Bug 923969 – dev-libs/apache-arrow-15.0.0 fails tests on x86 (32-bit?): arrow-compute-scalar-math-test, arrow-compute-aggregate-test, arrow-utility-test

• Status: CONFIRMED
• Product: Gentoo Linux

[Description matoro 2024-02-07 05:14:00 UTC]

All floating-point precision issues.

[ RUN      ] TestUnaryArithmeticFloating/1.Sqrt
/var/tmp/portage/dev-libs/apache-arrow-15.0.0/work/apache-arrow-15.0.0/cpp/src/arrow/testing/gtest_util.cc:143: Failure
Failed
Expected:
1.3407807929942596e+154
Actual:
1.3407807929942597e+154
/var/tmp/portage/dev-libs/apache-arrow-15.0.0/work/apache-arrow-15.0.0/cpp/src/arrow/testing/gtest_util.cc:143: Failure
Failed
Expected:
1.3407807929942596e+154
Actual:
1.3407807929942597e+154
[  FAILED  ] TestUnaryArithmeticFloating/1.Sqrt, where TypeParam = arrow::DoubleType (0 ms)


[ RUN      ] TestVarStdKernelMergeStability.Basics
/var/tmp/portage/dev-libs/apache-arrow-15.0.0/work/apache-arrow-15.0.0/cpp/src/arrow/compute/kernels/aggregate_test.cc:3231: Failure
Expected equality of these values:
  var->value
    Which is: 3872
  expected_var
    Which is: 3904
[  FAILED  ] TestVarStdKernelMergeStability.Basics (0 ms)


[ RUN      ] TestVarStdKernelRandom/4.Basics
/var/tmp/portage/dev-libs/apache-arrow-15.0.0/work/apache-arrow-15.0.0/cpp/src/arrow/compute/kernels/aggregate_test.cc:3231: Failure
Expected equality of these values:
  var->value
    Which is: 1011469379.3358333
  expected_var
    Which is: 1011469379.3358327
/var/tmp/portage/dev-libs/apache-arrow-15.0.0/work/apache-arrow-15.0.0/cpp/src/arrow/compute/kernels/aggregate_test.cc:3231: Failure
Expected equality of these values:
  var->value
    Which is: 1011966660.2499611
  expected_var
    Which is: 1011966660.2499604
[  FAILED  ] TestVarStdKernelRandom/4.Basics, where TypeParam = arrow::FloatType (0 ms)

Reproducible: Always

[Comment 1 matoro 2024-02-07 05:16:34 UTC]

Created attachment 884437 [details]
build.log and emerge --info

[Comment 2 matoro 2024-02-07 05:17:00 UTC]

Pending checking this on other 32-bit.

[Comment 3 Michał Górny 2024-02-12 18:11:32 UTC]

I can reproduce a subset of these failures but they go away with -ffpmath=sse.  I don't think there's much of a point in bothering with pre-SSE CPUs.

[Comment 4 Michał Górny 2024-02-14 10:03:19 UTC]

Unblocking per the -mfpmath=sse mail.