Fix modern MSVC compilation error on experimental/filesystem

[bob80905]

Modern MSVC deperecated experimental/filesystem.
This causes build failures when building DXC with MSVC.
The deprecation occurred on version 1922, so we should account for this removal, instead of silencing deprecation warnings.

[github-actions]

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff 14696d1cc10dd5fd802a28a751733747b11022ad 54e02abbd060c9f21a2ef13c6e12317b26b3847b -- tools/clang/unittests/HLSL/DxilContainerTest.cpp

View the diff from clang-format here.

diff --git a/tools/clang/unittests/HLSL/DxilContainerTest.cpp b/tools/clang/unittests/HLSL/DxilContainerTest.cpp
index 98153b24..90c09c66 100644
--- a/tools/clang/unittests/HLSL/DxilContainerTest.cpp
+++ b/tools/clang/unittests/HLSL/DxilContainerTest.cpp
@@ -80,17 +80,17 @@ using namespace fs;
 
 #include <codecvt>
 
-    // clang-format on
+// clang-format on
 
-    using namespace std;
-    using namespace hlsl_test;
+using namespace std;
+using namespace hlsl_test;
 #ifdef _WIN32
 
-    static uint8_t MaskCount(uint8_t V) {
-      DXASSERT_NOMSG(0 <= V && V <= 0xF);
-      static const uint8_t Count[16] = {0, 1, 1, 2, 1, 2, 2, 3,
-                                        1, 2, 2, 3, 2, 3, 3, 4};
-      return Count[V];
+static uint8_t MaskCount(uint8_t V) {
+  DXASSERT_NOMSG(0 <= V && V <= 0xF);
+  static const uint8_t Count[16] = {0, 1, 1, 2, 1, 2, 2, 3,
+                                    1, 2, 2, 3, 2, 3, 3, 4};
+  return Count[V];
 }
 #endif
 

• Check this box to apply formatting changes to this branch.

[farzonl]

Not sure which compiler this else case covers?

[bob80905]

I consulted an AI, and basically it looks like really old compilers would be covered by this case:
GCC 5.3 – 7.x
Pre-2014 GCC / Clang without __has_include
Clang with an old libstdc++/libc++

are some examples.

[farzonl]

Do we even support compilers that old for building dxc? LLVM has a cut off. Not sure if DXC ever required a minimum compiler version.

[hekota]

GCC 8 came out in 2018, and I don't think we should support compilers older than that. I think you can remove the #elif`` condition and just fall back to #include `.

[farzonl]

why do we need to check && \ (defined(__clang__) || !defined(__GNUC__) || __GNUC__ >= 8) why isn't just checking for the feature sufficient?

[bob80905]

As for GNUC >= 8: GCC 7 ships but doesn't actually define std::filesystem. Header exists, namespace doesn't -> compile error. GCC 8 is the first version where it really works.
For the defined(clang) check: Clang sets GNUC == 4, so the >= 8 check would wrongly reject it. The clause basically says "if it's Clang, skip the GCC version check."

[farzonl]

This looks correct. some stuff seems a bit extra but not a reason to hold up the pr.

[hekota]

Suggested change

	#if _MSC_VER < 1922
	#if (_MSC_VER >= 1920) && (_MSC_VER < 1922)

experimental::filesystem was enabled since version 1920, so it needs to be included in the check.

[hekota]

GCC 8 came out in 2018, and I don't think we should support compilers older than that. I think you can remove the #elif`` condition and just fall back to #include `.