Fix test for double truncation

Summary:
Work around Clang 13 treating the truncation fast path as UB and
optimising it to not work on compile time constants.

This appears to be a bug in clang, since adding the
`-fno-strict-float-cast-overflow` flag does not fix it. Nonetheless,
this diff also adds that flag, since it is generally desirable for
Hermes code.

Reviewed By: tmikov

Differential Revision: D31939409

fbshipit-source-id: b86830d129e6bf1b872e502294ab9ea1893ce242

cmake/modules/Hermes.cmake

@@ -419,6 +419,9 @@ if (GCC_COMPATIBLE)
   # Enable -Wdelete-non-virtual-dtor if available.
   add_flag_if_supported("-Wdelete-non-virtual-dtor" DELETE_NON_VIRTUAL_DTOR_FLAG)
 
+  # Avoid triggering arbitrary UB when converting doubles to ints.
+  add_flag_if_supported("-fno-strict-float-cast-overflow" NO_STRICT_FLOAT_CAST_OVERFLOW_FLAG)
+
   # Disable range loop analysis warnings.
   check_cxx_compiler_flag("-Wrange-loop-analysis" RANGE_ANALYSIS_FLAG)
   append_if(RANGE_ANALYSIS_FLAG "-Wno-range-loop-analysis" CMAKE_CXX_FLAGS)

unittests/Support/ConversionsTest.cpp

@@ -15,25 +15,30 @@ using namespace hermes;
 
 namespace {
 
+// Work around new versions of clang treating the truncation fast path as UB
+// and optimising it to not work on compile time constants.
+#ifdef __clang__
+LLVM_ATTRIBUTE_NOINLINE
+#endif
+int32_t toInt32Wrapper(double d) {
+  return hermes::truncateToInt32(d);
+}
+
 TEST(ConversionsTest, toInt32Test) {
-  EXPECT_EQ(0, hermes::truncateToInt32(0));
-  EXPECT_EQ(0, hermes::truncateToInt32(-0.1));
-  EXPECT_EQ(0, hermes::truncateToInt32(0.1));
-  EXPECT_EQ(1, hermes::truncateToInt32(1));
-  EXPECT_EQ(-1, hermes::truncateToInt32(-1.5));
-
-  EXPECT_EQ(1661992960, hermes::truncateToInt32(1e20));
-  EXPECT_EQ(-1661992960, hermes::truncateToInt32(-1e20));
-  EXPECT_EQ(-2147483648, hermes::truncateToInt32(-2147483648));
-
-  EXPECT_EQ(
-      0, hermes::truncateToInt32(std::numeric_limits<double>::infinity()));
-  EXPECT_EQ(
-      0, hermes::truncateToInt32(-std::numeric_limits<double>::infinity()));
-  EXPECT_EQ(
-      0, hermes::truncateToInt32(-std::numeric_limits<double>::quiet_NaN()));
-  EXPECT_EQ(
-      0, hermes::truncateToInt32(-std::numeric_limits<double>::denorm_min()));
+  EXPECT_EQ(0, toInt32Wrapper(0));
+  EXPECT_EQ(0, toInt32Wrapper(-0.1));
+  EXPECT_EQ(0, toInt32Wrapper(0.1));
+  EXPECT_EQ(1, toInt32Wrapper(1));
+  EXPECT_EQ(-1, toInt32Wrapper(-1.5));
+
+  EXPECT_EQ(1661992960, toInt32Wrapper(1e20));
+  EXPECT_EQ(-1661992960, toInt32Wrapper(-1e20));
+  EXPECT_EQ(-2147483648, toInt32Wrapper(-2147483648));
+
+  EXPECT_EQ(0, toInt32Wrapper(std::numeric_limits<double>::infinity()));
+  EXPECT_EQ(0, toInt32Wrapper(-std::numeric_limits<double>::infinity()));
+  EXPECT_EQ(0, toInt32Wrapper(-std::numeric_limits<double>::quiet_NaN()));
+  EXPECT_EQ(0, toInt32Wrapper(-std::numeric_limits<double>::denorm_min()));
 }
 
 TEST(ConversionsTest, toArrayIndexTest) {