Improved diagnostics for mismatched template constraints

[Paul Backus]

On Wednesday, 12 May 2021 at 12:54:24 UTC, Per Nordlöw wrote:
> I'm certain I've seen dmd be able to diagnose which 
> sub-expression of `isDumb` and `isInputRange` that evaluated to 
> false. Did I just dream this or is there a way? If there is a 
> way does this improvement still reside in a dmd PR?

It will diagnose which sub-expression of the template constraint 
failed, but it will not descend into templates. So, if you inline 
the constraint:

     void f(Args...)(Args args)
     if (Args.length == 2 && Args.length == 3)
     {
     }

You get the following error message:

     onlineapp.d(18): Error: template `onlineapp.f` cannot deduce 
function from argument types `!()(int, int)`, candidates are:
     onlineapp.d(6):        `f(Args...)(Args args)`
       with `Args = (int, int)`
       must satisfy the following constraint:
     `       Args.length == 3`

It might be possible to extend this to work with simple `enum` 
templates like `isDumb` and `isInputRange`, whose bodies consist 
of a single boolean expression, although I expect you would run 
into many of the same difficulties that caused past attempts at 
fixing issue 1807 [1] to fail.

[1] https://issues.dlang.org/show_bug.cgi?id=1807