Building Cython Extension Packages

This skill provides systematic approaches for building Cython extension packages, with emphasis on resolving compatibility issues that arise with modern Python and NumPy versions.

When to Use This Skill

• Installing legacy Cython packages that have not been updated for NumPy 2.0 or Python 3.x
• Encountering compilation errors related to deprecated NumPy types
• Building packages with

  .pyx

  Cython source files
• Installing extension packages to system or global Python environments
• Troubleshooting

  setup.py

  build failures for Cython projects

Recommended Approach

Phase 1: Pre-Build Analysis (Before Attempting to Build)

Conduct a comprehensive inventory of potential compatibility issues BEFORE attempting to build. This proactive approach prevents the inefficient cycle of build-fail-fix-rebuild.

1. Identify All Deprecated NumPy Types

Search for ALL known NumPy 2.0 deprecated type aliases simultaneously:

# Search in both Python (.py) and Cython (.pyx) files
grep -rn "np\.float[^0-9]" --include="*.py" --include="*.pyx" .
grep -rn "np\.int[^0-9]" --include="*.py" --include="*.pyx" .
grep -rn "np\.complex[^0-9]" --include="*.py" --include="*.pyx" .
grep -rn "np\.bool[^0-9]" --include="*.py" --include="*.pyx" .
grep -rn "np\.object[^0-9]" --include="*.py" --include="*.pyx" .
grep -rn "np\.str[^0-9]" --include="*.py" --include="*.pyx" .

2. Check for Python 3.x Compatibility Issues

# Common Python 2/3 compatibility issues
grep -rn "from fractions import gcd" --include="*.py" .
grep -rn "print " --include="*.py" .  # Python 2 print statements
grep -rn "xrange" --include="*.py" .
grep -rn "\.iteritems\|\.itervalues\|\.iterkeys" --include="*.py" .

3. Examine Cython Files Specifically

Cython

.pyx

# Look for cdef declarations with deprecated types
grep -rn "cdef.*np\." --include="*.pyx" .

Phase 2: Systematic Fixes

Apply fixes comprehensively before building, not reactively after each error.

NumPy Type Replacements:

np.float

np.float64

float

np.int

np.int64

int

np.complex

np.complex128

complex

np.bool

np.bool_

bool

np.object

np.object_

object

np.str

np.str_

str

Type Checking Considerations:

When replacing

isinstance()

# Instead of: isinstance(x, np.complex)
# Use: isinstance(x, (np.complexfloating, complex))
# This covers complex64, complex128, and Python's built-in complex

Python 3.x Fixes:

# gcd import
# Old: from fractions import gcd
# New: from math import gcd

Phase 3: Build and Install

Installation Types:

Understand the difference between installation methods:

• pip install .

  - Proper installation, copies package to site-packages

• pip install -e .

  - Editable/development install, links to source directory

When the task specifies "install to system/global Python environment," use

pip install .

-e

Clean Build Process:

# Clean any previous build artifacts
python setup.py clean --all
rm -rf build/ dist/ *.egg-info/

# Build and install
pip install .

Phase 4: Verification

1. Verify Installation Location

pip show <package-name>
python -c "import <package>; print(<package>.__file__)"

2. Test All Extension Modules

Run tests that specifically exercise Cython-compiled code paths:

# Run the test suite
python -m pytest tests/

# If specific test files exist for extensions, run those
python -m pytest tests/test_*.py -v

3. Import Verification

# Verify each compiled extension can be imported
import <package>.<cython_module>

4. Clean Rebuild Verification

After all fixes, perform a clean rebuild to ensure no stale artifacts:

pip uninstall <package> -y
rm -rf build/
pip install .

Common Pitfalls

1. Reactive vs. Proactive Fixing: Do not fix errors one-by-one as they appear. Search comprehensively first.

2. Ignoring Cython Files: The

   .pyx

   source files often contain the same deprecated types as

   .py

   files. Always check both.

3. Incomplete Type Coverage: When fixing

   isinstance()

   checks for complex numbers, account for all NumPy complex types (

   complex64

   ,

   complex128

   ) not just

   complex128

   .

4. Editable Install When Global Is Required: Using

   pip install -e .

   when the task requires a proper global installation leaves the package tied to the source directory.

5. Stale Build Artifacts: Failing to clean build directories before rebuilding can cause compiled extensions to use old code.

6. Missing Extension Tests: Verifying only Python code tests while leaving Cython extension tests unrun.

Reference Materials

For a comprehensive list of NumPy 2.0 migration changes, see

references/numpy2_migration.md