abseil-duration-factory-scale test fails on 32-bit IA32

[llvmbot]

Bugzilla Link	41460
Version	unspecified
OS	Linux
Attachments	clang build log on Archlinux32
Reporter	LLVM Bugzilla Contributor
CC	@EugeneZelenko

Extended Description

The clang-tidy testsuite fails on 32-bit IA32 for abseil-duration-factory-scale:

Failing Tests (1):
Clang Tools :: clang-tidy/abseil-duration-factory-scale.cpp

Expected Passes : 1126
Expected Failures : 1
Unsupported Tests : 3
Unexpected Failures: 1

FAILED: tools/extra/test/CMakeFiles/check-clang-tools
cd /build/clang/src/cfe-8.0.0.src/build/tools/extra/test && /usr/bin/python /usr/bin/lit -sv /build/clang/src/cfe-8.0.0.src/build/tools/extra/test
ninja: build stopped: subcommand failed.

The full build log is attached.

[kabe-gh]

I am also having trouble getting abseil-duration-factory-scale.cpp check on 32bit fail.

I tracked this down to clang-tidy/abseil/DurationFactoryScaleCheck.cpp:getNewScale()
comparing double float value to 1.0,
but 0.001 * 1000 doesn't equal 1.0 due to precision.

Following patch seems to fix this.
I don't know why x86_64 is working without this kind of patch.

Subject: fuzzy match double Multiplier == 1.0 in DurationFactoryScaleCheck.cpp
From: <kabe@>

Replace double comparison d == 1.0 with fuzzy match std::abs(d - 1.0) < 1e-14.
1e-14 is an arbitary epsilon.

Without this patch, clang-tidy did not downshift
  d = absl::Seconds(1e-3 * 30);
to
  d = absl::Milliseconds(30);


diff -up ./clang-tools-extra-13.0.1.src/clang-tidy/abseil/DurationFactoryScaleCheck.cpp.t64 ./clang-tools-extra-13.0.1.src/clang-tidy/abseil/DurationFactoryScaleCheck.cpp
--- ./clang-tools-extra-13.0.1.src/clang-tidy/abseil/DurationFactoryScaleCheck.cpp.t64  2022-01-21 06:31:59.000000000 +0900
+++ ./clang-tools-extra-13.0.1.src/clang-tidy/abseil/DurationFactoryScaleCheck.cpp      2022-10-11 22:27:36.694627456 +0900
@@ -96,12 +96,15 @@ getNewScaleSingleStep(DurationScale OldS
 // would produce a new scale.  If so, return it, otherwise `None`.
 static llvm::Optional<DurationScale> getNewScale(DurationScale OldScale,
                                                  double Multiplier) {
-  while (Multiplier != 1.0) {
+  //while (Multiplier != 1.0)
+  while (std::abs(Multiplier - 1.0) > 1e-14) {
     llvm::Optional<std::tuple<DurationScale, double>> Result =
         getNewScaleSingleStep(OldScale, Multiplier);
     if (!Result)
       break;
-    if (std::get<1>(*Result) == 1.0)
+    //if (std::get<1>(*Result) == 1.0)
+    //Above does not work properly when shifting up 0.001 to 1.0
+    if (std::abs(std::get<1>(*Result) - 1.0) < 1e-14)
       return std::get<0>(*Result);
     Multiplier = std::get<1>(*Result);
     OldScale = std::get<0>(*Result);