Repository navigation
[C++][Release] verify-rc-source-windows fails with latest simdjson (5.0.1) from conda #51690
Description
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
raulcd commentedon Oct 2, 2026
@pitrou @kou @Reranko05 is this a blocker? Should I just pin
simdjson < 5or use bundled on the verification job for Windows?raulcd commentedon Oct 2, 2026
Marking this potentially as 26.0.0 until feedback even though I think it's not a blocker.
kou commentedon Oct 2, 2026
I think that this is not a blocker.
It seems that
arrow/cpp/cmake_modules/SetupCxxFlags.cmake
Line 56 in e85e181
BTW, I feel that the approach used by #46963 is not a good approach because it always defines
__*__globally byadd_definitions()(other SIMD related flags cares aboutARROW_SIMD_LEVEL) and we should not define__*__macros as much as possible (I think that simdjson/simdjson#2864 (comment) is correct).Reranko05 commentedon Oct 3, 2026
@raulcd I don't think this is a blocker either.
Reranko05 commentedon Oct 6, 2026
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.kou commentedon Oct 16, 2026
If xtensor-stack/xsimd#1430 is merged, we can use
XSIMD_WITH_SSE4_2instead of__SSE4_1__and__SSE4_2__.Allow setting XSIMD_WITH_SSE3, SSSE3, SSE4_1 and SSE4_2 manually with…