is(typeof(...)) vs __traits(compiles, ...)

[Steven Schveighoffer]

On 2/21/20 5:15 AM, drug wrote:
> Currently this code does not compiles:
> ```
> unittest
> {
>      class MyClass
>      {
>          T opCall(T)(T p)
>          {
>              return p;
>          }
>      }
> 
>      import std.container.array : Array;
> 
>      Array!MyClass arr;
> }
> ```
> but if you comment out `opCall` in MyClass this code compiles. This is 
> caused by this in std.conv(4434):
> ```
>      static if (is(typeof(chunk = T(args))))
>                  chunk = T(args);
> ```
> The reason is that `is(typeof(chunk = T(args)))` returns true but does 
> not compiles becase MyClass has `opCall`, compiler calls `opCall` but it 
> needs `this` pointer that is unavailable. I replaced it by
> ```
>      static if (__traits(compiles, chunk = T(args)))
>                  chunk = T(args);
> ```
> it works but I'm not sure this good solution. The question is - 
> shouldn't `typeof` return false in this case? if so then the right fix 
> would be fix typeof.

This is a bug for is(typeof). It should indeed reject that call.

My understanding about __traits(compiles) is that it does some funky 
things in terms of allowing compilation that isn't normally allowed, 
which is a reason to prefer is(typeof). There are probably bugzilla 
issues on this. I remember compiler gurus (maybe Timon?) talking about 
this at one point.

I would say file an issue with a minimal test case. Any time you have:

static if(is(typeof(expr))) expr;

It should not error (excepting that there are some cases, such as 
expressions which can't technically be statements).

-Steve