key_field_test failures on s390x [FlatBuffers 22.11.23]

[musicinmybrain]

While updating the flatbuffers package in Fedora Linux (Rawhide, gcc 12.2.1) from 22.10.26 to 22.11.23, I found that the new key_field_test fails on the s390x architecture. This is the only big-endian primary architecture in Fedora Linux.

Test project /builddir/build/BUILD/flatbuffers-22.11.23/redhat-linux-build
    Start 1: flattests
1/1 Test #1: flattests ........................***Exception: SegFault  0.25 sec
EXPECTED: "3"
VALUE: "4"
TEST FAILED: /builddir/build/BUILD/flatbuffers-22.11.23/tests/key_field_test.cpp:59, 'sorted_bar_vec->Get(3)->b()' != '4' in 
EXPECTED: "1"
VALUE: "0"
TEST FAILED: /builddir/build/BUILD/flatbuffers-22.11.23/tests/key_field_test.cpp:60, 'nullptr' == 'sorted_bar_vec->LookupByKey( &flatbuffers::CastToArray(test_float_array1))' in 
0% tests passed, 1 tests failed out of 1
Total Test time (real) =   0.25 sec
The following tests FAILED:
	  1 - flattests (SEGFAULT)

While I don’t have interactive access to an s390x system, I am happy to test any patches or experiments via test builds of the RPM package.

[sunwen18]

I'm curious if this is related to how big endian and little endian treat float differently. The part that's failing is using float array as key, and the previous part of using uint8 array as key didn't fail.

[sunwen18]

Hi Ben, I'm wondering if there is a way that you could test my PR #7681 ?
I updated the unit test a bit and enforced EndianSwap before comparing in the KeyCompareWithValue function, I'm not totally sure about this, would be nice to know for sure. Thanks in advance!

[musicinmybrain]

Hi Ben, I'm wondering if there is a way that you could test my PR #7681 ? I updated the unit test a bit and enforced EndianSwap before comparing in the KeyCompareWithValue function, I'm not totally sure about this, would be nice to know for sure. Thanks in advance!

Based on a scratch build of the Fedora RPM with the patch applied, that doesn‘t seem to have solved the problem.

Test project /builddir/build/BUILD/flatbuffers-22.11.23/redhat-linux-build
    Start 1: flattests
1/1 Test #1: flattests ........................***Exception: SegFault  0.18 sec
EXPECTED: "2"
VALUE: "4"
TEST FAILED: /builddir/build/BUILD/flatbuffers-22.11.23/tests/key_field_test.cpp:69, 'sorted_bar_vec->Get(3)->b()' != '4' in 
EXPECTED: "1"
VALUE: "0"
TEST FAILED: /builddir/build/BUILD/flatbuffers-22.11.23/tests/key_field_test.cpp:74, 'nullptr' == 'sorted_bar_vec->LookupByKey(key_float_array)' in 
0% tests passed, 1 tests failed out of 1
Total Test time (real) =   0.18 sec
The following tests FAILED:
	  1 - flattests (SEGFAULT)

[sunwen18]

Hi Ben, I made one more update, mind give it another shot? Thanks!

[musicinmybrain]

I find that the current version still fails:

Test project /builddir/build/BUILD/flatbuffers-22.11.23/redhat-linux-build
    Start 1: flattests
1/1 Test #1: flattests ........................***Exception: SegFault  0.66 sec
EXPECTED: "2"
VALUE: "4"
TEST FAILED: /builddir/build/BUILD/flatbuffers-22.11.23/tests/key_field_test.cpp:69, 'sorted_bar_vec->Get(3)->b()' != '4' in 
EXPECTED: "1"
VALUE: "0"
TEST FAILED: /builddir/build/BUILD/flatbuffers-22.11.23/tests/key_field_test.cpp:74, 'nullptr' == 'sorted_bar_vec->LookupByKey(key_float_array)' in 
0% tests passed, 1 tests failed out of 1
Total Test time (real) =   0.66 sec
The following tests FAILED:
	  1 - flattests (SEGFAULT)

[sunwen18]

Hi Ben, I have merged #7681 to fix this, I have tested locally with building docker of multiarch/ubuntu-core:s390x-focal. Thing should work now for s390x if you wanna pull the latest master and give it a try.

[musicinmybrain]

Hi Ben, I have merged #7681 to fix this, I have tested locally with building docker of multiarch/ubuntu-core:s390x-focal. Thing should work now for s390x if you wanna pull the latest master and give it a try.

I tested #7681 as a patch to 22.11.23 (https://github.com/google/flatbuffers/pull/7681.patch), and I found that the tests do pass on s390x with the patch applied. Thank you for investigating this.

[dbaileychess]

Thanks! This should be in the 22.12.06 release, mind trying that as well?

[musicinmybrain]

Thanks! This should be in the 22.12.06 release, mind trying that as well?

Yes, the tests pass for the 22.12.06 release as well.