));
This means I can indeed test whether something _is_ an instance
of a template. It just doesn't take the `alias this` into
account. So I guess that's a bug?
Ok, now I'm pretty sure:
class V(string s) {
}
class S(int U) : V!"xyz" {
}
void main() {
S!10 a;
static if(is(typeof(a) : V!Args, Args...))
pragma(msg, Args);
}
This works, and it even correctly infers `Args` to be
`tuple("xyz")`. As `alias this` is supposed to be interchangeable
with subtyping, it must be a bug.