Fix Python 3.12 and NumPy 2.x compatibility issues

[sackio]

Summary

This PR fixes multiple compatibility issues preventing zigzag from building on Python 3.12 with NumPy 2.x and Cython 3.x. These changes enable zigzag to work with modern Python environments.

Problem

The current version fails to install on Python 3.12+ with errors:

1. Invalid pyproject.toml: error: invalid-pyproject-build-system-requires - invalid requirement: 'Cython>=^0.29'
2. Missing distutils: ModuleNotFoundError: No module named 'distutils' (removed in Python 3.12)
3. NumPy 2.x types: Invalid type errors with np.int_ (removed in NumPy 2.x)
4. Cython 3.x types: Invalid type errors with int_t (deprecated in Cython 3.x)
5. Build script errors: get_outputs() fails without build_py command

Changes

1. Fix pyproject.toml (commit e0c62a0)

File: pyproject.toml

Before:

requires = ["Cython>=^0.29", "setuptools", "wheel"]

After:

requires = ["Cython>=0.29", "setuptools", "wheel"]

Reason: The ^ operator is Poetry-specific syntax and not valid in PEP 508/PEP 440 dependency specifications used by build-system.requires.

2. Replace distutils with setuptools (commit 013d2cc)

File: build.py

Changes:

• from distutils.command.build_ext import build_ext → from setuptools.command.build_ext import build_ext
• from distutils.core import Distribution, Extension → from setuptools.dist import Distribution + from setuptools import Extension

Reason: Python 3.12 removed the distutils module. Use setuptools as the modern replacement.

3. Add setuptools to build requirements (commit 922933f)

File: pyproject.toml

Before:

requires = [
    "poetry-core>=1.0.0",
    "Cython>=0.29",
    "numpy >=1.21.1",
]

After:

requires = [
    "poetry-core>=1.0.0",
    "setuptools",
    "Cython>=0.29",
    "numpy >=1.21.1",
]

Reason: Since build.py now imports from setuptools, it must be in build-system.requires.

4. Fix NumPy 2.x type compatibility (commit 98167d0)

File: zigzag/core.pyx

Changes:

• Line 107: dtype=np.int_ → dtype=np.intp
• Line 213: dtype=np.int_ → dtype=np.intp

Reason: NumPy 2.0 removed np.int_ type alias. Use np.intp (platform-specific signed integer type used for indexing).

5. Fix Cython 3.x type compatibility (commit 7a2df2e)

File: zigzag/core.pyx

Changes:

• Added: cimport numpy as cnp
• Replaced all 12 occurrences of int_t with cnp.intp_t

Reason: The int_t type is deprecated in modern Cython with NumPy 2.x. Use cnp.intp_t for platform-specific integer types.

6. Add error handling to build script (commit 1b7bcf3)

File: build.py

Changes: Added try/except block with glob fallback for cmd.get_outputs():

try:
    outputs = cmd.get_outputs()
except (AttributeError, KeyError):
    # Fallback: manually find built extension files
    import glob
    outputs = glob.glob(os.path.join(cmd.build_lib, 'zigzag', '*.so'))
    outputs.extend(glob.glob(os.path.join(cmd.build_lib, 'zigzag', '*.pyd')))

Reason: get_outputs() can fail when build_py command doesn't exist in the minimal Distribution setup. Fallback uses glob to find built .so/.pyd files.

Testing

Tested successfully on:

• Python: 3.12
• NumPy: 2.2.6
• Cython: 0.29.37
• Platform: Linux x86_64 (Docker)

Build output:

Building wheel for zigzag (pyproject.toml): finished with status 'done'
Created wheel for zigzag: filename=zigzag-0.3.2-cp312-cp312-manylinux_2_41_x86_64.whl
Successfully installed zigzag-0.3.2

Import test:

import zigzag
print(zigzag.peak_valley_pivots([1, 2, 3, 2, 1], 0.1, -0.1))  # Works!

Impact

This fixes installation for:

• Dependent packages like smartmoneyconcepts that require zigzag
• Modern Python 3.12+ environments
• Projects using NumPy 2.x
• Docker containers with recent Python versions

Backward Compatibility

All changes maintain backward compatibility:

• Cython>=0.29 still accepts older versions
• setuptools is widely available
• np.intp works on all NumPy versions
• cnp.intp_t is the recommended modern Cython type

References

• PEP 508 - Dependency specification
• PEP 440 - Version Identification
• PEP 632 - Deprecate distutils
• NumPy 2.0 Migration Guide
• Cython NumPy Tutorial