[BUG] Non-type template parameter used as _type-id_ is not diagnosed

[JohelEGP]

Title: Non-type template parameter used as type-id is not diagnosed.

Description:

I encountered this when poking Cppfront to get it to emit a type-constraint.
The issue is alleviated by resolving #425,
as it would only happen with an NTTP by copy.

Minimal reproducer (https://cpp2.godbolt.org/z/WehjhjPjY):

t: @struct <T: std::regular> type = {
  v: T;
}
f: <T: std::regular> (v: T) = _ = v;
g: <T: std::regular> (forward v: T) = _ = v;
h: <T: std::regular> (move v: T) = _ = v;

main: () = {
  _ = :t = (0); // Not OK (works, should error)
  f(0); // error (by chance?)
  g(0); // error (by chance?)
  h(0); // Not OK (works, should error)
}

Commands:

cppfront main.cpp2
clang++17 -std=c++23 -stdlib=libc++ -lc++abi -pedantic-errors -Wall -Wextra -Wconversion -fsanitize=undefined -Werror=unused-result -I . main.cpp

Expected result: A diagnostic when using an NTTP as a type.

Actual result and error:

Cpp2 lowered to Cpp1:

//=== Cpp2 type declarations ====================================================


#include "cpp2util.h"

template<std::regular T> class t;
  

//=== Cpp2 type definitions and function declarations ===========================

template<std::regular T> class t {
  public: T v; 
};
template<std::regular T> auto f(cpp2::in<T> v) -> void;
template<std::regular T> auto g(auto&& v) -> void
CPP2_REQUIRES (std::is_same_v<CPP2_TYPEOF(v), T>)
;
template<std::regular T> auto h(T&& v) -> void;

auto main() -> int;
  

//=== Cpp2 function definitions =================================================


template<std::regular T> auto f(cpp2::in<T> v) -> void { (void) v;  }
template<std::regular T> auto g(auto&& v) -> void
requires (std::is_same_v<CPP2_TYPEOF(v), T>)
 { (void) CPP2_FORWARD(v);  }
template<std::regular T> auto h(T&& v) -> void { (void) std::move(v);  }

auto main() -> int{
  (void) t{0};  // Not OK (works, should error)
  f(0); // error (by chance?)
  g(0); // error (by chance?)
  h(0); // Not OK (works, should error)
}

Output:

main.cpp2:10:3: error: no matching function for call to 'f'
   10 |   f(0); // error (by chance?)
      |   ^
main.cpp2:4:31: note: candidate template ignored: couldn't infer template argument 'T'
    4 | template<std::regular T> auto f(cpp2::in<T> v) -> void { (void) v;  }
      |                               ^
main.cpp2:11:3: error: no matching function for call to 'g'
   11 |   g(0); // error (by chance?)
      |   ^
main.cpp2:5:31: note: candidate template ignored: couldn't infer template argument 'T'
    5 | template<std::regular T> auto g(auto&& v) -> void
      |                               ^
2 errors generated.

[JohelEGP]

The issue is alleviated by resolving #425,
as it would only happen with an NTTP by copy.

A by copy NTTP is actually the common case,
so fixing #425 is unlikely to alleviate anything.

[hsutter]

When you wrote <T: std::regular>, I think you meant to apply a concept constraint.

That isn't yet supported, but my intent is to spell it as <T: type is std::regular>.

The syntax <T: sometypeid> is always a value parameter (NTTP).

Please reopen if I misunderstood! Thanks.

[JohelEGP]

Yes, I know it's not supported, and did it intentionally anyways.
What I'm wondering is whether Cppfront should diagnose
the function definitions of those calls in the reproducer that say // Not OK (works, should error).
Because it knows T is an NTTP, yet allows T to be used as a type-id.

[JohelEGP]

That isn't yet supported, but my intent is to spell it as <T: type is std::regular>.

I'm supposing we will be able to omit type, and similarly declare NTTPs like <V: _ is std::regular>.
And similarly support this terse syntax on parameters, return types, and variable declarations (just like Cpp1), e.g.

f: <_: type is C, _: is C, _: _ is C> (_: _ is C, _: is C) -> _ is C { _: _ is C = …; }