Rust Backend failing with Python 3.14

[jeilealr]

Hi,

While creating and testing the GitHub Actions workflow for GSTools benchmarking , I noticed that the Rust backend through gstools_core segfaults with Python 3.14. The same minimal variogram smoke test works with any other version of python from 3.11 to 3.13 Python.

I am not sure yet whether this is caused by Python 3.14 compatibility, the gstools_core wheel/build, or something else. Below I show how to reproduce the error I've got.

Python 3.13 Control

Create an environment:

conda create -n gstools-rust-repro-py313 -y -c conda-forge \
  python=3.13 \
  numpy \
  scipy \
  gstools \
  gstools-cython \
  emcee \
  hankel \
  meshio \
  pyevtk \
  pip

Install gstools_core:

conda activate gstools-rust-repro-py313
python -m pip install gstools_core

Create and execute the following Rust variogram smoke test as rust_variogram_smoke.py:

import importlib.metadata as md

import numpy as np

import gstools as gs
import gstools_core  # noqa: F401


def version(name):
    try:
        return md.version(name)
    except md.PackageNotFoundError:
        return "not installed"


print(f"python packages:")
print(f"  gstools: {version('gstools')}")
print(f"  gstools-cython: {version('gstools-cython')}")
print(f"  gstools_core: {version('gstools_core')}")
print(f"  numpy: {version('numpy')}")
print(f"  scipy: {version('scipy')}")

gs_config = (
    gs.config._GSTOOLS_CORE_AVAIL,
    gs.config.USE_GSTOOLS_CORE,
    gs.config.NUM_THREADS,
)

try:
    gs.config._GSTOOLS_CORE_AVAIL = True
    gs.config.USE_GSTOOLS_CORE = True
    gs.config.NUM_THREADS = 2

    x = np.linspace(0.0, 10.0, 12)
    y = np.linspace(0.0, 5.0, 12)
    field = np.sin(x) + np.cos(y)
    bins = np.linspace(0.0, 8.0, 6)

    result = gs.vario_estimate(
        (x, y),
        field,
        bins,
        mesh_type="unstructured",
        return_counts=True,
    )
finally:
    (
        gs.config._GSTOOLS_CORE_AVAIL,
        gs.config.USE_GSTOOLS_CORE,
        gs.config.NUM_THREADS,
    ) = gs_config

print("Rust variogram smoke: PASS")
print(result)

The results I've got in this case was:

python packages:
  gstools: 1.7.0
  gstools-cython: 1.2.0
  gstools_core: 1.2.0
  numpy: 2.4.6
  scipy: 1.17.1
Rust variogram smoke: PASS

Python 3.14 Suspected Failure

Create the Python 3.14 environment:

conda create -n gstools-rust-repro-py314 -y -c conda-forge \
  python=3.14 \
  numpy \
  scipy \
  gstools \
  gstools-cython \
  emcee \
  hankel \
  meshio \
  pyevtk \
  pip

Install gstools_core:

conda activate gstools-rust-repro-py314
python -m pip install gstools_core

Run the same smoke test:

python rust_variogram_smoke.py

In this case the result was:

python packages:
  gstools: 1.7.0
  gstools-cython: 1.2.0
  gstools_core: 1.2.0
  numpy: 2.4.6
  scipy: 1.17.1
[1]    61058 segmentation fault  python 

As mentioned above, I am not yet sure why the Rust backend fails with Python 3.14. For now, I have decided to use only the Cython backend in the benchmarking workflow until this can be investigated and resolved.

The previous GitHub Actions workflow did not check Rust backend compatibility, so I think this may be the first time this issue has surfaced in CI.

How should we proceed with this? Who would be the best person to investigate or own this issue? Also, should Python 3.14 Rust-backend compatibility be treated as a priority?

[LSchueler]

Thanks for reporting this! I'll have a look into this.