Experimental macOS (Intel) build of riptide_cpp / riptable — findings & patch notes

[airoldi]

With the help of chat GPT PRo, I’ve been experimenting with building and using riptable / riptide_cpp on macOS (Intel).
I’m sharing these results because the C++ core actually compiles on macOS with some small portability tweaks — but the Python extension ultimately segfaults during module initialization.

This issue summarizes:

• ✔️ full environment details
• ✔️ the exact build steps
• ✔️ every error encountered along the way
• ✔️ patches applied (file-by-file)
• ❌ the final unresolved crash (PyModule_Create2 inside PyInit_riptide_cpp)
• ❓ questions/suggestions for maintainers

---

1. Environment

✔️ Hardware

• Machine: 2017 iMac (Intel)
• CPU: Intel(R) Xeon(R) W-2191B @ 2.30GHz
• Cores: Logical 36 / Physical 18
• Arch: x86_64

✔️ OS

macOS 15.7.2 (24G325)
Darwin Kernel Version 24.6.0

✔️ Conda / Python

conda 24.7.1
Environment: py311 → Python 3.11.9

✔️ Build Toolchain

• CMake 3.29.4 (conda-forge)
• Ninja 1.13.1
• AppleClang 17.0.0.17000013
• SDK: macOS15.5.sdk

✔️ Dependencies installed in env

• numpy 2.3.1
• tbb + tbb-devel
• benchmark
• zstd
• scikit-build-core

✔️ Version under test

• riptable 1.17.1 (PyPI)
• riptide_cpp 1.19.0 (PyPI)
• repo cloned from GitHub (git rev-parse HEAD for my build: INSERT HASH)

---

2. Summary of What Works / Fails

✔️ Works

• Manual CMake build of the C++ core (riptide_cpp)
• Building all test binaries:
  • riptide_test ✔️ builds
  • riptide_python_test ✔️ builds
  • sdsfile_test ✔️ builds

⚠️ Partial

• Some C++ tests run; at least one (riptide_test) crashes with Illegal instruction during test discovery.

❌ Fails

• pip install riptable or pip install . for riptide_cpp using scikit-build-core
  • cmake 4.1.3 invoked by backend fails due to:
    • Expectation of libpython3.11.a
    • Incorrect CLI argument: --=SEE_OVERRIDES_BELOW=--
• Python import riptide_cpp
  • Initially: ud2 trap inside GetProcCount()
  • After patching: segfault in PyImport_Import via SetupLogging()
  • After further patching: segfault inside PyModule_Create2 from PyInit_riptide_cpp

---

3. Step-by-Step Reproduction

3.1 pip install riptable failure (CMake invoked by scikit-build-core)

Errors:

WARNING: libpython3.11.a is not a real file
CMake Error: Unknown argument --=SEE_OVERRIDES_BELOW=--
*** CMake configuration failed

✔️ Reproduction:

conda create -n py311 python=3.11
conda activate py311
pip install riptable

---

4. Manual C++ Build Using CMake 3.29.4

4.1 Clone + configure

git clone https://github.com/rtosholdings/riptide_cpp.git
cd riptide_cpp
mkdir build && cd build

Python / NumPy hints

export PY_INCLUDE_DIR=$(python -c "import sysconfig; print(sysconfig.get_paths()['include'])")
export PY_LIBRARY=$(python -c "import sysconfig, os; print(os.path.join(sysconfig.get_config_var('LIBDIR'), 'libpython'+sysconfig.get_config_var('LDVERSION')+'.dylib'))")
export NUMPY_INCLUDE_DIR=$(python -c "import numpy; print(numpy.get_include())")

Configure:

cmake .. -DCMAKE_BUILD_TYPE=Release \
  -DPython3_INCLUDE_DIR="$PY_INCLUDE_DIR" \
  -DPython3_LIBRARY="$PY_LIBRARY" \
  -DPython3_NumPy_INCLUDE_DIRS="$NUMPY_INCLUDE_DIR"

---

5. Fixes Applied During Build (with error snippets)

✔️ FIX 1 — missing benchmark config

Could not find benchmarkConfig.cmake

✔️ FIX 2 — missing TBB config

Could not find TBBConfig.cmake

✔️ FIX 3 — unsupported JCC mitigation flags

-mbranches-within-32B-boundaries not supported by Apple's assembler

✔️ FIX 4 — AVX intrinsics without flags

requires target feature 'avx'

✔️ FIX 5 — std::format / std::make_format_args error

no matching function for call to 'make_format_args'

✔️ FIX 6 — template misuse in basic_ops.h

std::is_same_v<..., y> → y is not a type

✔️ FIX 7 — ud2 in GetProcCount

GetProcCount: ud2

✔️ FIX 8 — crash inside SetupLogging

PyImport_Import → SetupLogging() → PyInit_riptide_cpp

---

6. FINAL CRASH (Unresolved)

❌ Current Behavior:

python -c "import riptide_cpp"
Segmentation fault: 11

Backtrace:

frame #0: libpython3.11.dylib`PyModule_Create2 + 11
frame #1: riptide_cpp.cpython-311-darwin.so`PyInit_riptide_cpp + 255

Interpretation:

• The module initialization sequence appears incompatible with CPython 3.11’s expectations on macOS Intel.
• Likely an issue with the PyModuleDef structure or a corruption/assumption in PyInit_riptide_cpp.

---

7. Questions

• Is macOS (Intel) considered a supported or intended platform?
  While Linux/Windows are primary targets -- JCC flags, affinity code, and logging -- a macOS (Intel or Apple Silicon) would be most welcome.

• Would you consider a “best-effort macOS mode”?
  Examples:

  • Disable formatted logging
  • Fallback CPU detection
  • SIMDe-friendly dispatch
  • Avoid affinity logic

• Packaging:
  The scikit-build-core + CMake path currently cannot build on macOS conda environments due to reliance on libpython3.11.a and invalid flags.

• Module init:
  Any guidance on known-good patterns for PyModuleDef + Python 3.11?
  The crash is reliably inside PyModule_Create2.

---

8. Appendix — High-Level Patch List

• cmake/AddCommonSettings.cmake
• src/logging/logging_service.h
• src/basic_ops.h
• src/MathWorker.cpp
• src/Logger.cpp

Full diffs are available on request.

---