[jira] [Commented] (ARROW-17968) Arrow failed to run arrow-array-test and arrow-ipc-read-write-test on release and debug configuration with MSVC on windows

["Casey Carter (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-17968?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17617315#comment-17617315 ] 

Casey Carter commented on ARROW-17968:
--------------------------------------

The TestSwapEndianArrayData.RandomData is failing due to a combination of poor design and a change we made to the MSVC STL's `<random>` header which shipped in Visual Studio 2022 17.4 Preview 3. This test case generates an array of lengths that are NULL with very low probability (1%) and then asserts that the first and last entries in the array are not NULLs. Assuming a proper PRNG at least one of those asserts will fail for 1.99% of PRNG seeds. Our change to <random>'s `std::uniform_int_distribution` makes that distribution use random bits more efficiently, so the test case observes a different sequence of random numbers from the same seed which now fails. I validated this hypothesis with a quick hack: changing the seed used in the test case from 42 to 1729 results in the test passing.

It would be nice if this test case could be properly fixed, not by picking a different PRNG seed value that happens to succeed, but by forcing the first and last array entries to be non-NULL instead of choosing randomly.

> Arrow failed to run arrow-array-test and arrow-ipc-read-write-test on release and debug configuration with MSVC on windows
> --------------------------------------------------------------------------------------------------------------------------
>
>                 Key: ARROW-17968
>                 URL: https://issues.apache.org/jira/browse/ARROW-17968
>             Project: Apache Arrow
>          Issue Type: Bug
>          Components: C++
>         Environment: VS 2019 + Windows Server 2019
>            Reporter: Feng Nan Wei
>            Priority: Blocker
>         Attachments: ctest_debug.log
>
>
> Hi All,
> Arrow failed to run arrow-array-test on release configuration and failed to run arrow-array-test and arrow-ipc-read-write-test on debug configuration with MSVC on windows. it can be reproduced on d449d15 on master branch. Could you please help look at this issue?
>  
> Repro steps:
>  # git clone [https://github.com/apache/arrow] F:\gitP\apache\arrow
>  # cd F:\gitP\apache\arrow\cpp\build_amd64
>  # cmake -G "Visual Studio 16 2019" -A x64 -DCMAKE_SYSTEM_VERSION=10.0.18362.0 -DCMAKE_BUILD_TYPE=Debug -DARROW_BUILD_TESTS=ON -DBOOST_ROOT=F:\tools\boost_1_77_0\x64 .. 2>&1
>  # msbuild /maxcpucount:1 /p:Platform=x64 /p:Configuration=Debug arrow.sln /t:Rebuild 2>&1
>  # set ARROW_TEST_DATA=F:\gitP\apache\arrow\testing\data
>  # ctest -C Debug --output-on-failure --verbose 2>&1
>  
> Error info:
> 92% tests passed, 2 tests failed out of 26
> Label Time Summary:
> arrow-tests    =  79.99 sec*proc (26 tests)
> unittest       =  79.99 sec*proc (26 tests)
> Total Test time (real) =  80.54 sec
> The following tests FAILED:
>       1 - arrow-array-test (Failed)
>      24 - arrow-ipc-read-write-test (Failed)



--
This message was sent by Atlassian Jira
(v8.20.10#820010)