Remove CFFI in favor of native extension to support free-threaded mode

[adriencaccia]

Fix #86

[codspeed-hq]

CodSpeed Performance Report

Merging #96 will degrade performances by 24.69%

_{Comparing feat/remove-cffi (411adc2) with master (adee8a1)}

Summary

⚡ 47 improvements
❌ 22 regressions
✅ 98 untouched

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Mode	Benchmark	BASE	HEAD	Change
⚡	Simulation	test_iir_filter_process	35.2 µs	32.8 µs	+7.29%
❌	WallTime	test_iir_filter_process	2.6 µs	2.7 µs	-3.85%
⚡	Simulation	test_iir_filter_set_coefficients[a_coeffs0-b_coeffs0]	16.7 µs	14 µs	+19.03%
⚡	Simulation	test_make_allpass	46.4 µs	43.8 µs	+5.99%
⚡	Simulation	test_make_bandpass	47.1 µs	44.5 µs	+5.93%
⚡	Simulation	test_make_highpass	47.2 µs	44.5 µs	+6.13%
⚡	Simulation	test_make_highshelf	46.1 µs	43.6 µs	+5.72%
⚡	Simulation	test_make_lowpass	47 µs	44.5 µs	+5.81%
❌	WallTime	test_make_lowpass	5.3 µs	5.5 µs	-2.81%
⚡	Simulation	test_make_lowshelf	46.1 µs	43.6 µs	+5.8%
⚡	WallTime	test_make_lowshelf	7.4 µs	7.2 µs	+3.3%
❌	WallTime	test_make_peak	6 µs	6.2 µs	-2.73%
⚡	Simulation	test_make_peak	43.3 µs	40.9 µs	+5.9%
⚡	Simulation	test_color[graph0-3]	76.7 µs	74 µs	+3.68%
⚡	Simulation	test_combination_lists[0-0]	26.5 µs	23.9 µs	+10.78%
⚡	Simulation	test_combination_lists[4-2]	31.3 µs	28.7 µs	+8.81%
⚡	Simulation	test_combination_lists[5-4]	31.4 µs	28.9 µs	+8.65%
❌	WallTime	test_combination_sum[candidates0-8]	10.9 µs	11.1 µs	-2.62%
⚡	Simulation	test_combination_sum[candidates0-8]	43.6 µs	40.9 µs	+6.61%
⚡	Simulation	test_depth_first_search[4]	86.2 µs	84.2 µs	+2.34%
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

[Vizonex]

Fix #86

This is still a work in progress, it must be tested thoroughly in CI to ensure that it plays well with free-threaded versions of Python, using UV or not.

Sounds good. I can't wait to see if this will pass on through or not.

[adriencaccia]

Sounds good. I can't wait to see if this will pass on through or not.

It seems that there is an issue when running CODSPEED_LOG=debug codspeed --mode instrumentation -- uv run pytest tests/ --codspeed with python3.14t
Here is the error:

Fatal Python error:   ⠠ Running the benchmarks...                                                                                                                                                                                                                                                                      
preconfig_init_allocator:   ⠠ Running the benchmarks...                                                                                                                                                                                                                                                                
PYTHONMALLOC: unknown allocator
Python runtime state:   ⠠ Running the benchmarks...                                                                                                                                                                                                                                                                    
preinitializing  ⠠ Running the benchmarks...                                                                                                                                                                                                                                                                           


Stack (most recent call first):
  <tstate is freed>

In the runner we are setting the following:
https://github.com/CodSpeedHQ/runner/blob/e89b00befed092f3300c89b9f6577a00e025c944/src/run/runner/valgrind/measure.rs#L91

Do you know if there is any issue with free-threaded python and using PYTHONMALLOC?

[Vizonex]

Sounds good. I can't wait to see if this will pass on through or not.

It seems that there is an issue when running CODSPEED_LOG=debug codspeed --mode instrumentation -- uv run pytest tests/ --codspeed with python3.14t Here is the error:

Fatal Python error:   ⠠ Running the benchmarks...                                                                                                                                                                                                                                                                      
preconfig_init_allocator:   ⠠ Running the benchmarks...                                                                                                                                                                                                                                                                
PYTHONMALLOC: unknown allocator
Python runtime state:   ⠠ Running the benchmarks...                                                                                                                                                                                                                                                                    
preinitializing  ⠠ Running the benchmarks...                                                                                                                                                                                                                                                                           


Stack (most recent call first):
  <tstate is freed>

In the runner we are setting the following: https://github.com/CodSpeedHQ/runner/blob/e89b00befed092f3300c89b9f6577a00e025c944/src/run/runner/valgrind/measure.rs#L91

Do you know if there is any issue with free-threaded python and using PYTHONMALLOC?

@adriencaccia Sorry for my rather late response but there is one called PyMem_RawMalloc which can be ran without the gil same with PyMem_RawFree, PyMem_RawRealloc, PyMem_RawCalloc

[Copilot]

Pull request overview

This PR migrates from CFFI-based Python bindings to native CPython extension module to enable free-threaded Python support. The change removes the CFFI dependency and replaces it with a hand-written C extension module that wraps the existing instrument hooks library.

Key changes:

• Removed CFFI dependency and replaced with native C extension API
• Added new C module instrument_hooks_module.c implementing Python bindings using CPython's C API
• Updated InstrumentHooks class to use the new native module instead of CFFI's lib object
• Modified callgrind function calls to be methods on InstrumentHooks instead of lib properties

Reviewed changes

Copilot reviewed 7 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/pytest_codspeed/instruments/valgrind.py	Updated callgrind instrumentation calls to use instance methods instead of lib object
src/pytest_codspeed/instruments/hooks/instrument_hooks_module.c	New native C extension module implementing Python bindings for instrument hooks
src/pytest_codspeed/instruments/hooks/build.py	Removed CFFI build configuration file
src/pytest_codspeed/instruments/hooks/init.py	Refactored to use native extension module and added callgrind methods
setup.py	Replaced CFFI extension with native Extension setup
pyproject.toml	Removed CFFI dependency and updated package data configuration
.github/workflows/ci.yml	Added Python 3.14t (free-threaded) to test matrix

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The __del__ method with only a pass statement and comment should be removed entirely. An empty __del__ method can prevent proper garbage collection and the capsule destructor will be called automatically without defining __del__.