Cython3 (#128)

* make all cdef also noexcept for cython 3

.github/workflows/cibuildwheel.yml

@@ -11,6 +11,7 @@ jobs:
  name: Build wheels on ${{ matrix.os }}
  runs-on: ${{ matrix.os }}
  strategy:
+ fail-fast: false
  matrix:
  os: [ubuntu-latest, windows-latest, macos-latest]
 

nlsam/stabilizer.pyx

@@ -51,7 +51,7 @@ def multiprocess_stabilization(const floating[:,:,:] data, const floating1[:,:,:
  return out, eta
 
 
-cdef double chi_to_gauss(double m, double eta, double sigma, double N, double alpha, bint use_nan) nogil:
+cdef double chi_to_gauss(double m, double eta, double sigma, double N, double alpha, bint use_nan) noexcept nogil:
  """Maps the noisy signal intensity from a Rician/Non central chi distribution
  to its gaussian counterpart. See p. 4 of [1] eq. 12.
 
@@ -103,7 +103,7 @@ cdef double chi_to_gauss(double m, double eta, double sigma, double N, double al
  return inv_cdf_gauss
 
 
-cdef inline double _marcumq_cython(double a, double b, double M, double eps=1e-8) nogil:
+cdef inline double _marcumq_cython(double a, double b, double M, double eps=1e-8) noexcept nogil:
  """Computes the generalized Marcum Q function of order M.
  https://en.wikipedia.org/wiki/Marcum_Q-function
 
@@ -152,7 +152,7 @@ cdef inline double _marcumq_cython(double a, double b, double M, double eps=1e-8
 
 
 cdef double fixed_point_finder(double m_hat, double sigma, double N,
- bint clip_eta=True, int max_iter=100, double eps=1e-6) nogil:
+ bint clip_eta=True, int max_iter=100, double eps=1e-6) noexcept nogil:
  """Fixed point formula for finding eta. Table 1 p. 11 of [1]
 
  Input
@@ -220,13 +220,13 @@ cdef double fixed_point_finder(double m_hat, double sigma, double N,
  return t1
 
 
-cdef inline double _beta(double N) nogil:
+cdef inline double _beta(double N) noexcept nogil:
  """Helper function for xi, see p. 3 [1] just after eq. 8.
  Generalized version for non integer N"""
  return sqrt(2) * gamma(N + 0.5) / gamma(N)
 
 
-cdef inline double _fixed_point_k(double eta, double m, double sigma, double N) nogil:
+cdef inline double _fixed_point_k(double eta, double m, double sigma, double N) noexcept nogil:
  """Helper function for fixed_point_finder, see p. 11 [1] eq. D2."""
  cdef:
  double fpg, num, denom, h1f1m, h1f1p
@@ -242,7 +242,7 @@ cdef inline double _fixed_point_k(double eta, double m, double sigma, double N)
  return eta - num / denom
 
 
-cdef inline double _fixed_point_k_v2(double eta, double m, double sigma, double N) nogil:
+cdef inline double _fixed_point_k_v2(double eta, double m, double sigma, double N) noexcept nogil:
  """Helper function for fixed_point_finder, see p. 11 [1] eq. D3.
 
  This is a secret equation scheme which gives rise to a different fixed point iteration
@@ -260,7 +260,7 @@ cdef inline double _fixed_point_k_v2(double eta, double m, double sigma, double
 
  return eta + num / denom
 
-cdef inline double xi(double eta, double sigma, double N) nogil:
+cdef inline double xi(double eta, double sigma, double N) noexcept nogil:
  """Standard deviation scaling factor formula, see p. 3 of [1], eq. 10.
 
  Input
@@ -300,7 +300,7 @@ cdef inline double xi(double eta, double sigma, double N) nogil:
 
 
 # Helper function for the root finding loop
-cdef inline double k(double theta, double N, double r) nogil:
+cdef inline double k(double theta, double N, double r) noexcept nogil:
  cdef:
  # Fake SNR value for xi
  double eta = theta
@@ -317,7 +317,7 @@ cdef inline double k(double theta, double N, double r) nogil:
  return theta - num / denom
 
 
-cdef double root_finder(double r, double N, int max_iter=500, double eps=1e-6) nogil:
+cdef double root_finder(double r, double N, int max_iter=500, double eps=1e-6) noexcept nogil:
  cdef:
  double lower_bound = sqrt((2*N / xi(0.0, 1.0, N)) - 1)
  double t0, t1 = 0.0
@@ -354,7 +354,7 @@ def root_finder_loop(const floating[:] data, const floating1[:] sigma, const flo
  return corrected_sigma
 
 
-cdef inline bint check_high_SNR(double eta, double sigma, double N) nogil:
+cdef inline bint check_high_SNR(double eta, double sigma, double N) noexcept nogil:
  '''If the SNR is high enough against N, these corrections factors change basically nothing, so may as well return early.'''
  cdef:
  double SNR = eta / sigma

nlsam/utils.pyx

@@ -2,7 +2,7 @@
 
 import numpy as np
 
-cdef void _im2col3D(double[:,:,::1] A, double[::1,:] R, int[:] size) nogil:
+cdef void _im2col3D(double[:,:,::1] A, double[::1,:] R, int[:] size) noexcept nogil:
 
  cdef:
  Py_ssize_t k = 0, l = 0
@@ -50,7 +50,7 @@ cdef void _im2col3D_overlap(double[:,:,::1] A, double[::1,:] R, int[:] size, int
  k += 1
 
 
-cdef void _im2col4D(double[:,:,:,::1] A, double[::1,:] R, int[:] size) nogil:
+cdef void _im2col4D(double[:,:,:,::1] A, double[::1,:] R, int[:] size) noexcept nogil:
 
  cdef:
  Py_ssize_t k = 0, l = 0
@@ -142,7 +142,7 @@ cdef void _col2im3D_overlap(double[:,:,::1] A, double[:,:,::1] div, double[::1,:
  k += 1
 
 
-cdef void _col2im3D(double[:,:,::1] A, double[:,:,::1] div, double[::1,:] R, double[:] weights, int[:] block_shape) nogil:
+cdef void _col2im3D(double[:,:,::1] A, double[:,:,::1] div, double[::1,:] R, double[:] weights, int[:] block_shape) noexcept nogil:
 
  cdef:
  Py_ssize_t k = 0, l = 0
@@ -167,7 +167,7 @@ cdef void _col2im3D(double[:,:,::1] A, double[:,:,::1] div, double[::1,:] R, dou
  k += 1
 
 
-cdef void _col2im4D(double[:,:,:,::1] A, double[:,:,::1] div, double[::1,:] R, double[:] weights, int[:] block_shape) nogil:
+cdef void _col2im4D(double[:,:,:,::1] A, double[:,:,::1] div, double[::1,:] R, double[:] weights, int[:] block_shape) noexcept nogil:
  cdef:
  Py_ssize_t k = 0, l = 0
  Py_ssize_t a, b, c, d, m, n, o, p

pyproject.toml

@@ -1,5 +1,5 @@
 [build-system]
-requires = ["Cython>=0.29.14",
+requires = ["Cython>=0.29.33",
  "scipy>=1.5",
  "oldest-supported-numpy",
  "setuptools",
@@ -19,7 +19,7 @@ license = {text = "GPLv3"}
 dependencies = [
  'numpy>=1.15.4',
  'scipy>=1.5',
- 'cython>=0.29',
+ 'cython>=0.29.33',
  'nibabel>=2.0',
  'joblib>=0.14.1',
  'autodmri>=0.2.1',