[BUG] Non-empty line after one-liner type with requires clause fails assertion

[JohelEGP]

Title: Non-empty line after one-liner type with requires clause fails assertion.

Minimal reproducer (https://cpp2.godbolt.org/z/bfh5bfTj4):

t: type requires true = { }
main: () = { }

Commands:

cppfront -clean-cpp1 main.cpp2

Actual result and error:

Output.

cppfront: source/cppfront.cpp:523: void cpp2::positional_printer::align_to(cpp2::source_position): Assertion `psource && 0 <= curr_pos.lineno && curr_pos.lineno < std::ssize(psource->get_lines())+1' failed.

[filipsajdak]

I have an easy fix for the assertion:

diff --git a/source/cppfront.cpp b/source/cppfront.cpp
index 5b89964..44cc52e 100644
--- a/source/cppfront.cpp
+++ b/source/cppfront.cpp
@@ -5094,7 +5094,7 @@ public:
                 if (n.requires_clause_expression) {
                     printer.print_cpp2("requires( ", n.requires_pos);
                     emit(*n.requires_clause_expression);
-                    printer.print_cpp2(" )\n\n", n.requires_pos);
+                    printer.print_cpp2(" )\n", n.requires_pos);
                 }
 
                 printer.print_cpp2("class ", n.position());

But even without the parsing error, the result cpp1 code is not good (check here: https://cpp2.godbolt.org/z/eGMMz8Txj)

#include "cpp2util.h"

requires( true )

class t;
requires( true )

class t {
      public: t() = default;
      public: t(t const&) = delete; /* No 'that' constructor, suppress copy */
      public: auto operator=(t const&) -> void = delete;

};

auto main() -> int;

auto main() -> int{}

error:

build/_cppfront/main.cpp:5:1: error: expected unqualified-id before 'requires'
    5 | requires( true )
      | ^~~~~~~~
build/_cppfront/main.cpp:8:1: error: expected unqualified-id before 'requires'
    8 | requires( true )
      | ^~~~~~~~
make[2]: *** [CMakeFiles/test.dir/build.make:81: CMakeFiles/test.dir/_cppfront/main.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:83: CMakeFiles/test.dir/all] Error 2

[JohelEGP]

Hmmm... yes.
It's because t isn't a templated entity.
Just add a template-head to it.
Thank you for looking into this!

[filipsajdak]

The most probably requires clause for non-templates should be a cpp2 parsing error.

[JohelEGP]

That's actually a Cpp1 semantic rule
that could be elevated to a Cpp2 parser error,
but so could many other Cpp1 semantic rules.

[hsutter]

Thanks!
I think I'd prefer leaving the semantic rule enforced by the Cpp1 compiler, as long as it's reasonably readable and points to the right Cpp2 line number.

[hsutter]

Re elevating the Cpp1 semantic rule to Cpp2, repeating rationale from #528 for visibility:

I do think there's value in diagnosing a requires on a non-templated type for type declarations.

Rationale:

• writing a stray requires on a function already gives a nice Cpp1 diagnostic (e.g., MSVC says error C7599: 't::f': a trailing requires clause is only allowed on a templated function)
• writing a stray requires on a type currently lowers it to a naked requires at global scope before the type which leads to a harder-to-understand Cpp1 diagnostic (e.g., MSVC says error C2059: syntax error: 'requires', which is correct and indicates the rough issue, but it's not as specific)

So diagnosing the type case in Cpp2 delivers the primary diagnostic usability improvement.