Correction for short function syntax, closes #356

[filipsajdak]

The current implementation does not work for the following code:

main: () = {
    :() = 1;
    [[assert: 1]]
}

It fails with the error:

error: subscript expression [ ] must not be empty (if you were trying to name a C-style array type, use 'std::array' instead) (at '[')

This change introduces a minor correction that moves back parsing to a semicolon (to simulate a double semicolon) for short syntax.

It is not done in the following cases:

:() = 1;(); // immediately called lambda
f(a,b,:() = 1;); // last argument in function call
f(a,:() = 1;,c); // first or in the middle argument

After this change, the original issue is solved.
All regression tests pass. Closes #356

[JohelEGP]

I can confirm that the original issue is solved with this. Thank you.

[JohelEGP]

Let me add that I'm not sure of the implications of this solution. I hadn't thought of IIFE for simple function expressions.

I tested the following on branch main. The handling of IIFEs is inconsistent. Is that because it is a semantic exception that applies to a function scope? In particular, for main and the type, they are all in a statement (as opposed to the declaration-seq of the top-level). See https://godbolt.org/z/vYsqTe3Gq.

Cpp2 code demonstrating the issue.

a0: std::any = :() -> _ = { return 0; }();

// // error: unexpected end of source file
// b0: std::any = :() -> _ = { return 0; }(
// );

// // error: unexpected end of source file
// c0: std::any = :() -> _ = { return 0; }
// ();

d0: std::any = :() -> _ = 0;();

// // error: unexpected end of source file
// e0: std::any = :() -> _ = 0;(
// );

f0: std::any = :() -> _ = 0;
(); // mixed2.cpp2:18:2: error: expected unqualified-id

main: () = {
  a1: std::any = :() -> _ = { return 0; }();

  b1: std::any = :() -> _ = { return 0; }(
  );

  c1: std::any = :() -> _ = { return 0; }
  ();

  d1: std::any = :() -> _ = 0;();

  e1: std::any = :() -> _ = 0;(
  );

  f1: std::any = :() -> _ = 0;
  ();
}

x: type = {
  // Enabled individually, all generate:
  // ```C++
  //   private: std::any 𝘓2 {[]() -> auto;
  // ()}; public: x() = default; x(x const&) = delete; auto operator=(x const&) -> void = delete;
  // ```

  // a2: std::any = :() -> _ = { return 0; }();

  // b2: std::any = :() -> _ = { return 0; }(
  // );

  // c2: std::any = :() -> _ = { return 0; }
  // ();

  // d2: std::any = :() -> _ = 0;();

  // e2: std::any = :() -> _ = 0;(
  // );

  // f2: std::any = :() -> _ = 0;
  // ();
}

Should I open an issue for this?

[filipsajdak]

No need for a new issue as this is not yet merged.

[filipsajdak]

For the following code:

c0: std::any = :() -> _ = { return 0; }
();

first line is cpp2, the second cpp1:

+/* 2 */ c0: std::any = :() -> _ = { return 0; }
+/* 1 */ ();

Nevertheless it should just inform that there is an issue with missing semicolon.

[filipsajdak]

I have pushed the fix - the issue is with the error() method used to report errors.
The method uses curr() internally, which throws when cppfront has processed all tokens.

cppfront/source/parse.h

Line 2760 in f83ca9b

m += std::string(" (at '") + curr().to_string(true) + "')";

cppfront/source/parse.h

Lines 2787 to 2795 in f83ca9b

	auto curr() const
	-> token const&
	{
	if (done()) {
	throw std::runtime_error("unexpected end of source file");
	}
	return (*tokens_)[pos];
	}

[filipsajdak]

Now the error looks like the following:

bug_352_error_message_semicolon.cpp2...
bug_352_error_message_semicolon.cpp2(1,39): error: ill-formed initializer (after '}')
bug_352_error_message_semicolon.cpp2(1,1): error: unexpected text at end of Cpp2 code section (at 'c0')
bug_352_error_message_semicolon.cpp2(1,0): error: parse failed for section starting here

[JohelEGP]

Yes, with the build from 2b730e8.

[hsutter]

Looks good, thanks!