Description
Describe the issue:
Using Numpy 2.x (tried 2.02, 2.1.2) I can no longer access module data.
After f2py I get the pyf file
! -*- f90 -*-
! Note: the context of this file is case sensitive.
python module _solver ! in
interface ! in :_solver
module types ! in :_solver:solver.f90
integer, parameter,optional :: int32=selected_int_kind(8)
integer, parameter,optional :: int64=selected_int_kind(16)
integer, parameter,optional :: real64=selected_real_kind(15)
end module types
module lanedata ! in :_solver:solver.f90
use types, only: real64,int32
integer(kind=4), parameter,optional :: maxdata=1048575
real(kind=8), allocatable,dimension(:,:) :: theta
integer(kind=4) :: ndata
end module lanedata
subroutine freelanedata ! in :_solver:solver.f90
use lanedata, only: ndata,theta
end subroutine freelanedata
subroutine lane(dx,n,w) ! in :_solver:solver.f90
use types, only: int32,real64
use lanedata, only: maxdata,theta,ndata
real(kind=8) intent(in) :: dx
real(kind=8) intent(in) :: n
real(kind=8) intent(in) :: w
end subroutine lane
subroutine rk4(x0,y0,y1,dx,n,w,z0,z1) ! in :_solver:solver.f90
use types, only: real64
real(kind=8) intent(in) :: x0
real(kind=8) intent(in) :: y0
real(kind=8) intent(in) :: y1
real(kind=8) intent(in) :: dx
real(kind=8) intent(in) :: n
real(kind=8) intent(in) :: w
real(kind=8) intent(out) :: z0
real(kind=8) intent(out) :: z1
end subroutine rk4
end interface
end python module _solver
! This file was auto-generated with f2py (version:2.1.2).
! See:
! https://web.archive.org/web/20140822061353/http://cens.ioc.ee/projects/f2py2ebut when I try to access the data, as suggested in the online doc (just checking that this is still as it used to be, https://numpy.org/doc/2.1/f2py/python-usage.html#fortran-90-module-data), I get
In [2]: print(laneemden._solver.__doc__)
This module '_solver' is auto-generated with f2py (version:2.1.2).
Functions:
freelanedata()
lane(dx,n,w)
z0,z1 = rk4(x0,y0,y1,dx,n,w)
.
or
In [3]: dir(laneemden._solver)
Out[3]:
['__doc__',
'__f2py_numpy_version__',
'__file__',
'__loader__',
'__name__',
'__package__',
'__solver_error',
'__spec__',
'__version__',
'freelanedata',
'lane',
'rk4']
so, not trace of module lanedata. This used to work (and still does) in Numpy 1.x.
Reproduce the code example:
from laneemden._solver import lanedataHere my code, solver.f90
! 20111212 Alexander Heger
! 1/z**2 d/dz (z**2 d/dz theta(z)) + theta(z)**n = 0
! for small z one can approximate
! theta(z) = 1. + (-1/6.)*z**2 + (n/120.)*z**4 + O(z**6)
! Therefore lim(z)-->0 d**2 theta(z)/d z**2 = -1/3
! 20120423 Alexander Heger
! if we include a constant rotation rate Omega the equation becomes
! 1/z**2 d/dz (z**2 d/dz theta(z)) + theta(z)**n - w = 0
! where
! w = W/rho_c
! W = 2 Omega**2 / 4 pi G
! for small z one can approximate
! theta(z) = 1. + (w - 1)/6. *z**2 + (1.-w)*(n/120.)*z**4 + O(z**6)
! Therefore lim(z)-->0 d**2 theta(z)/d z**2 = (w-1.)/3
module types
implicit none
INTEGER, PARAMETER :: int32 = SELECTED_INT_KIND(8)
INTEGER, PARAMETER :: int64 = SELECTED_INT_KIND(16)
INTEGER, PARAMETER :: real64 = SELECTED_REAL_KIND(15)
end module types
module lanedata
use types, only: &
real64, int32
implicit none
save
integer(kind=int32), parameter :: &
maxdata = 2**20-1
real(kind=real64), dimension(:,:), allocatable :: &
theta
integer(kind=int32) :: &
ndata
end module lanedata
subroutine freelanedata()
use lanedata, only: &
ndata, theta
implicit none
if (allocated(theta)) deallocate(theta)
ndata = -1
end subroutine freelanedata
subroutine lane(dx, n, w)
! lane emden integration, return data,
! including first invalid (rho < 0) point for interpolation.
use types, only: &
int32, real64
use lanedata, only: &
maxdata, theta, ndata
implicit none
!f2py real(kind=real64), intent(in) :: dx, n, w
real(kind=real64), intent(in) :: &
dx, n, w
real(kind=real64) :: &
x
integer(kind=int32) :: &
i, j
real(kind=real64), dimension(0:1) :: &
y, z
real(kind=real64), dimension(:,:), allocatable :: &
theta_
if (allocated(theta)) deallocate(theta)
j = maxdata
allocate(theta_(0:j, 0:1))
x = 0.d0
y(0:1) = (/1.d0, 0.d0/)
i = 0
theta_(i,:) = y(:)
do while (.True.)
call rk4(x,y(0),y(1),dx,n,w,z(0),z(1))
if (i == j) then
j = j + maxdata
allocate(theta(0:j,0:1))
theta(0:i,:) = theta_(0:i,:)
deallocate(theta_)
call move_alloc(theta, theta_)
endif
i = i+1
theta_(i, :) = z(:)
x = x + dx
y(:) = z(:)
if (y(0) < 0.d0) exit
enddo
ndata = i
allocate(theta(0:ndata, 0:1))
theta(0:ndata,0:1) = theta_(0:ndata,0:1)
deallocate(theta_)
end subroutine lane
subroutine rk4(x0, y0, y1, dx, n, w, z0, z1)
use types, only: &
real64
implicit none
real(kind=real64), parameter :: &
p13 = 1.d0 / 3.d0, &
p16 = 1.d0 / 6.d0
real(kind=real64), intent(in) :: &
x0, y0, y1
real(kind=real64), intent(in) :: &
dx, n, w
real(kind=real64), intent(out) :: &
z0, z1
real(kind=real64) :: &
xh, dh
real(kind=real64) :: &
k10, k11, k20, k21, k30, k31, k40, k41
!f2py real(8), intent(in) :: x0, dx, n, w
!f2py real(8), intent(in) :: y0, y1
!f2py real(8), intent(out) :: z0, z1
xh = x0 + 0.5d0 * dx
dh = 0.5d0 * dx
k10 = y1
if (x0 == 0) then
k11 = (w - 1.d0) * p13
else
k11 = -2.d0 / x0 * y1 - (max(y0, 0.d0))**n + w
endif
k20 = y1 + dh*k11
k21 = -2.d0 / xh * k20 - (max(y0 + dh*k10, 0.d0))**n
k30 = y1 + dh*k21
k31 = -2.d0 / xh * k30 - (max(y0 + dh*k20, 0.d0))**n
k40 = y1 + dx*k31
k41 = -2.d0 / (x0+dx) * k40 - (max(y0 + dx*k30,0.d0))**n
z0 = y0 + dx*(k10 + 2.d0 * (k20 + k30) + k40) * p16
z1 = y1 + dx*(k11 + 2.d0 * (k21 + k31) + k41) * p16
end subroutine rk4
I use a custom build package that does not work with meson (and sadly, the usually very helpful Numpy developers out of principle refused to include fixes to obvious bugs in the build script to this day and a small patch to make my script continue to work, for both of which I had posted patches) and that are lengthy, so I won't include.
The issue seems to be that there is on interface in module lanedata. If I include an interface into the definition of module lanedata so it becomes, e.g.,
module lanedata
use types, only: &
real64, int32
implicit none
save
integer(kind=int32), parameter :: &
maxdata = 2**20-1
real(kind=real64), dimension(:,:), allocatable :: &
theta
integer(kind=int32) :: &
ndata
contains
subroutine phoney
end subroutine phoney
end module lanedatathen the module is included and an interface generated. This behaviour to skip modules w/o interface contents (just adding the contains line but an empty section does not suffice).
Maybe the Python 1.x behaviour can be restored, as it is common to have plain data modules w/o definition of functions or subroutines.
It is also possible that I missed a flag / behaviour change, my apologies in this case and please advise.
Error message:
---------------------------------------------------------------------------
ImportError Traceback (most recent call last)
Cell In[4], line 1
----> 1 from laneemden._solver import lanedata
ImportError: cannot import name 'lanedata' from 'laneemden._solver' (/home/alex/python/source/laneemden/_solver.cpython-311-x86_64-linux-gnu.so)Python and NumPy Versions:
2.1.2
3.11.10 (main, Sep 8 2024, 14:25:06) [GCC 14.2.1 20240801 (Red Hat 14.2.1-1)]
Runtime Environment:
[{'numpy_version': '2.1.2',
'python': '3.11.10 (main, Sep 8 2024, 14:25:06) [GCC 14.2.1 20240801 (Red '
'Hat 14.2.1-1)]',
'uname': uname_result(system='Linux', node='w.2sn.net', release='6.11.3-200.fc40.x86_64', version='#1 SMP PREEMPT_DYNAMIC Thu Oct 10 22:31:19 UTC 2024', machine='x86_64')},
{'simd_extensions': {'baseline': ['SSE', 'SSE2', 'SSE3'],
'found': ['SSSE3',
'SSE41',
'POPCNT',
'SSE42',
'AVX',
'F16C',
'FMA3',
'AVX2',
'AVX512F',
'AVX512CD',
'AVX512_SKX'],
'not_found': ['AVX512_KNL',
'AVX512_KNM',
'AVX512_CLX',
'AVX512_CNL',
'AVX512_ICL']}},
{'architecture': 'SkylakeX',
'filepath': '/home/alex/Python_3.11.10/lib/python3.11/site-packages/numpy.libs/libscipy_openblas64_-ff651d7f.so',
'internal_api': 'openblas',
'num_threads': 16,
'prefix': 'libscipy_openblas',
'threading_layer': 'pthreads',
'user_api': 'blas',
'version': '0.3.27'}]
Context for the issue:
breaks code, can no longer access module data that worked in Numpy < 2.0.