[libcxx-commits] [libcxx] [libc++] LWG2381: Inconsistency in parsing floating point numbers (PR #77948)

[Mark de Wever via libcxx-commits]

================
@@ -193,6 +319,7 @@ int main(int, char**)
         assert(err == ios.failbit);
         assert(v == INFINITY);
     }
+#if LDBL_MANT_DIG >= 64
----------------
mordante wrote:

I initially expected that, but these tests are running with clang-cl (and mingw). I expect at least clang-cl uses LDBL_MANT_DIG == 53 and thus give the same results as MSVC STL.

I'm especially puzzled why these infinity tests fail while the infinity test above passes.
The `304888344611713860501504000000` test seems more likely to fail due to a lack of precision.
(I see this test is listed in https://github.com/microsoft/STL/blob/main/tests/libcxx/expected_results.txt)

Did you run these tests with MSVC STL and checked why they failed?

https://github.com/llvm/llvm-project/pull/77948