Emit a diagnostic warning about parsing ambiguity in a conjunction

[BaLiKfromUA]

Addresses #11

Testing performed

• Added a basic parser test to catch a warning message, tested locally.

> env LIT_FILTER=cxx2x-ambig-reflect-expr.cpp ninja check-clang -C build-llvm/

ninja: Entering directory `build-llvm/'
[89/90] Running the Clang regression tests
llvm-lit: /home/balik/Desktop/clang-p2996/llvm/utils/lit/lit/llvm/config.py:509: note: using clang: /home/balik/Desktop/clang-p2996/build-llvm/bin/clang

Testing Time: 0.04s

Total Discovered Tests: 19357
  Excluded: 19356 (99.99%)
  Passed  :     1 (0.01%)

• Manually tested by compiling a code with mentioned issue

#include <experimental/meta>
#include <iostream>

enum MyEnum { x, y, e = -1, f, z = 99 };

int main() {
  MyEnum value{};
  constexpr auto reflValue = ^value; 
  
  if (reflValue != ^MyEnum && true) {
    std::cout << "YES" << std::endl;
  } else {
    std::cout << "NO" << std::endl;
  }
}

> build-llvm/bin/clang++ -std=c++26 -freflection -Wl,-rpath,/home/balik/Development/clang-p2996/build-llvm/lib/x86_64-unknown-linux-gnu -L/home/balik/Development/clang-p2996/build-llvm/lib/x86_64-unknown-linux-gnu -stdlib=libc++ -lc++ -lc++abi -o a.out main.cpp

main.cpp:10:28: warning: given expression may be misinterpreted due to 'T &&' type parsing logic. did you mean '... (^T) && ...'?
   10 |   if (reflValue != ^MyEnum && true) {
      |                            ^
main.cpp:10:31: error: expected ')'
   10 |   if (reflValue != ^MyEnum && true) {
      |                               ^
main.cpp:10:6: note: to match this '('
   10 |   if (reflValue != ^MyEnum && true) {
      |      ^
1 warning and 1 error generated.

Additional context

PR is ready for review IMO.

Last issue left on my side:

• Didn't manage to capture a type name to display it in a warning message instead of T. Should be possible but need more investigation.

[katzdm]

Let's use e.g., expected-warning@-1, rather than hardcode a fixed line number, to make these tests a bit more resilient to subsequent changes.

(same for other checks below)

[BaLiKfromUA]

Fixed bfdfdd6

[katzdm]

I think there's an important missing case here, which is that e.g.,:

constexpr auto r1 = ^int &;
constexpr auto r2 = ^int &&;

should both be ok without warning.

[BaLiKfromUA]

Nice catch, missed it!

Fixed a736536

[katzdm]

nit: maybe something like:

token '&&' binds left to the type 'int &&'; did you mean '(^int) &&'?

[BaLiKfromUA]

Changed. But I still cannot get name of type :(
So message looks like this: token '&' binds left to type 'T &'; did you mean '(^T) &'?

I checked similar places in a code to get an idea where to get type name but in all methods names are empty for my tests.

[katzdm]

Unfortunately, there are a lot more cases here. I'm not sure this approach will work (although it might - not really sure one way or the other).

r == ^int& ? 1 : 0
condition ? ^int& : ^int;
^int&(*)(bool)
^int& != r
^int& == r
constexpr auto r {^int&};

In addition to those, in the presence of some user-defined

consteval result_type operator???(std::meta::info R, MyStupidClass);

You could have ^int& + o, along with: -, *, /, %, ], |, || <=, >=, >.

Those are all the cases I can think of off the top of my head, but I may have forgotten some.

[BaLiKfromUA]

thank you! gonna think more how to address it

[BaLiKfromUA]

Gonna open another PR once I found proper solution