Question on generated cpp for constexpr object

[mike919192]

Hi. I finally got a chance today to try out the constexpr functionality that was added recently. Functions worked exactly as I expected, but I was surprised by the generated cpp for constexpr objects.

This is the example from the last update

// object with "always the same value" (constexpr object)
forty_two: i64 == 42;

compiles to this cpp code

//actual cpp code
cpp2::i64 const& forty_two = 42;

but I was expecting something more like this cpp code

//expected cpp code
constexpr cpp2::i64 forty_two{42};

The actual code cannot be used where constexpr is required. For example:

//produces error
test_array: std::array<uint, forty_two> = 0;

hello.cpp2: In function ‘int main()’:
hello.cpp2:11:21: error: the value of ‘forty_two’ is not usable in a constant expression
   11 |     test: std::array<uint, forty_two> = 0;
      |                     ^~~~~~~~~
hello.cpp2:10:22: note: ‘forty_two’ was not initialized with a constant expression
   10 |     forty_two: i64 == 42;
      |                      ^~~~     
hello.cpp2:11:30: error: the value of ‘forty_two’ is not usable in a constant expression
   11 |     test: std::array<uint, forty_two> = 0;
      |                              ^
hello.cpp2:10:22: note: ‘forty_two’ was not initialized with a constant expression
   10 |     forty_two: i64 == 42;
      |                      ^~~~     
hello.cpp2:11:21: note: in template argument for type ‘long unsigned int’
   11 |     test: std::array<uint, forty_two> = 0;

Is this the intended design? Because it doesn't seem to be usable when constexpr is required.

[JohelEGP]

This is the intended design, although perhaps outdated.
At b589f5d#commitcomment-129028450, I argue that this should also lower to a constexpr object.

[mike919192]

Ok thanks for the reply. So is it correct to say that there is currently no pure cpp2 way to write this?

constexpr cpp2::i64 forty_two{42};

Seems like the best that is currently possible would be to used mixed cpp / cpp2. But that is also limited because I guess you would not be able to have a constexpr object declared inside of a cpp2 function.

[JohelEGP]

You can do this: #568 (comment).

[JohelEGP]

So is it correct to say that there is currently no pure cpp2 way to write this?

That's right.

This is an alternative (https://cpp2.godbolt.org/z/GsM5hj7zd):

constant: <V: _> type == std::integral_constant<decltype(V), V>;
main: () = {
  ROWS: type == constant<3>;
  static_assert(ROWS() == 3);
  static_assert(ROWS::value == 3);
}

[mike919192]

Thanks. That's usable as a workaround. I was just very confused when I first tried the code from the Autumn update since it says its supposed to be a constexpr object but isnt really, at least for now. Maybe the implementation is still WIP.

[hsutter]

forty_two: i64 == 42;

Thanks! This is just an oversight. Will fix.

Note that when the above is a non-local variable, it does lower to a constexpr variable: cpp2::i64 inline constexpr forty_two = 42;

When I made the change, I meant (but forgot) to check that the same (except not inline) was now being done for local variables for consistency.

Note that if you do want the equivalent of a local const&, you can still write it using block/statement scope variables which default to in semantics:

(v2 := v)  // v2 is a vector<T> const&
for v2 do (s)
    std::cout << "(s)$\n";

[hsutter]

Fixed in: 3512ecd

Thanks again.