Visitar URL original
[C++][Release] verify-rc-source-windows fails with latest simdjson (5.0.1) from conda · Issue #51690 · apache/arrow · GitHub
Skip to content

[C++][Release] verify-rc-source-windows fails with latest simdjson (5.0.1) from conda #51690

Description

@raulcd

Describe the bug, including details regarding any error messages, version, and platform.

Verification job for Windows is failing with the following:

   unity_1_cxx.cxx
  unity_0_cxx.cxx
C:\tmp\arrow-verify-release\conda-env\Library\include\simdjson.h(97474,68): error C3861: '__builtin_ctz': identifier not found [D:\a\crossbow\crossbow\arrow\cpp\build\src\arrow\arrow_json_shared.vcxproj]
  (compiling source file 'CMakeFiles/arrow_json_shared.dir/Unity/unity_0_cxx.cxx')
  
C:\tmp\arrow-verify-release\conda-env\Library\include\simdjson.h(97474,68): error C3861: '__builtin_ctz': identifier not found [D:\a\crossbow\crossbow\arrow\cpp\build\src\arrow\arrow_json_shared.vcxproj]
  (compiling source file 'CMakeFiles/arrow_json_shared.dir/Unity/unity_1_cxx.cxx')
  
  Creating directories for 'mimalloc_ep'
  Building Custom Rule D:/a/crossbow/crossbow/arrow/cpp/CMakeLists.txt

I think it is related to the following issue and the latest simdjson (simdjson-5.0.1) on conda.

Component(s)

C++, Release

Activity

  1. raulcd commented on Oct 2, 2026

    @raulcd
    MemberAuthor

    @pitrou @kou @Reranko05 is this a blocker? Should I just pin simdjson < 5 or use bundled on the verification job for Windows?

  2. added this to the 26.0.0 milestone on Oct 2, 2026
  3. raulcd commented on Oct 2, 2026

    @raulcd
    MemberAuthor

    Marking this potentially as 26.0.0 until feedback even though I think it's not a blocker.

  4. kou commented on Oct 2, 2026

    @kou
    Member

    I think that this is not a blocker.

    It seems that

    add_definitions(-D__SSE2__ -D__SSE4_1__ -D__SSE4_2__)
    introduced by #46963 is the cause of this. It's already included in our past releases. We can fix it in the next release.

    BTW, I feel that the approach used by #46963 is not a good approach because it always defines __*__ globally by add_definitions() (other SIMD related flags cares about ARROW_SIMD_LEVEL) and we should not define __*__ macros as much as possible (I think that simdjson/simdjson#2864 (comment) is correct).

  5. Reranko05 commented on Oct 3, 2026

    @Reranko05
    Collaborator

    @raulcd I don't think this is a blocker either.

  6. removed this from the 26.0.0 milestone on Oct 5, 2026
  7. Reranko05 commented on Oct 6, 2026

    @Reranko05
    Collaborator

    One other related point: simdjson 5.0.2 also includes a fix for an empty-input issue in truncated_bytes() (simdjson#2889) that we're currently working around in #51038. So once we fix the __*__ definitions on our side and can bump simdjson, we'll also pick up that fix.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions