[BUG] `@print` shortcomings

[JohelEGP]

Title: Single-expression functions print with extra semicolon.

Description:

This makes the printed code invalid Cpp2:

example.cpp2(4,51): error: a single-expression function should end with a single semicolon (at ';')

Minimal reproducer (https://cpp2.godbolt.org/z/jGb31dK74):

#include <functional>
t: @struct @print type = {
  x: std::function<void()> = :() = _ = 0;
  y: std::function<void()> = :() 0;
}
main: () = { }

Commands:

cppfront main.cpp2
clang++18 -std=c++23 -stdlib=libc++ -lc++abi -pedantic-errors -Wall -Wextra -Wconversion -Werror=unused-result -I . main.cpp

Expected result: Printed Cpp2 code to be valid.

Actual result and error: public x: std::function<void()> = :() = _ = 0;;.

Cpp2 lowered to Cpp1:

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


#include "cpp2util.h"


class t;
  

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

#include <functional>
class t {
  public: std::function<void()> x {[]() -> void { static_cast<void>(0);  }}; 
  public: std::function<void()> y {[]() -> auto { return 0;  }}; 
};
auto main() -> int;

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


auto main() -> int{}

Output:

main.cpp2...

t: type = 
{
    public x: std::function<void()> = :() = _ = 0;;

    public y: std::function<void()> = :() -> move _ = 0;;
}
 ok (mixed Cpp1/Cpp2, Cpp2 code passes safety checks)

See also:

• fix(parse): allow (to omit) semicolon after _unnamed-declaration_ #479.

[JohelEGP]

This is #479's test:

u: @struct @print type = {
  t: type = {
    operator[]: (this, f) = {}
  }

  main: () -> int = {
    (x := t()) { x[:() -> _ = 0]; }
    (x := t()) { x[:() -> _ = 0;]; }

    assert(!(:() = 0; is int));

    _ = :i32 = 0;
    assert(true);

    return :i32 = 0;
  }

  a: i32 == :i32 = 0;
  b: i32 = 0;
}

main: () = {}

What it prints:

u: type = {
  public t: type = {
    operator[]: (in this, in f: ) = {}
  }

  public main: () -> move int = {
    (in x := t()) { x[:() -> move _ = 0]; }
    (in x := t()) { x[:() -> move _ = 0;]; }
    assert(!(:() = 0; is int))
    _ = :i32 = 0;
    ;
    assert(true)
    return :i32 = 0;
    ;
  }

  public a: i32 == :i32 = 0;
  ;

  public b: i32 = 0;
}

And where the errors are at:

u: type = {
  public t: type = {
    operator[]: (in this, in f: ) = { }
  }

  public main: () -> move int = {
    (in x := t()) { x[:() -> move _ = 0]; }
    (in x := t()) { x[:() -> move _ = 0;]; }
    assert(!(:() = 0; is int)); // Added ';'.
    _ = :i32 = 0;               // x.cpp2(10,5): error: missing ';' after contract-statement (at '_')
    // ;                        // x.cpp2(11,5): error: empty statement is not allowed - remove extra semicolon (at ';')
    assert(true);    // Added ';'.
    return :i32 = 0; // x.cpp2(13,5): error: missing ';' after contract-statement (at 'return')
    // ;             // x.cpp2(14,5): error: empty statement is not allowed - remove extra semicolon (at ';')
  }

  public a: i32 == :i32 = 0;
  // ; // x.cpp2(18,3): error: empty statement is not allowed - remove extra semicolon (at ';')

  public b: i32 = 0;
}

• asserts need ;. Resolved with commit b41f60e.
• Statements ending in ; need to remove any extra ;.

[JohelEGP]

Also, deduced type parameters print with : but without a type/placeholder. Apparently, it works.
And function expressions in initializers of constexpr variables have an extra line break.
See https://cpp2.godbolt.org/z/aoY3zqEW7.

t: @struct @print type = {
  i: int == :(_) 0;(0);
}
main: () = { }

t: type = 
{
    public i: int == 
:(in _:) -> move _ = 0;(0);
}

[JohelEGP]

@structs print without @struct,
so they can be invalid Cpp2
and be interpreted by the user wrongly.

@struct doesn't generate Cpp2 code,
but instead affects lowering.
So it should be @printed.

[AbhinavK00]

Very sorry for the questions but what does print metafunction do? Also, is it confirmed that multiple metafunctions can be applied on the same type?

[JohelEGP]

For @print, when you use it, like at 586ff20#diff-4ccd4b8a84887f1aadc839722d871182e36bd2a986f4bae0dfa9edfdfe1a3952,
cppfront will pretty-print it during compilation: 586ff20#diff-86374c72ff7e9b814ef80bc78f78f4cac4189bf2b4fbb78be253303ff4ee9d4b.
It can be useful to see what a metafunction generates.
For example, see #821 (comment).

Yes, metafunctions form a pipeline: #447 (comment).

[AbhinavK00]

Thanks for the answer!

[hsutter]

Thanks!

I think I got them all, including the ones in your test cases and a few more... but please check and reopen with a new test case if I missed some.

[JohelEGP]

Thank you.
How about #840 (comment)?

@structs print without @struct,
so they can be invalid Cpp2
and be interpreted by the user wrongly.

@struct doesn't generate Cpp2 code,
but instead affects lowering.
So it should be @printed.

[hsutter]

@struct doesn't generate Cpp2 code,
but instead affects lowering.
So it should be @printed.

Good point, that's the simplest fix for now so I'll implement it that way. Thanks!

[JohelEGP]

With program-defined metafunctions (#907), that can be generally true for all metafunctions.
So it might be better if the pretty print visualizer could print processed metafunctions in a C comment.